Re: [syzbot] Monthly lsm report (Apr 2026)

[Jarkko Sakkinen <jarkko@kernel.org>]

On Tue, Apr 14, 2026 at 10:02:13AM -0400, Paul Moore wrote:
> > <2> 68      Yes   possible deadlock in keyring_clear (3)
> >                   https://syzkaller.appspot.com/bug?extid=f55b043dacf43776b50c
> 
> Jarkko, David,
> 
> Do we have a fix for the keyring_clear() issue, or is it not a real problem?

Sorry for not meeting the timeline I promised.

Anyhow, let's on the issue.

There's really just two alternatives to resolve [1]:

A. balance_pgdat() acquires keyring semaphore before __fs_reclaim_acquire(),
   and a non-locking-acquiring  aking __keyring_clear() would be called
   inside fscrypt_put_master_key().
B. keyring_clear() is deferred and we accept that quota is not
   immediately released.

Fixing this from the user process side doing kzalloc() is of course unrealistic,
and unstable fix.

So.. I don't think this is keyring issue per se. This is fscrypt issue
mainly, aand depending on whether A or B are used to sort this out,
possibly also kswapd issue.

Or this is my analysis (which could be wrong of course) after couple
hours looking into it.

[1] https://lore.kernel.org/all/68e54915.a00a0220.298cc0.0480.GAE@google.com/T/

BR, Jarkko