Feedback-ID: i0fe9450f:Fastmail
Date: Tue, 26 May 2026 14:20:52 +0900
From: Michael Paquier <michael@paquier.xyz>
To: Euler Taveira <euler@eulerto.com>
Cc: Postgres hackers <pgsql-hackers@lists.postgresql.org>,
	=?iso-8859-1?Q?=C1lvaro?= Herrera <alvherre@kurilemu.de>
Subject: Re: NULL pointer dereference in syslogger with load_libraries() and
 -DEXEC_BACKEND at startup
Message-ID: <ahUttFTpHvcunR0y@paquier.xyz>
References: <ahP-JT4ZRPyobnLb@paquier.xyz>
 <ahQGnTqb7i0UUz70@paquier.xyz>
 <746eb248-0787-4633-94fe-f3dbef249f85@app.fastmail.com>
MIME-Version: 1.0
Content-Type: multipart/signed; micalg=pgp-sha512;
	protocol="application/pgp-signature"; boundary="Azo1zuNv3kijY5/m"
Content-Disposition: inline
In-Reply-To: <746eb248-0787-4633-94fe-f3dbef249f85@app.fastmail.com>
Archived-At: 
 <https://www.postgresql.org/message-id/ahUttFTpHvcunR0y%40paquier.xyz>
Precedence: bulk


--Azo1zuNv3kijY5/m
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Mon, May 25, 2026 at 11:39:54PM -0300, Euler Taveira wrote:
> It seems I was too optimistic about this patch. Since the commit 0c8e082f=
ba8
> sets MyBackendType to B_LOGGER earlier, it breaks the following assumptio=
n in
> syslogger.c.
>=20
>         /*
>          * If we're told to write to a structured log file, but it's not =
open,
>          * dump the data to syslogFile (which is always open) instead.  T=
his can
>          * happen if structured output is enabled after postmaster start =
and we've
>          * been unable to open logFile.  There are also race conditions d=
uring a
>          * parameter change whereby backends might send us structured out=
put
>          * before we open the logFile or after we close it.  Writing form=
atted
>          * output to the regular log file isn't great, but it beats dropp=
ing log
>          * output on the floor.
>=20
> It shouldn't assume syslogFile is always open. The send_message_to_server=
_log()
> shouldn't be executing the following code path in this case.

This comment has been rewritten in 2022 when jsonlog was added in
dc686681e079, but this assumption is much older than that:
bff84a547d71, where the syslogger has been made more robust.  And we
assume that the log file is always open, so this change is breaking
an old assumption.

> It could set MyBackendType only if child_type !=3D B_LOGGER during
> launch_backend.c and set it at the same code path as in the past. However=
, I
> consider this solution ugly and hackish.

While thinking about an approach that could allow to keep
0c8e082fba8d, I was wondering whether we should have a boolean flag
that tracks if the log file is opened or not that gets set (we should
not care about the reset) when we are done with its creation, but I'm
feeling that this makes the logic feeble.  We know only rely on
MyBackendType for the job which means to complicate all these checks.
The part that makes me uneasy is that the logging facility should be
robust by design, and simpler is always better IMO.

An exception where we don't set MyBackendType and have an exception
for this corresponding child_type value does not really feel right to
me either..  As a whole, I am not sure to like what has been done
here.
--
Michael

--Azo1zuNv3kijY5/m
Content-Type: application/pgp-signature; name=signature.asc

-----BEGIN PGP SIGNATURE-----
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=Jr9a
-----END PGP SIGNATURE-----

--Azo1zuNv3kijY5/m--