Re: SECCOMP_RET_USER_NOTIF: listener improvements
From: Christian Brauner
Date: Wed Apr 24 2019 - 11:23:12 EST
On Wed, Apr 24, 2019 at 09:20:01AM -0600, Tycho Andersen wrote:
> On Wed, Apr 24, 2019 at 05:04:26PM +0200, Christian Brauner wrote:
> > Hey everyone,
> > So I was working on making use of the seccomp listener stuff and I
> > stumbled upon a problem. Imagine a scenario where:
> > 1. Task T1 installs Filter F1 and gets and listener fd for that filter FD1
> > 2. T1 sends FD1 via SCM_RIGHTS to task T2
> > T2 now holds a reference to the same underlying struct file as FD1 via FD2
> > 3. T2 registers FD2 in an event loop and starts listening for events
> > 4. T1 exits and wipes FD1
> > Now, T2 still holds a reference to the filter via FD2 which references
> > the same underlying file as FD1 which has the seccomp filter stashed in
> > private_data.
> > So T2 will never get notified that the filter is essentially unused and
> > doesn't know when to exit, i.e. it has no way of telling when T1 and all
> > of its children using the same filter are gone.
> > I think we should have a way to do this
> Since the only way we ever allow creating a struct file * that points
> to a struct seccomp_filter *, if there is a notifier attached, the
> number of tasks still being monitored by a particular filter should be
> filter->usage - 1 (assuming there is a notifier attached). So we could
> augment __put_seccomp_filter() to check for this and send out a
> message with a SECCOMP_NOTIF_FLAG_DEAD flag or something.
> > *or* alternatively have a way to attach a process to an existing
> > filter.
> I also think this wouldn't be too hard, since the struct file * has a
> reference to the filter. So I guess the question is: which of these
> makes more sense?
Or we do both... But overall I'm very much in favor of the option to
attach a task to an existing filter. Re-creating a seccomp context is
extremely brittle and error-prone. This would take a way a major pain