From: "Darrick J. Wong" <djwong@kernel.org>
To: Josef Bacik <josef@toxicpanda.com>
Cc: fstests@vger.kernel.org
Subject: Re: [PATCH] fstests: fix bug with _check_dmesg "some pattern"
Date: Fri, 21 Nov 2025 11:21:25 -0800 [thread overview]
Message-ID: <20251121192125.GP196366@frogsfrogsfrogs> (raw)
In-Reply-To: <16110e3dac5c7476dd5c74bfb50efb4dc1aac5f0.1763747729.git.josef@toxicpanda.com>
On Fri, Nov 21, 2025 at 12:55:34PM -0500, Josef Bacik wrote:
> I've been rigging up Claude to run fstests via an MCP, and part of this
> means it mounts a 9p file system to save the fstests results. This was
> giving me occasional ENODATA errors, which I dug into. Turns out it's a
> race between truncate and read on 9p which is completely normal for 9p,
> but it's uncovering an actual bug in _check_dmesg. The way bash works is
> when you do
>
> cat foo | whatever > foo
>
> bash will setup the various communication channels first, which is the |
> and >, and then it will execute the commands. This means that it
> truncates foo FIRST, and then it launches cat. Which means we'd race
> with truncate and sometimes get data, sometimes get nothing. And even
> worse on 9p we'd get ENODATA because we get a 0 read back when we
> thought the isize was reasonable.
>
> Fix this by changing this case in _check_dmesg to redirect to another
> file, and move that file over $seqres.dmesg to be checked later.
>
> Signed-off-by: Josef Bacik <josef@toxicpanda.com>
> ---
> common/rc | 9 ++++++++-
> 1 file changed, 8 insertions(+), 1 deletion(-)
>
> diff --git a/common/rc b/common/rc
> index a10ac177..947685df 100644
> --- a/common/rc
> +++ b/common/rc
> @@ -4844,7 +4844,14 @@ _check_dmesg()
> # test failures. Custom filters are then applied afterwards.
> _dmesg_since_test_start | _check_dmesg_filter > $seqres.dmesg
> if [ -n "$1" ]; then
> - cat $seqres.dmesg | $1 > $seqres.dmesg
> + # We use $seqres.dmesg.tmp because shell will setup the
> + # redirections first, which means we would truncate
> + # $seqres.dmesg before cat'ing it. This is racy so sometimes it
> + # would work, but othertimes it would silently fail, or in the
> + # case of 9p you'd get ENODATA. We must redirect into a
> + # temporary file and then move it.
> + cat $seqres.dmesg | $1 > $seqres.dmesg.tmp && \
> + mv $seqres.dmesg.tmp $seqres.dmesg
I wonder if this would be simpler put:
if [ -n "$1" ]; then
_dmesg_since_test_start | _check_dmesg_filter | $1 > $seqres.dme
else
_dmesg_since_test_start | _check_dmesg_filter > $seqres.dmesg
fi
Fewer tempfiles, etc.
--D
> fi
>
> grep -E -q -e "kernel BUG at" \
> --
> 2.51.1
>
>
prev parent reply other threads:[~2025-11-21 19:21 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-11-21 17:55 [PATCH] fstests: fix bug with _check_dmesg "some pattern" Josef Bacik
2025-11-21 19:21 ` Darrick J. Wong [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20251121192125.GP196366@frogsfrogsfrogs \
--to=djwong@kernel.org \
--cc=fstests@vger.kernel.org \
--cc=josef@toxicpanda.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox