[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <aRSng1I6l1f7l7EB@infradead.org>
Date: Wed, 12 Nov 2025 07:28:03 -0800
From: Christoph Hellwig <hch@...radead.org>
To: Raphael Pinsonneault-Thibeault <rpthibeault@...il.com>
Cc: cem@...nel.org, djwong@...nel.org, chandanbabu@...nel.org,
bfoster@...hat.com, linux-xfs@...r.kernel.org,
linux-kernel@...r.kernel.org, linux-kernel-mentees@...ts.linux.dev,
skhan@...uxfoundation.org,
syzbot+9f6d080dece587cfdd4c@...kaller.appspotmail.com
Subject: Re: [PATCH] xfs: ensure log recovery buffer is resized to avoid OOB
On Wed, Nov 12, 2025 at 09:10:34AM -0500, Raphael Pinsonneault-Thibeault wrote:
> Fix by removing the check for xlog_rec_header h_version, since the code
> is already within the if(xfs_has_logv2) path. The CRC checksum will
> reject the bad record anyway, this fix is to ensure we can read the
> entire buffer without an OOB.
Thanks for the fix and the very detailed commit message explaining
the logic. I think this should work, but I suspect the better fix
would be to just reject the mount for
h_size > XLOG_HEADER_CYCLE_SIZE && !XLOG_VERSION_2
because the larger h_size can't work for v1 logs, and the log stripe
unit adjustment is also a v2 feature, so it really should not have
been applied even accidentally in mkfs.
> Can xfs_has_logv2() and xlog_rec_header h_version ever disagree?
They should not, but I'm pretty sure if we give syzbot enough time
it'll craft an image doing that :) So we better add sanity checks
for that now.
Powered by blists - more mailing lists