Re: [RFC PATCH v2 2/3] security: Add PR_CAPBSET_DROP_MASK for process-wide bounding-set drops
From: Christian Brauner
Date: Fri Nov 20 2026 - 08:54:05 EST
On Wed, Nov 18, 2026 at 11:15:17AM +0800, Jinjie Ruan wrote:
> Do I need to resubmit RFC v3 based on your suggestion and address the
> sashiko review issues besides sighand?
>
> Or wait for the results of new discussions?
The signal_struct part has to go. I think this is also what Serge's
question was getting it. And imho it's the real problem in this series.
The caps a task has are described by its struct cred and the whole
kernel relies on a cred being a complete, immutable snapshot of that.
With a pending drop in signal_struct that isn't true anymore.
Anything that stashes creds and uses them later, override_creds(),
io_uring personalities, file->f_cred, LSM hooks, sees a bounding set
the task doesn't actually have.
And every place that looks at the bounding set has to remember to fold
in the signal_struct mask. Any place that forgets is a bypass. That's an
entirely new bug class. So not going to happen, sorry.