Re: [PATCH net v2 0/2] netconsole: Fix userdata race condition

[Gustavo Luiz Duarte <gustavold@gmail.com>]

On Wed, Oct 22, 2025 at 10:01 PM Jakub Kicinski <kuba@kernel.org> wrote:
>
> On Wed, 22 Oct 2025 10:39:56 -0700 Gustavo Luiz Duarte wrote:
> > This series fixes a race condition in netconsole's userdata handling
> > where concurrent message transmission could read partially updated
> > userdata fields, resulting in corrupted netconsole output.
> >
> > The first patch adds a selftest that reproduces the race condition by
> > continuously sending messages while rapidly changing userdata values,
> > detecting any torn reads in the output.
> >
> > The second patch fixes the issue by ensuring update_userdata() holds
> > the target_list_lock while updating both extradata_complete and
> > userdata_length, preventing readers from seeing inconsistent state.
> >
> > This targets net tree as it fixes a bug introduced in commit df03f830d099
> > ("net: netconsole: cache userdata formatted string in netconsole_target").
>
> This test is skipping on debug kernel builds in netdev CI.
>
> TAP version 13
> 1..1
> # overriding timeout to 360
> # selftests: drivers/net: netcons_race_userdata.sh
> # socat died before we could check 10000 messages. Skipping test.
> ok 1 selftests: drivers/net: netcons_race_userdata.sh # SKIP
>
> We can't have skips for SW tests.
>
> I think Breno was fighting with a similar problem in the past.
> Not sure what he ended up doing. Maybe just leave it at the print?
> Don't actually mark the test as skipped?
>
> Slightly more advanced option is to only do that if KSFT_MACHINE_SLOW
> per:
> https://github.com/linux-netdev/nipa/wiki/How-to-run-netdev-selftests-CI-style#dealing-with-slow-runners-in-performancelatency-tests

There are two reasons for hitting this skip.
1. The hardcoded 2s timeout in listen_port_and_save_to() expired
2. socat died or failed to start for mysterious reasons

#1 should probably be a success (we ran the test for this long and no
corruption found), and for #2 we can try to return whatever exit code
socat give us.
Retrieving socat return code is a bit tricky because we are running it
in a subshell, but we can save it in a temp file.

I can also send a follow up patch to use a longer timeout in
listen_port_and_save_to() if KSFT_MACHINE_SLOW


> --
> pw-bot: cr