Horton

Re: Non-replayable WAL records through overflows and >MaxAllocSize lengths

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: Matthias van de Meent <boekewurm+postgres@gmail.com>
Cc: Alvaro Herrera <alvherre@alvh.no-ip.org>, Andres Freund <andres@anarazel.de>, Ian Lawrence Barwick <barwick@gmail.com>, Heikki Linnakangas <hlinnaka@iki.fi>, David Zhang <david.zhang@highgo.ca>, Robert Haas <robertmhaas@gmail.com>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Date: 2023-04-06T23:35:05Z
Lists: pgsql-hackers

Attachments

On Fri, Apr 07, 2023 at 08:08:34AM +0900, Michael Paquier wrote:
> So bumping mainrdata_len to uint64 is actually not entirely in line
> with this code.  Well, it will work because we'd still fail a couple
> of lines down, but perhaps its readability should be improved so as
> we have an extra check in this code path to make sure that
> mainrdata_len is not higher than PG_UINT32_MAX, then use an
> intermediate casted variable before saving the length in the record
> data to make clear that the type of the main static length in
> xloginsert.c is not the same as what a record has?  The v10 I sent
> previously blocked this possibility, but not v11.

So, I was thinking about something like the attached tweaking this
point, the error details a bit, applying an indentation and writing a
commit message...  Matthias?
--
Michael

Commits

  1. Add more protections in WAL record APIs against overflows

  2. Add overflow protection for block-related data in WAL records

  3. Change internal RelFileNode references to RelFileNumber or RelFileLocator.

  4. Revamp the WAL record format.