public inbox for [email protected]help / color / mirror / Atom feed
[PATCH v7 1/5] Add pgstat_report_anytime_stat() for periodic stats flushing 148+ messages / 1 participants [nested] [flat]
* [PATCH v7 1/5] Add pgstat_report_anytime_stat() for periodic stats flushing @ 2026-01-05 09:41 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-01-05 09:41 UTC (permalink / raw) Long running transactions can accumulate significant statistics (WAL, IO, ...) that remain unflushed until the transaction ends. This delays visibility of resource usage in monitoring views like pg_stat_io and pg_stat_wal and produces spikes when flushed. This commit introduces pgstat_report_anytime_stat(), which flushes non transactional statistics even inside active transactions. A new timeout handler fires every second (if enabled while adding pending stats) to call this function, ensuring timely stats visibility without waiting for transaction completion. Implementation details: - Add PgStat_FlushMode enum to classify stats kinds: * FLUSH_ANYTIME: Stats that can always be flushed (WAL, IO, ...) * FLUSH_AT_TXN_BOUNDARY: Stats requiring transaction boundaries - Modify pgstat_flush_pending_entries() and pgstat_flush_fixed_stats() to accept a boolean anytime_only parameter: * When false: flushes all stats (existing behavior) * When true: flushes only FLUSH_ANYTIME stats and skips FLUSH_AT_TXN_BOUNDARY stats - The flush_pending_cb and flush_static_cb callbacks now receive an anytime_only boolean parameter. Most of the time it's not used (except for assertions), but it's preparatory work for moving the relations stats to anytime (without introducin a new callback). - Add pgstat_schedule_anytime_update() macro to schedule the next anytime flush, relying on PGSTAT_MIN_INTERVAL The force parameter in pgstat_report_anytime_stat() is currently unused (always called with force=false) but reserved for future use cases requiring immediate flushing. --- src/backend/access/transam/xlog.c | 6 + src/backend/postmaster/bgwriter.c | 9 +- src/backend/postmaster/checkpointer.c | 10 +- src/backend/postmaster/startup.c | 2 + src/backend/postmaster/walsummarizer.c | 9 +- src/backend/postmaster/walwriter.c | 9 +- src/backend/replication/walreceiver.c | 9 +- src/backend/replication/walsender.c | 8 +- src/backend/tcop/postgres.c | 12 ++ src/backend/utils/activity/pgstat.c | 112 ++++++++++++++---- src/backend/utils/activity/pgstat_backend.c | 13 +- src/backend/utils/activity/pgstat_bgwriter.c | 2 +- .../utils/activity/pgstat_checkpointer.c | 2 +- src/backend/utils/activity/pgstat_database.c | 2 +- src/backend/utils/activity/pgstat_function.c | 4 +- src/backend/utils/activity/pgstat_io.c | 10 +- src/backend/utils/activity/pgstat_relation.c | 12 +- src/backend/utils/activity/pgstat_slru.c | 6 +- .../utils/activity/pgstat_subscription.c | 4 +- src/backend/utils/activity/pgstat_wal.c | 10 +- src/backend/utils/init/globals.c | 1 + src/backend/utils/init/postinit.c | 3 + src/include/miscadmin.h | 1 + src/include/pgstat.h | 16 +++ src/include/utils/pgstat_internal.h | 52 ++++++-- src/include/utils/timeout.h | 1 + .../test_custom_stats/test_custom_var_stats.c | 4 +- src/tools/pgindent/typedefs.list | 1 + 28 files changed, 264 insertions(+), 66 deletions(-) 10.8% src/backend/postmaster/ 6.0% src/backend/replication/ 50.7% src/backend/utils/activity/ 6.0% src/backend/ 19.3% src/include/utils/ 5.6% src/include/ diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 13ec6225b85..d01b11c7470 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -1085,6 +1085,9 @@ XLogInsertRecord(XLogRecData *rdata, pgWalUsage.wal_fpi += num_fpi; pgWalUsage.wal_fpi_bytes += fpi_bytes; + /* Schedule next anytime stats update timeout */ + pgstat_schedule_anytime_update(); + /* Required for the flush of pending stats WAL data */ pgstat_report_fixed = true; } @@ -2066,6 +2069,9 @@ AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli, bool opportunistic) pgWalUsage.wal_buffers_full++; TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_DONE(); + /* Schedule next anytime stats update timeout */ + pgstat_schedule_anytime_update(); + /* * Required for the flush of pending stats WAL data, per * update of pgWalUsage. diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 0956bd39a85..059c601c3b8 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -49,7 +49,9 @@ #include "storage/smgr.h" #include "storage/standby.h" #include "utils/memutils.h" +#include "utils/pgstat_internal.h" #include "utils/resowner.h" +#include "utils/timeout.h" #include "utils/timestamp.h" /* @@ -103,7 +105,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) pqsignal(SIGINT, SIG_IGN); pqsignal(SIGTERM, SignalHandlerForShutdownRequest); /* SIGQUIT handler was already set up by InitPostmasterChild */ - pqsignal(SIGALRM, SIG_IGN); + InitializeTimeouts(); /* establishes SIGALRM handler */ pqsignal(SIGPIPE, SIG_IGN); pqsignal(SIGUSR1, procsignal_sigusr1_handler); pqsignal(SIGUSR2, SIG_IGN); @@ -113,6 +115,11 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) */ pqsignal(SIGCHLD, SIG_DFL); + /* + * Register timeouts needed + */ + RegisterTimeout(ANYTIME_STATS_UPDATE_TIMEOUT, AnytimeStatsUpdateTimeoutHandler); + /* * We just started, assume there has been either a shutdown or * end-of-recovery snapshot. diff --git a/src/backend/postmaster/checkpointer.c b/src/backend/postmaster/checkpointer.c index e03c19123bc..e11c4b099c8 100644 --- a/src/backend/postmaster/checkpointer.c +++ b/src/backend/postmaster/checkpointer.c @@ -66,8 +66,9 @@ #include "utils/acl.h" #include "utils/guc.h" #include "utils/memutils.h" +#include "utils/pgstat_internal.h" #include "utils/resowner.h" - +#include "utils/timeout.h" /*---------- * Shared memory area for communication between checkpointer and backends @@ -215,7 +216,7 @@ CheckpointerMain(const void *startup_data, size_t startup_data_len) pqsignal(SIGINT, ReqShutdownXLOG); pqsignal(SIGTERM, SIG_IGN); /* ignore SIGTERM */ /* SIGQUIT handler was already set up by InitPostmasterChild */ - pqsignal(SIGALRM, SIG_IGN); + InitializeTimeouts(); /* establishes SIGALRM handler */ pqsignal(SIGPIPE, SIG_IGN); pqsignal(SIGUSR1, procsignal_sigusr1_handler); pqsignal(SIGUSR2, SignalHandlerForShutdownRequest); @@ -225,6 +226,11 @@ CheckpointerMain(const void *startup_data, size_t startup_data_len) */ pqsignal(SIGCHLD, SIG_DFL); + /* + * Register timeouts needed + */ + RegisterTimeout(ANYTIME_STATS_UPDATE_TIMEOUT, AnytimeStatsUpdateTimeoutHandler); + /* * Initialize so that first time-driven event happens at the correct time. */ diff --git a/src/backend/postmaster/startup.c b/src/backend/postmaster/startup.c index cdbe53dd262..4954fe425b7 100644 --- a/src/backend/postmaster/startup.c +++ b/src/backend/postmaster/startup.c @@ -32,6 +32,7 @@ #include "storage/standby.h" #include "utils/guc.h" #include "utils/memutils.h" +#include "utils/pgstat_internal.h" #include "utils/timeout.h" @@ -245,6 +246,7 @@ StartupProcessMain(const void *startup_data, size_t startup_data_len) RegisterTimeout(STANDBY_DEADLOCK_TIMEOUT, StandbyDeadLockHandler); RegisterTimeout(STANDBY_TIMEOUT, StandbyTimeoutHandler); RegisterTimeout(STANDBY_LOCK_TIMEOUT, StandbyLockTimeoutHandler); + RegisterTimeout(ANYTIME_STATS_UPDATE_TIMEOUT, AnytimeStatsUpdateTimeoutHandler); /* * Unblock signals (they were blocked when the postmaster forked us) diff --git a/src/backend/postmaster/walsummarizer.c b/src/backend/postmaster/walsummarizer.c index 2d8f57099fd..9f8ef8159d1 100644 --- a/src/backend/postmaster/walsummarizer.c +++ b/src/backend/postmaster/walsummarizer.c @@ -48,6 +48,8 @@ #include "storage/shmem.h" #include "utils/guc.h" #include "utils/memutils.h" +#include "utils/pgstat_internal.h" +#include "utils/timeout.h" #include "utils/wait_event.h" /* @@ -249,7 +251,7 @@ WalSummarizerMain(const void *startup_data, size_t startup_data_len) pqsignal(SIGINT, SignalHandlerForShutdownRequest); pqsignal(SIGTERM, SignalHandlerForShutdownRequest); /* SIGQUIT handler was already set up by InitPostmasterChild */ - pqsignal(SIGALRM, SIG_IGN); + InitializeTimeouts(); /* establishes SIGALRM handler */ pqsignal(SIGPIPE, SIG_IGN); pqsignal(SIGUSR1, procsignal_sigusr1_handler); pqsignal(SIGUSR2, SIG_IGN); /* not used */ @@ -271,6 +273,11 @@ WalSummarizerMain(const void *startup_data, size_t startup_data_len) */ pqsignal(SIGCHLD, SIG_DFL); + /* + * Register timeouts needed + */ + RegisterTimeout(ANYTIME_STATS_UPDATE_TIMEOUT, AnytimeStatsUpdateTimeoutHandler); + /* * If an exception is encountered, processing resumes here. */ diff --git a/src/backend/postmaster/walwriter.c b/src/backend/postmaster/walwriter.c index 23e79a32345..ded0f250288 100644 --- a/src/backend/postmaster/walwriter.c +++ b/src/backend/postmaster/walwriter.c @@ -61,7 +61,9 @@ #include "storage/smgr.h" #include "utils/hsearch.h" #include "utils/memutils.h" +#include "utils/pgstat_internal.h" #include "utils/resowner.h" +#include "utils/timeout.h" /* @@ -106,7 +108,7 @@ WalWriterMain(const void *startup_data, size_t startup_data_len) pqsignal(SIGINT, SignalHandlerForShutdownRequest); pqsignal(SIGTERM, SignalHandlerForShutdownRequest); /* SIGQUIT handler was already set up by InitPostmasterChild */ - pqsignal(SIGALRM, SIG_IGN); + InitializeTimeouts(); /* establishes SIGALRM handler */ pqsignal(SIGPIPE, SIG_IGN); pqsignal(SIGUSR1, procsignal_sigusr1_handler); pqsignal(SIGUSR2, SIG_IGN); /* not used */ @@ -116,6 +118,11 @@ WalWriterMain(const void *startup_data, size_t startup_data_len) */ pqsignal(SIGCHLD, SIG_DFL); + /* + * Register timeouts needed + */ + RegisterTimeout(ANYTIME_STATS_UPDATE_TIMEOUT, AnytimeStatsUpdateTimeoutHandler); + /* * Create a memory context that we will do all our work in. We do this so * that we can reset the context during error recovery and thereby avoid diff --git a/src/backend/replication/walreceiver.c b/src/backend/replication/walreceiver.c index 10e64a7d1f4..11b7c114d3b 100644 --- a/src/backend/replication/walreceiver.c +++ b/src/backend/replication/walreceiver.c @@ -77,7 +77,9 @@ #include "utils/builtins.h" #include "utils/guc.h" #include "utils/pg_lsn.h" +#include "utils/pgstat_internal.h" #include "utils/ps_status.h" +#include "utils/timeout.h" #include "utils/timestamp.h" @@ -252,7 +254,7 @@ WalReceiverMain(const void *startup_data, size_t startup_data_len) pqsignal(SIGINT, SIG_IGN); pqsignal(SIGTERM, die); /* request shutdown */ /* SIGQUIT handler was already set up by InitPostmasterChild */ - pqsignal(SIGALRM, SIG_IGN); + InitializeTimeouts(); /* establishes SIGALRM handler */ pqsignal(SIGPIPE, SIG_IGN); pqsignal(SIGUSR1, procsignal_sigusr1_handler); pqsignal(SIGUSR2, SIG_IGN); @@ -260,6 +262,11 @@ WalReceiverMain(const void *startup_data, size_t startup_data_len) /* Reset some signals that are accepted by postmaster but not here */ pqsignal(SIGCHLD, SIG_DFL); + /* + * Register timeouts needed + */ + RegisterTimeout(ANYTIME_STATS_UPDATE_TIMEOUT, AnytimeStatsUpdateTimeoutHandler); + /* Load the libpq-specific functions */ load_file("libpqwalreceiver", false); if (WalReceiverFunctions == NULL) diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c index 2cde8ebc729..a7214d0dc6f 100644 --- a/src/backend/replication/walsender.c +++ b/src/backend/replication/walsender.c @@ -1987,8 +1987,8 @@ WalSndWaitForWal(XLogRecPtr loc) if (TimestampDifferenceExceeds(last_flush, now, WALSENDER_STATS_FLUSH_INTERVAL)) { - pgstat_flush_io(false); - (void) pgstat_flush_backend(false, PGSTAT_BACKEND_FLUSH_IO); + pgstat_flush_io(false, true); + (void) pgstat_flush_backend(false, PGSTAT_BACKEND_FLUSH_IO, true); last_flush = now; } @@ -3016,8 +3016,8 @@ WalSndLoop(WalSndSendDataCallback send_data) if (TimestampDifferenceExceeds(last_flush, now, WALSENDER_STATS_FLUSH_INTERVAL)) { - pgstat_flush_io(false); - (void) pgstat_flush_backend(false, PGSTAT_BACKEND_FLUSH_IO); + pgstat_flush_io(false, true); + (void) pgstat_flush_backend(false, PGSTAT_BACKEND_FLUSH_IO, true); last_flush = now; } diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c index 21de158adbb..2089de782d5 100644 --- a/src/backend/tcop/postgres.c +++ b/src/backend/tcop/postgres.c @@ -3564,6 +3564,18 @@ ProcessInterrupts(void) pgstat_report_stat(true); } + /* + * Flush stats outside of transaction boundary if the timeout fired. + * Unlike transactional stats, these can be flushed even inside a running + * transaction. + */ + if (AnytimeStatsUpdateTimeoutPending) + { + AnytimeStatsUpdateTimeoutPending = false; + + pgstat_report_anytime_stat(false); + } + if (ProcSignalBarrierPending) ProcessProcSignalBarrier(); diff --git a/src/backend/utils/activity/pgstat.c b/src/backend/utils/activity/pgstat.c index 11bb71cad5a..a4ff64dc5ce 100644 --- a/src/backend/utils/activity/pgstat.c +++ b/src/backend/utils/activity/pgstat.c @@ -112,6 +112,7 @@ #include "utils/guc_hooks.h" #include "utils/memutils.h" #include "utils/pgstat_internal.h" +#include "utils/timeout.h" #include "utils/timestamp.h" @@ -122,8 +123,6 @@ * ---------- */ -/* minimum interval non-forced stats flushes.*/ -#define PGSTAT_MIN_INTERVAL 1000 /* how long until to block flushing pending stats updates */ #define PGSTAT_MAX_INTERVAL 60000 /* when to call pgstat_report_stat() again, even when idle */ @@ -187,7 +186,8 @@ static void pgstat_init_snapshot_fixed(void); static void pgstat_reset_after_failure(void); -static bool pgstat_flush_pending_entries(bool nowait); +static bool pgstat_flush_pending_entries(bool nowait, bool anytime_only); +static bool pgstat_flush_fixed_stats(bool nowait, bool anytime_only); static void pgstat_prep_snapshot(void); static void pgstat_build_snapshot(void); @@ -288,6 +288,7 @@ static const PgStat_KindInfo pgstat_kind_builtin_infos[PGSTAT_KIND_BUILTIN_SIZE] .fixed_amount = false, .write_to_file = true, + .flush_mode = FLUSH_AT_TXN_BOUNDARY, /* so pg_stat_database entries can be seen in all databases */ .accessed_across_databases = true, @@ -305,6 +306,7 @@ static const PgStat_KindInfo pgstat_kind_builtin_infos[PGSTAT_KIND_BUILTIN_SIZE] .fixed_amount = false, .write_to_file = true, + .flush_mode = FLUSH_AT_TXN_BOUNDARY, .shared_size = sizeof(PgStatShared_Relation), .shared_data_off = offsetof(PgStatShared_Relation, stats), @@ -321,6 +323,7 @@ static const PgStat_KindInfo pgstat_kind_builtin_infos[PGSTAT_KIND_BUILTIN_SIZE] .fixed_amount = false, .write_to_file = true, + .flush_mode = FLUSH_AT_TXN_BOUNDARY, .shared_size = sizeof(PgStatShared_Function), .shared_data_off = offsetof(PgStatShared_Function, stats), @@ -336,6 +339,7 @@ static const PgStat_KindInfo pgstat_kind_builtin_infos[PGSTAT_KIND_BUILTIN_SIZE] .fixed_amount = false, .write_to_file = true, + .flush_mode = FLUSH_AT_TXN_BOUNDARY, .accessed_across_databases = true, @@ -353,6 +357,7 @@ static const PgStat_KindInfo pgstat_kind_builtin_infos[PGSTAT_KIND_BUILTIN_SIZE] .fixed_amount = false, .write_to_file = true, + .flush_mode = FLUSH_AT_TXN_BOUNDARY, /* so pg_stat_subscription_stats entries can be seen in all databases */ .accessed_across_databases = true, @@ -370,6 +375,7 @@ static const PgStat_KindInfo pgstat_kind_builtin_infos[PGSTAT_KIND_BUILTIN_SIZE] .fixed_amount = false, .write_to_file = false, + .flush_mode = FLUSH_ANYTIME, .accessed_across_databases = true, @@ -436,6 +442,7 @@ static const PgStat_KindInfo pgstat_kind_builtin_infos[PGSTAT_KIND_BUILTIN_SIZE] .fixed_amount = true, .write_to_file = true, + .flush_mode = FLUSH_ANYTIME, .snapshot_ctl_off = offsetof(PgStat_Snapshot, io), .shared_ctl_off = offsetof(PgStat_ShmemControl, io), @@ -453,6 +460,7 @@ static const PgStat_KindInfo pgstat_kind_builtin_infos[PGSTAT_KIND_BUILTIN_SIZE] .fixed_amount = true, .write_to_file = true, + .flush_mode = FLUSH_ANYTIME, .snapshot_ctl_off = offsetof(PgStat_Snapshot, slru), .shared_ctl_off = offsetof(PgStat_ShmemControl, slru), @@ -470,6 +478,7 @@ static const PgStat_KindInfo pgstat_kind_builtin_infos[PGSTAT_KIND_BUILTIN_SIZE] .fixed_amount = true, .write_to_file = true, + .flush_mode = FLUSH_ANYTIME, .snapshot_ctl_off = offsetof(PgStat_Snapshot, wal), .shared_ctl_off = offsetof(PgStat_ShmemControl, wal), @@ -775,23 +784,11 @@ pgstat_report_stat(bool force) partial_flush = false; /* flush of variable-numbered stats tracked in pending entries list */ - partial_flush |= pgstat_flush_pending_entries(nowait); + partial_flush |= pgstat_flush_pending_entries(nowait, false); /* flush of other stats kinds */ if (pgstat_report_fixed) - { - for (PgStat_Kind kind = PGSTAT_KIND_MIN; kind <= PGSTAT_KIND_MAX; kind++) - { - const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); - - if (!kind_info) - continue; - if (!kind_info->flush_static_cb) - continue; - - partial_flush |= kind_info->flush_static_cb(nowait); - } - } + partial_flush |= pgstat_flush_fixed_stats(nowait, false); last_flush = now; @@ -1293,7 +1290,8 @@ pgstat_prep_pending_entry(PgStat_Kind kind, Oid dboid, uint64 objid, bool *creat if (entry_ref->pending == NULL) { - size_t entrysize = pgstat_get_kind_info(kind)->pending_size; + const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); + size_t entrysize = kind_info->pending_size; Assert(entrysize != (size_t) -1); @@ -1345,9 +1343,14 @@ pgstat_delete_pending_entry(PgStat_EntryRef *entry_ref) /* * Flush out pending variable-numbered stats. + * + * If anytime_only is true, only flushes FLUSH_ANYTIME entries. + * This is safe to call inside transactions. + * + * If anytime_only is false, flushes all entries. */ static bool -pgstat_flush_pending_entries(bool nowait) +pgstat_flush_pending_entries(bool nowait, bool anytime_only) { bool have_pending = false; dlist_node *cur = NULL; @@ -1377,8 +1380,22 @@ pgstat_flush_pending_entries(bool nowait) Assert(!kind_info->fixed_amount); Assert(kind_info->flush_pending_cb != NULL); + /* Skip transactional stats if we're in anytime_only mode */ + if (anytime_only && kind_info->flush_mode == FLUSH_AT_TXN_BOUNDARY) + { + have_pending = true; + + if (dlist_has_next(&pgStatPending, cur)) + next = dlist_next_node(&pgStatPending, cur); + else + next = NULL; + + cur = next; + continue; + } + /* flush the stats, if possible */ - did_flush = kind_info->flush_pending_cb(entry_ref, nowait); + did_flush = kind_info->flush_pending_cb(entry_ref, nowait, anytime_only); Assert(did_flush || nowait); @@ -1402,6 +1419,33 @@ pgstat_flush_pending_entries(bool nowait) return have_pending; } +/* + * Flush fixed-amount stats. + * + * If anytime_only is true, only flushes FLUSH_ANYTIME stats (safe inside transactions). + * If anytime_only is false, flushes all stats with flush_static_cb. + */ +static bool +pgstat_flush_fixed_stats(bool nowait, bool anytime_only) +{ + bool partial_flush = false; + + for (PgStat_Kind kind = PGSTAT_KIND_MIN; kind <= PGSTAT_KIND_MAX; kind++) + { + const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); + + if (!kind_info || !kind_info->flush_static_cb) + continue; + + /* Skip transactional stats if we're in anytime_only mode */ + if (anytime_only && kind_info->flush_mode == FLUSH_AT_TXN_BOUNDARY) + continue; + + partial_flush |= kind_info->flush_static_cb(nowait, anytime_only); + } + + return partial_flush; +} /* ------------------------------------------------------------ * Helper / infrastructure functions @@ -2119,3 +2163,31 @@ assign_stats_fetch_consistency(int newval, void *extra) if (pgstat_fetch_consistency != newval) force_stats_snapshot_clear = true; } + +/* + * Flushes only FLUSH_ANYTIME stats using non-blocking locks. Transactional + * stats (FLUSH_AT_TXN_BOUNDARY) remain pending until transaction boundary. + * Safe to call inside transactions. + */ +void +pgstat_report_anytime_stat(bool force) +{ + bool nowait = !force; + + pgstat_assert_is_up(); + + /* Flush stats outside of transaction boundary */ + pgstat_flush_pending_entries(nowait, true); + pgstat_flush_fixed_stats(nowait, true); +} + +/* + * Timeout handler for flushing anytime stats. + */ +void +AnytimeStatsUpdateTimeoutHandler(void) +{ + AnytimeStatsUpdateTimeoutPending = true; + InterruptPending = true; + SetLatch(MyLatch); +} diff --git a/src/backend/utils/activity/pgstat_backend.c b/src/backend/utils/activity/pgstat_backend.c index f2f8d3ff75f..b09316d3ab3 100644 --- a/src/backend/utils/activity/pgstat_backend.c +++ b/src/backend/utils/activity/pgstat_backend.c @@ -31,6 +31,7 @@ #include "storage/procarray.h" #include "utils/memutils.h" #include "utils/pgstat_internal.h" +#include "utils/timeout.h" /* * Backend statistics counts waiting to be flushed out. These counters may be @@ -66,6 +67,9 @@ pgstat_count_backend_io_op_time(IOObject io_object, IOContext io_context, INSTR_TIME_ADD(PendingBackendStats.pending_io.pending_times[io_object][io_context][io_op], io_time); + /* Schedule next anytime stats update timeout */ + pgstat_schedule_anytime_update(); + backend_has_iostats = true; pgstat_report_fixed = true; } @@ -82,6 +86,9 @@ pgstat_count_backend_io_op(IOObject io_object, IOContext io_context, PendingBackendStats.pending_io.counts[io_object][io_context][io_op] += cnt; PendingBackendStats.pending_io.bytes[io_object][io_context][io_op] += bytes; + /* Schedule next anytime stats update timeout */ + pgstat_schedule_anytime_update(); + backend_has_iostats = true; pgstat_report_fixed = true; } @@ -268,7 +275,7 @@ pgstat_flush_backend_entry_wal(PgStat_EntryRef *entry_ref) * if some statistics could not be flushed due to lock contention. */ bool -pgstat_flush_backend(bool nowait, bits32 flags) +pgstat_flush_backend(bool nowait, bits32 flags, bool anytime_only) { PgStat_EntryRef *entry_ref; bool has_pending_data = false; @@ -311,9 +318,9 @@ pgstat_flush_backend(bool nowait, bits32 flags) * If some stats could not be flushed due to lock contention, return true. */ bool -pgstat_backend_flush_cb(bool nowait) +pgstat_backend_flush_cb(bool nowait, bool anytime_only) { - return pgstat_flush_backend(nowait, PGSTAT_BACKEND_FLUSH_ALL); + return pgstat_flush_backend(nowait, PGSTAT_BACKEND_FLUSH_ALL, anytime_only); } /* diff --git a/src/backend/utils/activity/pgstat_bgwriter.c b/src/backend/utils/activity/pgstat_bgwriter.c index ed2fd801189..1c5f0c3ec40 100644 --- a/src/backend/utils/activity/pgstat_bgwriter.c +++ b/src/backend/utils/activity/pgstat_bgwriter.c @@ -61,7 +61,7 @@ pgstat_report_bgwriter(void) /* * Report IO statistics */ - pgstat_flush_io(false); + pgstat_flush_io(false, true); } /* diff --git a/src/backend/utils/activity/pgstat_checkpointer.c b/src/backend/utils/activity/pgstat_checkpointer.c index 1f70194b7a7..2d89a082464 100644 --- a/src/backend/utils/activity/pgstat_checkpointer.c +++ b/src/backend/utils/activity/pgstat_checkpointer.c @@ -68,7 +68,7 @@ pgstat_report_checkpointer(void) /* * Report IO statistics */ - pgstat_flush_io(false); + pgstat_flush_io(false, true); } /* diff --git a/src/backend/utils/activity/pgstat_database.c b/src/backend/utils/activity/pgstat_database.c index 933dcb5cae5..8e86df60461 100644 --- a/src/backend/utils/activity/pgstat_database.c +++ b/src/backend/utils/activity/pgstat_database.c @@ -435,7 +435,7 @@ pgstat_reset_database_timestamp(Oid dboid, TimestampTz ts) * false without flushing the entry. Otherwise returns true. */ bool -pgstat_database_flush_cb(PgStat_EntryRef *entry_ref, bool nowait) +pgstat_database_flush_cb(PgStat_EntryRef *entry_ref, bool nowait, bool anytime_only) { PgStatShared_Database *sharedent; PgStat_StatDBEntry *pendingent; diff --git a/src/backend/utils/activity/pgstat_function.c b/src/backend/utils/activity/pgstat_function.c index e6b84283c6c..5ba4958382f 100644 --- a/src/backend/utils/activity/pgstat_function.c +++ b/src/backend/utils/activity/pgstat_function.c @@ -190,11 +190,13 @@ pgstat_end_function_usage(PgStat_FunctionCallUsage *fcu, bool finalize) * false without flushing the entry. Otherwise returns true. */ bool -pgstat_function_flush_cb(PgStat_EntryRef *entry_ref, bool nowait) +pgstat_function_flush_cb(PgStat_EntryRef *entry_ref, bool nowait, bool anytime_only) { PgStat_FunctionCounts *localent; PgStatShared_Function *shfuncent; + Assert(!anytime_only); + localent = (PgStat_FunctionCounts *) entry_ref->pending; shfuncent = (PgStatShared_Function *) entry_ref->shared_stats; diff --git a/src/backend/utils/activity/pgstat_io.c b/src/backend/utils/activity/pgstat_io.c index 28de24538dc..7cd32900236 100644 --- a/src/backend/utils/activity/pgstat_io.c +++ b/src/backend/utils/activity/pgstat_io.c @@ -19,6 +19,7 @@ #include "executor/instrument.h" #include "storage/bufmgr.h" #include "utils/pgstat_internal.h" +#include "utils/timeout.h" static PgStat_PendingIO PendingIOStats; static bool have_iostats = false; @@ -79,6 +80,9 @@ pgstat_count_io_op(IOObject io_object, IOContext io_context, IOOp io_op, /* Add the per-backend counts */ pgstat_count_backend_io_op(io_object, io_context, io_op, cnt, bytes); + /* Schedule next anytime stats update timeout */ + pgstat_schedule_anytime_update(); + have_iostats = true; pgstat_report_fixed = true; } @@ -172,9 +176,9 @@ pgstat_fetch_stat_io(void) * Simpler wrapper of pgstat_io_flush_cb() */ void -pgstat_flush_io(bool nowait) +pgstat_flush_io(bool nowait, bool anytime_only) { - (void) pgstat_io_flush_cb(nowait); + (void) pgstat_io_flush_cb(nowait, anytime_only); } /* @@ -186,7 +190,7 @@ pgstat_flush_io(bool nowait) * acquired. Otherwise, return false. */ bool -pgstat_io_flush_cb(bool nowait) +pgstat_io_flush_cb(bool nowait, bool anytime_only) { LWLock *bktype_lock; PgStat_BktypeIO *bktype_shstats; diff --git a/src/backend/utils/activity/pgstat_relation.c b/src/backend/utils/activity/pgstat_relation.c index bc8c43b96aa..04d21483d93 100644 --- a/src/backend/utils/activity/pgstat_relation.c +++ b/src/backend/utils/activity/pgstat_relation.c @@ -267,8 +267,8 @@ pgstat_report_vacuum(Relation rel, PgStat_Counter livetuples, * is done -- which will likely vacuum many relations -- or until the * VACUUM command has processed all tables and committed. */ - pgstat_flush_io(false); - (void) pgstat_flush_backend(false, PGSTAT_BACKEND_FLUSH_IO); + pgstat_flush_io(false, true); + (void) pgstat_flush_backend(false, PGSTAT_BACKEND_FLUSH_IO, true); } /* @@ -362,8 +362,8 @@ pgstat_report_analyze(Relation rel, pgstat_unlock_entry(entry_ref); /* see pgstat_report_vacuum() */ - pgstat_flush_io(false); - (void) pgstat_flush_backend(false, PGSTAT_BACKEND_FLUSH_IO); + pgstat_flush_io(false, true); + (void) pgstat_flush_backend(false, PGSTAT_BACKEND_FLUSH_IO, true); } /* @@ -812,7 +812,7 @@ pgstat_twophase_postabort(FullTransactionId fxid, uint16 info, * entry when successfully flushing. */ bool -pgstat_relation_flush_cb(PgStat_EntryRef *entry_ref, bool nowait) +pgstat_relation_flush_cb(PgStat_EntryRef *entry_ref, bool nowait, bool anytime_only) { Oid dboid; PgStat_TableStatus *lstats; /* pending stats entry */ @@ -820,6 +820,8 @@ pgstat_relation_flush_cb(PgStat_EntryRef *entry_ref, bool nowait) PgStat_StatTabEntry *tabentry; /* table entry of shared stats */ PgStat_StatDBEntry *dbentry; /* pending database entry */ + Assert(!anytime_only); + dboid = entry_ref->shared_entry->key.dboid; lstats = (PgStat_TableStatus *) entry_ref->pending; shtabstats = (PgStatShared_Relation *) entry_ref->shared_stats; diff --git a/src/backend/utils/activity/pgstat_slru.c b/src/backend/utils/activity/pgstat_slru.c index 2190f388eae..bf8a4d58673 100644 --- a/src/backend/utils/activity/pgstat_slru.c +++ b/src/backend/utils/activity/pgstat_slru.c @@ -19,6 +19,7 @@ #include "utils/pgstat_internal.h" #include "utils/timestamp.h" +#include "utils/timeout.h" static inline PgStat_SLRUStats *get_slru_entry(int slru_idx); @@ -139,7 +140,7 @@ pgstat_get_slru_index(const char *name) * acquired. Otherwise return false. */ bool -pgstat_slru_flush_cb(bool nowait) +pgstat_slru_flush_cb(bool nowait, bool anytime_only) { PgStatShared_SLRU *stats_shmem = &pgStatLocal.shmem->slru; int i; @@ -223,6 +224,9 @@ get_slru_entry(int slru_idx) Assert((slru_idx >= 0) && (slru_idx < SLRU_NUM_ELEMENTS)); + /* Schedule next anytime stats update timeout */ + pgstat_schedule_anytime_update(); + have_slrustats = true; pgstat_report_fixed = true; diff --git a/src/backend/utils/activity/pgstat_subscription.c b/src/backend/utils/activity/pgstat_subscription.c index 500b1899188..c4614817966 100644 --- a/src/backend/utils/activity/pgstat_subscription.c +++ b/src/backend/utils/activity/pgstat_subscription.c @@ -116,11 +116,13 @@ pgstat_fetch_stat_subscription(Oid subid) * false without flushing the entry. Otherwise returns true. */ bool -pgstat_subscription_flush_cb(PgStat_EntryRef *entry_ref, bool nowait) +pgstat_subscription_flush_cb(PgStat_EntryRef *entry_ref, bool nowait, bool anytime_only) { PgStat_BackendSubEntry *localent; PgStatShared_Subscription *shsubent; + Assert(!anytime_only); + localent = (PgStat_BackendSubEntry *) entry_ref->pending; shsubent = (PgStatShared_Subscription *) entry_ref->shared_stats; diff --git a/src/backend/utils/activity/pgstat_wal.c b/src/backend/utils/activity/pgstat_wal.c index 183e0a7a97b..2c2f3f10e10 100644 --- a/src/backend/utils/activity/pgstat_wal.c +++ b/src/backend/utils/activity/pgstat_wal.c @@ -51,12 +51,12 @@ pgstat_report_wal(bool force) nowait = !force; /* flush wal stats */ - (void) pgstat_wal_flush_cb(nowait); - pgstat_flush_backend(nowait, PGSTAT_BACKEND_FLUSH_WAL); + (void) pgstat_wal_flush_cb(nowait, true); + (void) pgstat_flush_backend(nowait, PGSTAT_BACKEND_FLUSH_WAL, true); /* flush IO stats */ - pgstat_flush_io(nowait); - (void) pgstat_flush_backend(nowait, PGSTAT_BACKEND_FLUSH_IO); + pgstat_flush_io(nowait, true); + (void) pgstat_flush_backend(nowait, PGSTAT_BACKEND_FLUSH_IO, true); } /* @@ -88,7 +88,7 @@ pgstat_wal_have_pending(void) * acquired. Otherwise return false. */ bool -pgstat_wal_flush_cb(bool nowait) +pgstat_wal_flush_cb(bool nowait, bool anytime_only) { PgStatShared_Wal *stats_shmem = &pgStatLocal.shmem->wal; WalUsage wal_usage_diff = {0}; diff --git a/src/backend/utils/init/globals.c b/src/backend/utils/init/globals.c index 36ad708b360..ad44826c39e 100644 --- a/src/backend/utils/init/globals.c +++ b/src/backend/utils/init/globals.c @@ -40,6 +40,7 @@ volatile sig_atomic_t IdleSessionTimeoutPending = false; volatile sig_atomic_t ProcSignalBarrierPending = false; volatile sig_atomic_t LogMemoryContextPending = false; volatile sig_atomic_t IdleStatsUpdateTimeoutPending = false; +volatile sig_atomic_t AnytimeStatsUpdateTimeoutPending = false; volatile uint32 InterruptHoldoffCount = 0; volatile uint32 QueryCancelHoldoffCount = 0; volatile uint32 CritSectionCount = 0; diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c index b59e08605cc..eeeac1bf39a 100644 --- a/src/backend/utils/init/postinit.c +++ b/src/backend/utils/init/postinit.c @@ -64,6 +64,7 @@ #include "utils/injection_point.h" #include "utils/memutils.h" #include "utils/pg_locale.h" +#include "utils/pgstat_internal.h" #include "utils/portal.h" #include "utils/ps_status.h" #include "utils/snapmgr.h" @@ -773,6 +774,8 @@ InitPostgres(const char *in_dbname, Oid dboid, RegisterTimeout(CLIENT_CONNECTION_CHECK_TIMEOUT, ClientCheckTimeoutHandler); RegisterTimeout(IDLE_STATS_UPDATE_TIMEOUT, IdleStatsUpdateTimeoutHandler); + RegisterTimeout(ANYTIME_STATS_UPDATE_TIMEOUT, + AnytimeStatsUpdateTimeoutHandler); } /* diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index f16f35659b9..84e698da214 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -96,6 +96,7 @@ extern PGDLLIMPORT volatile sig_atomic_t IdleSessionTimeoutPending; extern PGDLLIMPORT volatile sig_atomic_t ProcSignalBarrierPending; extern PGDLLIMPORT volatile sig_atomic_t LogMemoryContextPending; extern PGDLLIMPORT volatile sig_atomic_t IdleStatsUpdateTimeoutPending; +extern PGDLLIMPORT volatile sig_atomic_t AnytimeStatsUpdateTimeoutPending; extern PGDLLIMPORT volatile sig_atomic_t CheckClientConnectionPending; extern PGDLLIMPORT volatile sig_atomic_t ClientConnectionLost; diff --git a/src/include/pgstat.h b/src/include/pgstat.h index fff7ecc2533..b340a680614 100644 --- a/src/include/pgstat.h +++ b/src/include/pgstat.h @@ -35,6 +35,9 @@ /* Default directory to store temporary statistics data in */ #define PG_STAT_TMP_DIR "pg_stat_tmp" +/* Minimum interval non-forced stats flushes */ +#define PGSTAT_MIN_INTERVAL 1000 + /* Values for track_functions GUC variable --- order is significant! */ typedef enum TrackFunctionsLevel { @@ -533,8 +536,21 @@ extern void pgstat_initialize(void); /* Functions called from backends */ extern long pgstat_report_stat(bool force); +extern void pgstat_report_anytime_stat(bool force); extern void pgstat_force_next_flush(void); +/* + * Schedule the next anytime stats update timeout. + * + * This should be called whenever accumulating statistics that support + * FLUSH_ANYTIME flushing mode. + */ +#define pgstat_schedule_anytime_update() \ + do { \ + if (IsUnderPostmaster && !get_timeout_active(ANYTIME_STATS_UPDATE_TIMEOUT)) \ + enable_timeout_after(ANYTIME_STATS_UPDATE_TIMEOUT, PGSTAT_MIN_INTERVAL); \ + } while (0) + extern void pgstat_reset_counters(void); extern void pgstat_reset(PgStat_Kind kind, Oid dboid, uint64 objid); extern void pgstat_reset_of_kind(PgStat_Kind kind); diff --git a/src/include/utils/pgstat_internal.h b/src/include/utils/pgstat_internal.h index 9b8fbae00ed..607f4255268 100644 --- a/src/include/utils/pgstat_internal.h +++ b/src/include/utils/pgstat_internal.h @@ -224,6 +224,19 @@ typedef struct PgStat_SubXactStatus PgStat_TableXactStatus *first; /* head of list for this subxact */ } PgStat_SubXactStatus; +/* + * Flush mode for statistics kinds. + * + * FLUSH_AT_TXN_BOUNDARY has to be the first because we want it to be the + * default value. + */ +typedef enum PgStat_FlushMode +{ + FLUSH_AT_TXN_BOUNDARY, /* All fields can only be flushed at + * transaction boundary */ + FLUSH_ANYTIME, /* All fields can be flushed anytime, + * including within transactions */ +} PgStat_FlushMode; /* * Metadata for a specific kind of statistics. @@ -251,6 +264,16 @@ typedef struct PgStat_KindInfo */ bool track_entry_count:1; + /* + * The mode of when to flush stats. See PgStat_FlushMode for more details. + * + * This member only has meaning for statistics kinds that accumulate + * pending stats and use flush callbacks. For kinds that write directly to + * shared memory (e.g., archiver, bgwriter, checkpointer), this member has + * no effect. + */ + PgStat_FlushMode flush_mode; + /* * The size of an entry in the shared stats hash table (pointed to by * PgStatShared_HashEntry->body). For fixed-numbered statistics, this is @@ -297,8 +320,10 @@ typedef struct PgStat_KindInfo * For variable-numbered stats: flush pending stats. Required if pending * data is used. See flush_static_cb when dealing with stats data that * that cannot use PgStat_EntryRef->pending. + * + * The anytime_only parameter indicates whether this is an anytime flush. */ - bool (*flush_pending_cb) (PgStat_EntryRef *sr, bool nowait); + bool (*flush_pending_cb) (PgStat_EntryRef *sr, bool nowait, bool anytime_only); /* * For variable-numbered stats: delete pending stats. Optional. @@ -366,8 +391,10 @@ typedef struct PgStat_KindInfo * * "pgstat_report_fixed" needs to be set to trigger the flush of pending * stats. + * + * The anytime_only parameter indicates whether this is an anytime flush. */ - bool (*flush_static_cb) (bool nowait); + bool (*flush_static_cb) (bool nowait, bool anytime_only); /* * For fixed-numbered statistics: Reset All. @@ -677,6 +704,7 @@ extern PgStat_EntryRef *pgstat_fetch_pending_entry(PgStat_Kind kind, extern void *pgstat_fetch_entry(PgStat_Kind kind, Oid dboid, uint64 objid); extern void pgstat_snapshot_fixed(PgStat_Kind kind); +extern void AnytimeStatsUpdateTimeoutHandler(void); /* @@ -696,8 +724,8 @@ extern void pgstat_archiver_snapshot_cb(void); #define PGSTAT_BACKEND_FLUSH_WAL (1 << 1) /* Flush WAL statistics */ #define PGSTAT_BACKEND_FLUSH_ALL (PGSTAT_BACKEND_FLUSH_IO | PGSTAT_BACKEND_FLUSH_WAL) -extern bool pgstat_flush_backend(bool nowait, bits32 flags); -extern bool pgstat_backend_flush_cb(bool nowait); +extern bool pgstat_flush_backend(bool nowait, bits32 flags, bool anytime_only); +extern bool pgstat_backend_flush_cb(bool nowait, bool anytime_only); extern void pgstat_backend_reset_timestamp_cb(PgStatShared_Common *header, TimestampTz ts); @@ -729,7 +757,7 @@ extern void AtEOXact_PgStat_Database(bool isCommit, bool parallel); extern PgStat_StatDBEntry *pgstat_prep_database_pending(Oid dboid); extern void pgstat_reset_database_timestamp(Oid dboid, TimestampTz ts); -extern bool pgstat_database_flush_cb(PgStat_EntryRef *entry_ref, bool nowait); +extern bool pgstat_database_flush_cb(PgStat_EntryRef *entry_ref, bool nowait, bool anytime_only); extern void pgstat_database_reset_timestamp_cb(PgStatShared_Common *header, TimestampTz ts); @@ -737,7 +765,7 @@ extern void pgstat_database_reset_timestamp_cb(PgStatShared_Common *header, Time * Functions in pgstat_function.c */ -extern bool pgstat_function_flush_cb(PgStat_EntryRef *entry_ref, bool nowait); +extern bool pgstat_function_flush_cb(PgStat_EntryRef *entry_ref, bool nowait, bool anytime_only); extern void pgstat_function_reset_timestamp_cb(PgStatShared_Common *header, TimestampTz ts); @@ -745,9 +773,9 @@ extern void pgstat_function_reset_timestamp_cb(PgStatShared_Common *header, Time * Functions in pgstat_io.c */ -extern void pgstat_flush_io(bool nowait); +extern void pgstat_flush_io(bool nowait, bool anytime_only); -extern bool pgstat_io_flush_cb(bool nowait); +extern bool pgstat_io_flush_cb(bool nowait, bool anytime_only); extern void pgstat_io_init_shmem_cb(void *stats); extern void pgstat_io_reset_all_cb(TimestampTz ts); extern void pgstat_io_snapshot_cb(void); @@ -762,7 +790,7 @@ extern void AtEOSubXact_PgStat_Relations(PgStat_SubXactStatus *xact_state, bool extern void AtPrepare_PgStat_Relations(PgStat_SubXactStatus *xact_state); extern void PostPrepare_PgStat_Relations(PgStat_SubXactStatus *xact_state); -extern bool pgstat_relation_flush_cb(PgStat_EntryRef *entry_ref, bool nowait); +extern bool pgstat_relation_flush_cb(PgStat_EntryRef *entry_ref, bool nowait, bool anytime_only); extern void pgstat_relation_delete_pending_cb(PgStat_EntryRef *entry_ref); extern void pgstat_relation_reset_timestamp_cb(PgStatShared_Common *header, TimestampTz ts); @@ -809,7 +837,7 @@ extern PgStatShared_Common *pgstat_init_entry(PgStat_Kind kind, * Functions in pgstat_slru.c */ -extern bool pgstat_slru_flush_cb(bool nowait); +extern bool pgstat_slru_flush_cb(bool nowait, bool anytime_only); extern void pgstat_slru_init_shmem_cb(void *stats); extern void pgstat_slru_reset_all_cb(TimestampTz ts); extern void pgstat_slru_snapshot_cb(void); @@ -820,7 +848,7 @@ extern void pgstat_slru_snapshot_cb(void); */ extern void pgstat_wal_init_backend_cb(void); -extern bool pgstat_wal_flush_cb(bool nowait); +extern bool pgstat_wal_flush_cb(bool nowait, bool anytime_only); extern void pgstat_wal_init_shmem_cb(void *stats); extern void pgstat_wal_reset_all_cb(TimestampTz ts); extern void pgstat_wal_snapshot_cb(void); @@ -830,7 +858,7 @@ extern void pgstat_wal_snapshot_cb(void); * Functions in pgstat_subscription.c */ -extern bool pgstat_subscription_flush_cb(PgStat_EntryRef *entry_ref, bool nowait); +extern bool pgstat_subscription_flush_cb(PgStat_EntryRef *entry_ref, bool nowait, bool anytime_only); extern void pgstat_subscription_reset_timestamp_cb(PgStatShared_Common *header, TimestampTz ts); diff --git a/src/include/utils/timeout.h b/src/include/utils/timeout.h index 0965b590b34..10723bb664c 100644 --- a/src/include/utils/timeout.h +++ b/src/include/utils/timeout.h @@ -35,6 +35,7 @@ typedef enum TimeoutId IDLE_SESSION_TIMEOUT, IDLE_STATS_UPDATE_TIMEOUT, CLIENT_CONNECTION_CHECK_TIMEOUT, + ANYTIME_STATS_UPDATE_TIMEOUT, STARTUP_PROGRESS_TIMEOUT, /* First user-definable timeout reason */ USER_TIMEOUT, diff --git a/src/test/modules/test_custom_stats/test_custom_var_stats.c b/src/test/modules/test_custom_stats/test_custom_var_stats.c index 64a8fe63cce..bc0b5d6e0eb 100644 --- a/src/test/modules/test_custom_stats/test_custom_var_stats.c +++ b/src/test/modules/test_custom_stats/test_custom_var_stats.c @@ -83,7 +83,7 @@ static dsa_area *custom_stats_description_dsa = NULL; /* Flush callback: merge pending stats into shared memory */ static bool test_custom_stats_var_flush_pending_cb(PgStat_EntryRef *entry_ref, - bool nowait); + bool nowait, bool anytime_only); /* Serialization callback: write auxiliary entry data */ static void test_custom_stats_var_to_serialized_data(const PgStat_HashKey *key, @@ -150,7 +150,7 @@ _PG_init(void) * Returns false only if nowait=true and lock acquisition fails. */ static bool -test_custom_stats_var_flush_pending_cb(PgStat_EntryRef *entry_ref, bool nowait) +test_custom_stats_var_flush_pending_cb(PgStat_EntryRef *entry_ref, bool nowait, bool anytime_only) { PgStat_StatCustomVarEntry *pending_entry; PgStatShared_CustomVarEntry *shared_entry; diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 241945734ec..1dbc4b96f51 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2271,6 +2271,7 @@ PgStat_Counter PgStat_EntryRef PgStat_EntryRefHashEntry PgStat_FetchConsistency +PgStat_FlushMode PgStat_FunctionCallUsage PgStat_FunctionCounts PgStat_HashKey -- 2.34.1 --C2xzVbxFmVrP7k6O Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v7-0002-Add-anytime-flush-tests-for-custom-stats.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-04 04:47 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-04 04:47 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: --- src/backend/commands/user.c | 246 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 191 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..6d8e2fa8813 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,92 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1181,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1263,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1274,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1582,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1681,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --xlWYagR37YwJXkF8-- ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v3 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Surya Poondla <[email protected]> Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --Dvg/WemKV0I3T/Od Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v3-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
* [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() @ 2026-07-06 08:28 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 148+ messages in thread From: Bertrand Drouvot @ 2026-07-06 08:28 UTC (permalink / raw) DropRole() and GrantRole() resolve the role name to an OID before acquiring LockSharedObject() on the role. A concurrent session that commits a DROP ROLE between the read and the lock acquisition leaves the first session acting on a stale OID. This commit fixes the races by using the same approach as RangeVarGetRelidExtended(): It encapsulates name resolution, permission checking (via a caller-supplied callback), and lock acquisition inside a retry loop driven by SharedInvalidMessageCounter. If invalidation messages arrive between name resolution and locking, indicating concurrent DDL, the function retries. The lock is kept across retries and only released if the name resolves to a different OID on the next iteration. Two callbacks are provided: - RoleNameCallbackForDropRole(): checks current/session user, superuser attribute, and ADMIN OPTION privilege before locking. This is similar to what DropRole() is currently doing before LockSharedObject(). - RoleNameCallbackForGrantRole(): calls check_role_membership_authorization() to verify the current user can grant/revoke membership. This is similar to what GrantRole() is currently doing before calling AddRoleMems()/DelRoleMems(). DropRole() and GrantRole() now call RoleNameGetOid() with appropriate lock levels. AlterRole() does not need the fix because it calls CatalogTupleUpdate() on the pg_authid tuple before AddRoleMems(), which blocks a concurrent DROP ROLE. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/aki6fMNLUx6%2BBR8K%40bdtpg --- src/backend/commands/user.c | 248 ++++++++++++++++++++++++++---------- src/include/commands/user.h | 9 ++ 2 files changed, 193 insertions(+), 64 deletions(-) 95.5% src/backend/commands/ 4.4% src/include/commands/ diff --git a/src/backend/commands/user.c b/src/backend/commands/user.c index be11c49f919..5b869e91c17 100644 --- a/src/backend/commands/user.c +++ b/src/backend/commands/user.c @@ -34,6 +34,7 @@ #include "miscadmin.h" #include "port/pg_bitutils.h" #include "storage/lmgr.h" +#include "storage/sinval.h" #include "utils/acl.h" #include "utils/builtins.h" #include "utils/catcache.h" @@ -116,6 +117,10 @@ static void plan_recursive_revoke(CatCList *memlist, bool revoke_admin_option_only, DropBehavior behavior); static void InitGrantRoleOptions(GrantRoleOptions *popt); +static void RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); +static void RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); /* Check if current user has createrole privileges */ @@ -126,6 +131,94 @@ have_createrole_privilege(void) } +/* + * RoleNameGetOid + * Given a role name, look up its OID, lock it, and return the OID. + * + * This follows the same pattern as RangeVarGetRelidExtended(): + * name resolution, permission check (via callback), and lock acquisition are + * performed inside a retry loop. If invalidation messages arrive during the + * process (indicating concurrent DDL), we retry to ensure the name still + * resolves to the same OID. + * + * The callback is invoked before locking, giving callers a chance to check + * permissions. It receives the current rolename, the resolved OID, the + * previous OID (InvalidOid on first iteration), and a caller-supplied arg. + * If the callback raises an error, the function aborts without locking. + * + * If missing_ok is true and the role does not exist, returns InvalidOid. + * Otherwise, raises an error. + */ +Oid +RoleNameGetOid(const char *rolename, LOCKMODE lockmode, bool missing_ok, + RoleNameGetOidCallback callback, void *callback_arg) +{ + uint64 inval_count; + Oid roleid; + Oid oldroleid = InvalidOid; + bool retry = false; + + for (;;) + { + /* + * Remember the current invalidation count so we can detect concurrent + * DDL after locking. + */ + inval_count = SharedInvalidMessageCounter; + + /* Look up the role name */ + roleid = get_role_oid(rolename, true); + + if (!OidIsValid(roleid)) + { + if (retry) + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + if (!missing_ok) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + return InvalidOid; + } + + /* + * Invoke caller-supplied callback before locking. This is a good + * place to check permissions: we haven't taken the lock yet, but we + * know the OID we intend to lock. If concurrent DDL changes things, + * the callback will be invoked again on the next iteration. + */ + if (callback) + callback(rolename, roleid, oldroleid, callback_arg); + + /* + * If upon retry we get back the same OID, the invalidation messages + * did not change the final answer. So we're done. + * + * If we got a different OID, we've locked the role that used to have + * this name rather than the one that does now. Release the old lock. + */ + if (retry) + { + if (roleid == oldroleid) + break; + UnlockSharedObject(AuthIdRelationId, oldroleid, 0, lockmode); + } + + /* Lock the role */ + LockSharedObject(AuthIdRelationId, roleid, 0, lockmode); + + /* If no invalidation messages were processed, we're done */ + if (inval_count == SharedInvalidMessageCounter) + break; + + /* Something may have changed, retry */ + retry = true; + oldroleid = roleid; + } + + return roleid; +} + + /* * CREATE ROLE */ @@ -1090,6 +1183,59 @@ AlterRoleSet(AlterRoleSetStmt *stmt) } +/* + * Before acquiring a role lock for DROP ROLE, check that the role is not the + * current/session user and that the caller has sufficient privileges to drop it. + */ +static void +RoleNameCallbackForDropRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + HeapTuple tuple; + Form_pg_authid roleform; + + if (roleid == GetUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetOuterUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("current user cannot be dropped"))); + if (roleid == GetSessionUserId()) + ereport(ERROR, + (errcode(ERRCODE_OBJECT_IN_USE), + errmsg("session user cannot be dropped"))); + + tuple = SearchSysCache1(AUTHOID, ObjectIdGetDatum(roleid)); + if (!HeapTupleIsValid(tuple)) + ereport(ERROR, + (errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("role \"%s\" does not exist", rolename))); + + roleform = (Form_pg_authid) GETSTRUCT(tuple); + + /* + * For safety's sake, we allow createrole holders to drop ordinary roles + * but not superuser roles, and only if they also have ADMIN OPTION. + */ + if (roleform->rolsuper && !superuser()) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", + "SUPERUSER", "SUPERUSER"))); + if (!is_admin_of_role(GetUserId(), roleid)) + ereport(ERROR, + (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied to drop role"), + errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", + "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); + + ReleaseSysCache(tuple); +} + + /* * DROP ROLE */ @@ -1119,9 +1265,7 @@ DropRole(DropRoleStmt *stmt) { RoleSpec *rolspec = lfirst(item); char *role; - HeapTuple tuple, - tmp_tuple; - Form_pg_authid roleform; + HeapTuple tmp_tuple; ScanKeyData scankey; SysScanDesc sscan; Oid roleid; @@ -1132,71 +1276,27 @@ DropRole(DropRoleStmt *stmt) errmsg("cannot use special role specifier in DROP ROLE"))); role = rolspec->rolename; - tuple = SearchSysCache1(AUTHNAME, PointerGetDatum(role)); - if (!HeapTupleIsValid(tuple)) - { - if (!stmt->missing_ok) - { - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("role \"%s\" does not exist", role))); - } - else - { - ereport(NOTICE, - (errmsg("role \"%s\" does not exist, skipping", - role))); - } + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP or ALTER commits between name + * resolution and lock acquisition. + */ + roleid = RoleNameGetOid(role, AccessExclusiveLock, stmt->missing_ok, + RoleNameCallbackForDropRole, NULL); + if (!OidIsValid(roleid)) + { + /* missing_ok case: role doesn't exist */ + ereport(NOTICE, + (errmsg("role \"%s\" does not exist, skipping", + role))); continue; } - roleform = (Form_pg_authid) GETSTRUCT(tuple); - roleid = roleform->oid; - - if (roleid == GetUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetOuterUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("current user cannot be dropped"))); - if (roleid == GetSessionUserId()) - ereport(ERROR, - (errcode(ERRCODE_OBJECT_IN_USE), - errmsg("session user cannot be dropped"))); - - /* - * For safety's sake, we allow createrole holders to drop ordinary - * roles but not superuser roles, and only if they also have ADMIN - * OPTION. - */ - if (roleform->rolsuper && !superuser()) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute may drop roles with the %s attribute.", - "SUPERUSER", "SUPERUSER"))); - if (!is_admin_of_role(GetUserId(), roleid)) - ereport(ERROR, - (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), - errmsg("permission denied to drop role"), - errdetail("Only roles with the %s attribute and the %s option on role \"%s\" may drop this role.", - "CREATEROLE", "ADMIN", NameStr(roleform->rolname)))); - /* DROP hook for the role being removed */ InvokeObjectDropHook(AuthIdRelationId, roleid, 0); - /* Don't leak the syscache tuple */ - ReleaseSysCache(tuple); - - /* - * Lock the role, so nobody can add dependencies to her while we drop - * her. We keep the lock until the end of transaction. - */ - LockSharedObject(AuthIdRelationId, roleid, 0, AccessExclusiveLock); - /* * If there is a pg_auth_members entry that has one of the roles to be * dropped as the roleid or member, it should be silently removed, but @@ -1484,6 +1584,21 @@ RenameRole(const char *oldname, const char *newname) return address; } +/* + * Before acquiring a role lock for GRANT/REVOKE, check that the current user + * has authorization to grant/revoke membership in the specified role. + */ +static void +RoleNameCallbackForGrantRole(const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg) +{ + bool is_grant = *((bool *) callback_arg); + Oid currentUserId = GetUserId(); + + check_role_membership_authorization(currentUserId, roleid, is_grant); +} + + /* * GrantRoleStmt * @@ -1568,9 +1683,14 @@ GrantRole(ParseState *pstate, GrantRoleStmt *stmt) (errcode(ERRCODE_INVALID_GRANT_OPERATION), errmsg("column names cannot be included in GRANT/REVOKE ROLE"))); - roleid = get_role_oid(rolename, false); - check_role_membership_authorization(currentUserId, - roleid, stmt->is_grant); + /* + * Use RoleNameGetOid to resolve the name, check permissions, and lock + * the role atomically with a retry loop. This prevents race + * conditions where a concurrent DROP commits between name resolution + * and lock acquisition. + */ + roleid = RoleNameGetOid(rolename, ShareUpdateExclusiveLock, false, + RoleNameCallbackForGrantRole, &stmt->is_grant); if (stmt->is_grant) AddRoleMems(currentUserId, rolename, roleid, stmt->grantee_roles, grantee_ids, diff --git a/src/include/commands/user.h b/src/include/commands/user.h index 97dcb93791b..17263452250 100644 --- a/src/include/commands/user.h +++ b/src/include/commands/user.h @@ -21,6 +21,15 @@ extern PGDLLIMPORT int Password_encryption; /* values from enum PasswordType */ extern PGDLLIMPORT char *createrole_self_grant; +/* Callback for RoleNameGetOid, invoked after name resolution but before locking */ +typedef void (*RoleNameGetOidCallback) (const char *rolename, Oid roleid, + Oid oldroleid, void *callback_arg); + +extern Oid RoleNameGetOid(const char *rolename, LOCKMODE lockmode, + bool missing_ok, + RoleNameGetOidCallback callback, + void *callback_arg); + /* Hook to check passwords in CreateRole() and AlterRole() */ typedef void (*check_password_hook_type) (const char *username, const char *shadow_pass, PasswordType password_type, Datum validuntil_time, bool validuntil_null); -- 2.34.1 --cOHldBwZEf1OqVbN Content-Type: text/x-diff; charset=us-ascii Content-Disposition: attachment; filename="v2-0002-Protect-role-resolution-in-roleSpecsToIds-against.patch" ^ permalink raw reply [nested|flat] 148+ messages in thread
end of thread, other threads:[~2026-07-06 08:28 UTC | newest] Thread overview: 148+ messages (download: mbox mbox.gz follow: Atom feed) -- links below jump to the message on this page -- 2026-01-05 09:41 [PATCH v7 1/5] Add pgstat_report_anytime_stat() for periodic stats flushing Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-04 04:47 [PATCH v1] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v3 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]> 2026-07-06 08:28 [PATCH v2 1/2] Add RoleNameGetOid() with invalidation-based retry loop for DropRole()/GrantRole() Bertrand Drouvot <[email protected]>
This inbox is served by agora; see mirroring instructions for how to clone and mirror all data and code used for this inbox