Date: Thu, 21 Mar 2024 20:07:14 -0400
From: Bruce Momjian <bruce@momjian.us>
To: Thomas Munro <thomas.munro@gmail.com>
Cc: walther@technowledgy.de, Peter Eisentraut <peter@eisentraut.org>,
	Christophe Pettus <xof@thebuild.com>,
	Andrew Dunstan <andrew@dunslane.net>,
	PostgreSQL Bugs <pgsql-bugs@lists.postgresql.org>,
	Tom Lane <tgl@sss.pgh.pa.us>
Subject: Re: Regression tests fail with musl libc because libpq.so can't be
 loaded
Message-ID: <ZfzLsiXbguTDatiZ@momjian.us>
References: <4093813.1710899588@sss.pgh.pa.us>
 <CA+hUKGKNK5V8XwwJJZm36s3EUy8V51xu4XiE8=26n=Wq3OGd4A@mail.gmail.com>
 <4095435.1710900216@sss.pgh.pa.us>
 <CA+hUKGK2cSx3dveMqBD+E9N=zJ0WawBcbk_jnPqmdWBM0DT_3Q@mail.gmail.com>
 <477fa513-6bf0-456d-8d24-1e22a19df671@technowledgy.de>
 <ZfrmN4WHuJgD95fw@momjian.us>
 <CA+hUKGJzC4rr2Yr=QZyx9W83Wv=TVJHndKF1anjNLNXVUM7TqQ@mail.gmail.com>
 <f98cd8de-1c66-491a-8409-e62c09932080@technowledgy.de>
 <36690fcf-f0c8-4d60-af49-6a7efa0e2b14@technowledgy.de>
 <CA+hUKGJAkTZUa9JGr-zsadaqsN=K0P71t-1L2t2=MQFDMc4GuQ@mail.gmail.com>
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="TLvhRbpbklWUu3IV"
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: 
 <CA+hUKGJAkTZUa9JGr-zsadaqsN=K0P71t-1L2t2=MQFDMc4GuQ@mail.gmail.com>
Archived-At: 
 <https://www.postgresql.org/message-id/ZfzLsiXbguTDatiZ%40momjian.us>
Precedence: bulk


--TLvhRbpbklWUu3IV
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit

On Fri, Mar 22, 2024 at 11:42:52AM +1300, Thomas Munro wrote:
> On Fri, Mar 22, 2024 at 9:30 AM <walther@technowledgy.de> wrote:
> > > 4. The upstream (musl) suggestion of which I sent a PoC was to "exec
> > > yourself with a bigger argv".
> >
> > We could do this in HEAD now ...
> 
> Just a thought: if we want to go this way, do we need a new exec call?
>  We already control the initial exec in pg_ctl.c.
> 
> > > Could we even use the exec-approach as the fallback in all other cases
> > > except BSDs and Windows and get rid of PS_USE_NONE?
> >
> > ... and then remove PS_USE_NONE at the beginning of the v18 cycle.
> >
> > This would give a bit more time for those "other systems", which were
> > previously falling back PS_USE_NONE and would then clobber argv, too.
> 
> RIght.  It's unspecified by POSIX whether ps shows changes to those
> strings (and there are systems that don't), but it can't hurt to do so
> anyway, and it'd be better than having a PS_USE_NONE code path that is
> untested.  I dimly recall that it turned out that PS_USE_NONE was
> actually broken for a while without anyone noticing.

Actually, I was thinking the opposite.  Since the musl libc is widely
used, it will be tested, and I don't want to disable process display
updates for such a common platform.

I suggest we use the #ifdef test to continue our existing behavior for
the libraries we know about, like glibc, and use the LD_* process title
truncation hack for libc's we don't recognize.

Attached is a prototype patch which implements this based on previous
patches.

-- 
  Bruce Momjian  <bruce@momjian.us>        https://momjian.us
  EDB                                      https://enterprisedb.com

  Only you can decide what is important to you.

--TLvhRbpbklWUu3IV
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment; filename="proctitle.diff"

diff --git a/src/backend/utils/misc/ps_status.c b/src/backend/utils/misc/ps_status.c
index 5d829e6e483..154862f9fb0 100644
--- a/src/backend/utils/misc/ps_status.c
+++ b/src/backend/utils/misc/ps_status.c
@@ -151,7 +151,32 @@ save_ps_display_args(int argc, char **argv)
 		for (i = 0; environ[i] != NULL; i++)
 		{
 			if (end_of_area + 1 == environ[i])
-				end_of_area = environ[i] + strlen(environ[i]);
+			{
+
+				/* Do this for Linux libc libraries that might be unsafe. */
+#if ! defined(__linux__) || (! defined(__GLIBC__) && defined(__UCLIBC__ ))
+				/*
+				 * If we see variables that the musl runtime linker is known
+				 * to stash pointers to, give up so we don't break later calls
+				 * to dlopen().
+				 */
+				if (strstr(environ[i], "LD_LIBRARY_PATH=") == environ[i] ||
+					strstr(environ[i], "LD_PRELOAD=") == environ[i])
+				{
+					/*
+					 * We can overwrite the name, but stop at the equals sign.
+					 * Future loops will not find contiguous space, but we
+					 * don't break early because we want to count the total
+					 * number.
+					 */
+					end_of_area = strchr(environ[i], '=');
+				}
+				else
+#endif
+				{
+					end_of_area = environ[i] + strlen(environ[i]);
+				}
+			}
 		}
 
 		ps_buffer = argv[0];

--TLvhRbpbklWUu3IV--