Date: Mon, 24 Jun 2024 13:42:21 -0700
From: Noah Misch <noah@leadboat.com>
To: Andrew Dunstan <andrew@dunslane.net>
Cc: pgsql-committers@lists.postgresql.org
Subject: Re: pgsql: Skip some permissions checks on Cygwin
Message-ID: <20240624204221.42.nmisch@google.com>
References: <E1sHjJc-001KCN-H6@gemulon.postgresql.org>
 <20240624010311.0b.nmisch@google.com>
 <fb2c5702-a51d-4d15-a523-83a0443f0df5@dunslane.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <fb2c5702-a51d-4d15-a523-83a0443f0df5@dunslane.net>
User-Agent: Mutt/2.2.12 (2023-09-09)
Archived-At: 
 <https://www.postgresql.org/message-id/20240624204221.42.nmisch%40google.com>
Precedence: bulk

On Mon, Jun 24, 2024 at 10:17:32AM -0400, Andrew Dunstan wrote:
> On 2024-06-23 Su 9:03 PM, Noah Misch wrote:
> > On Thu, Jun 13, 2024 at 12:12:29PM +0000, Andrew Dunstan wrote:
> > > Skip some permissions checks on Cygwin
> > > 
> > > These are checks that are already skipped on other Windows systems.
> > >   	skip "unix-style permissions not supported on Windows", 2
> > > -	  if ($windows_os);
> > > +	  if ($windows_os || $Config::Config{osname} eq 'cygwin');
> > > -	skip "group access not supported on Windows", 3 if ($windows_os);
> > > +	skip "group access not supported on Windows", 3
> > > +	  if ($windows_os || $Config::Config{osname} eq 'cygwin');
> > Cygwin does support Unix-style permissions, so I'm not following the rationale
> > for this change.  Can you say more?
> 
> It seems to have some difficulties with group permissions in some cases. I
> improved the error message a bit, and got a bunch that all look like this:
> 
> build/testrun/initdb/001_initdb/log/regress_log_001_initdb:*main::STDERR/home/Administrator/postgresql/build/testrun/initdb/001_initdb/data/tmp_test_qlhf/data_group/global/6303
> mode must be 0640, not 0600
> build/testrun/initdb/001_initdb/log/regress_log_001_initdb:*main::STDERR/home/Administrator/postgresql/build/testrun/initdb/001_initdb/data/tmp_test_qlhf/data_group/global/pg_control
> mode must be 0640, not 0600
> build/testrun/initdb/001_initdb/log/regress_log_001_initdb:*main::STDERR/home/Administrator/postgresql/build/testrun/initdb/001_initdb/data/tmp_test_qlhf/data_group/global/pg_filenode.map
> mode must be 0640, not 0600
> build/testrun/initdb/001_initdb/log/regress_log_001_initdb:*main::STDERR/home/Administrator/postgresql/build/testrun/initdb/001_initdb/data/tmp_test_qlhf/data_group/base/5
> mode must be 0750, not 0000000000000000000000000000000000000700
> build/testrun/initdb/001_initdb/log/regress_log_001_initdb:*main::STDERR/home/Administrator/postgresql/build/testrun/initdb/001_initdb/data/tmp_test_qlhf/data_group/base/5/112
> mode must be 0640, not 0600
> build/testrun/initdb/001_initdb/log/regress_log_001_initdb:*main::STDERR/home/Administrator/postgresql/build/testrun/initdb/001_initdb/data/tmp_test_qlhf/data_group/base/5/113
> mode must be 0640, not 0600
> build/testrun/initdb/001_initdb/log/regress_log_001_initdb:*main::STDERR/home/Administrator/postgresql/build/testrun/initdb/001_initdb/data/tmp_test_qlhf/data_group/base/5/1247
> mode must be 0640, not 0600
> build/testrun/initdb/001_initdb/log/regress_log_001_initdb:*main::STDERR/home/Administrator/postgresql/build/testrun/initdb/001_initdb/data/tmp_test_qlhf/data_group/base/5/1247_fsm
> mode must be 0640, not 0600
> build/testrun/initdb/001_initdb/log/regress_log_001_initdb:*main::STDERR/home/Administrator/postgresql/build/testrun/initdb/001_initdb/data/tmp_test_qlhf/data_group/base/5/1247_vm
> mode must be 0640, not 0600
> build/testrun/initdb/001_initdb/log/regress_log_001_initdb:*main::STDERR/home/Administrator/postgresql/build/testrun/initdb/001_initdb/data/tmp_test_qlhf/data_group/base/5/1249
> mode must be 0640, not 0600
> 
> 
> Apparently there's some issue with the group None. I don't mind digging
> further if you think it might be worthwhile, but it seemed to me that since

I'll guess digging further is not worthwhile.  Even if it is a PostgreSQL bug,
the consequences seem limited to unintentionally withholding group access.
There's a remote chance you're using a "noacl" mount, where this result could
fit.

> we didn't mind too much skipping these checks on native Windows we wouldn't
> mind on Cygwin either.

The reasons for skipping the test are different on native Windows (does not
provide POSIX file modes) vs. Cygwin (has POSIX file modes, but the
combination of PostgreSQL and Cygwin somehow reaches the wrong mode).

> Up to now we haven't been covering Cygwin with TAP tests, and I'd like to
> change that.

Thanks for doing that.