Date: Mon, 11 Apr 2022 12:40:51 -0700
From: Andres Freund <andres@anarazel.de>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: anisimow.d@gmail.com, pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #17462: Invalid memory access in heapam_tuple_lock
Message-ID: <20220411194051.sovwssdawxoh4m5h@alap3.anarazel.de>
References: <17462-9c98a0f00df9bd36@postgresql.org>
 <286748.1649692519@sss.pgh.pa.us>
 <288544.1649693991@sss.pgh.pa.us>
 <20220411173417.ocwjn5xrkqqjiysl@alap3.anarazel.de>
 <352840.1649699498@sss.pgh.pa.us>
 <20220411190042.o5ebdhc5qsejleff@alap3.anarazel.de>
 <496872.1649705112@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <496872.1649705112@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/20220411194051.sovwssdawxoh4m5h%40alap3.anarazel.de>
Precedence: bulk

Hi,

On 2022-04-11 15:25:12 -0400, Tom Lane wrote:
> Andres Freund <andres@anarazel.de> writes:
> > Another approach is to extend the SatisfiesDirty approach and store the tid of
> > the next tuple version in addition the xmin/xmax we already store. And have
> > heap_fetch() always set t_data to NULL if the snapshot check fails.
>
> That seems like a fairly clean idea, although I think we can't use it
> in the back branches without an ABI break.  We're not going to find a
> TID's worth of padding space in struct SnapshotData.

Right. There's enough space on x86-64, just not contiuous. But not on 32bit
x86, so even if we were willing to live with the ugliness of splitting
ItemPointerData across fields temporarily (which I don't think we would)...

I guess we could put members of SnapshotData into a union with ItemPointerData
that aren't used by InitDirtySnapshot()/HeapTupleSatisfiesDirty().

E.g. ph_node - which fairly fundamentally won't be used by dirty snapshots,
and seems unlikely to be used by any extensions? And even if, it'd cause a
compile-time breakage for such extensions, not a silent ABI breakage...

Greetings,

Andres Freund