Date: Sun, 24 Jul 2022 20:40:40 -0700
From: Nathan Bossart <nathandbossart@gmail.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Kyotaro Horiguchi <horikyota.ntt@gmail.com>, michael@paquier.xyz,
	gurjeet@singh.im, pgsql-hackers@postgresql.org
Subject: Re: Unprivileged user can induce crash by using an SUSET param in
 PGOPTIONS
Message-ID: <20220725034040.GA4085281@nathanxps13>
References: 
 <CABwTF4U9v4u3L2GWU6ZVxK-6NtvEMarrgAj2Vq8=wE_g1Tkncw@mail.gmail.com>
 <CABwTF4WXCbjEpVRq+G+MZeJEXU=2evYMb_DVdzXpWap5cczkcQ@mail.gmail.com>
 <Ytn2ssEKdzPFM3Vf@paquier.xyz>
 <20220722.151600.1094104349120879204.horikyota.ntt@gmail.com>
 <729330.1658516182@sss.pgh.pa.us>
 <20220722223359.GB3998906@nathanxps13>
 <763512.1658529844@sss.pgh.pa.us>
 <20220722225615.GC3998906@nathanxps13>
 <954936.1658597004@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <954936.1658597004@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/20220725034040.GA4085281%40nathanxps13>
Precedence: bulk

On Sat, Jul 23, 2022 at 01:23:24PM -0400, Tom Lane wrote:
> -	/*
> -	 * process any libraries that should be preloaded at backend start (this
> -	 * likewise can't be done until GUC settings are complete)
> -	 */
> -	process_session_preload_libraries();

This patch essentially moveѕ the call to
process_session_preload_libraries() to earlier in PostgresMain().  The
discussion upthread seems to indicate that this is okay.  I did notice that
the log_disconnections handler won't be set up yet, so failures due to
session_preload_libraries won't be logged the same way as before.  Also,
the call to pgstat_report_connect() won't happen.  Neither of these strikes
me as particularly bad, but it seemed worth noting.

> + *	load_session_libraries: TRUE to honor session_preload_libraries

nitpick: Should we call out local_preload_libraries here, too?

> +	if (load_session_libraries)
> +		process_session_preload_libraries();

I noticed that a couple of places check whether whereToSendOutput is set to
DestRemote to determine if this is an interactive session.  Maybe something
like

	if (whereToSendOutput == DestRemote && !am_walsender)

would be a reasonably future-proof way to avoid the need for a new
InitPostgres() argument.

Otherwise, the patch looks good to me.

-- 
Nathan Bossart
Amazon Web Services: https://aws.amazon.com