f_modown and LSM inconsistency (was [PATCH v2 1/4] Landlock: Add signal control)

[Christian Brauner]

On Fri, Aug 09, 2024 at 04:00:41PM GMT, Jann Horn wrote:
> On Fri, Aug 9, 2024 at 3:18 PM Mickaël Salaün <mic at digikod.net> wrote:
> > Talking about f_modown() and security_file_set_fowner(), it looks like
> > there are some issues:
> >
> > On Fri, Aug 09, 2024 at 02:44:06PM +0200, Jann Horn wrote:
> > > On Fri, Aug 9, 2024 at 12:59 PM Mickaël Salaün <mic at digikod.net> wrote:
> >
> > [...]
> >
> > > > BTW, I don't understand why neither SELinux nor Smack use (explicit)
> > > > atomic operations nor lock.
> > >
> > > Yeah, I think they're sloppy and kinda wrong - but it sorta works in
> > > practice mostly because they don't have to do any refcounting around
> > > this?
> > >
> > > > And it looks weird that
> > > > security_file_set_fowner() isn't called by f_modown() with the same
> > > > locking to avoid races.
> > >
> > > True. I imagine maybe the thought behind this design could have been
> > > that LSMs should have their own locking, and that calling an LSM hook
> > > with IRQs off is a little weird? But the way the LSMs actually use the
> > > hook now, it might make sense to call the LSM with the lock held and
> > > IRQs off...
> > >
> >
> > Would it be OK (for VFS, SELinux, and Smack maintainers) to move the
> > security_file_set_fowner() call into f_modown(), especially where
> > UID/EUID are populated.  That would only call security_file_set_fowner()
> > when the fown is actually set, which I think could also fix a bug for
> > SELinux and Smack.
> >
> > Could we replace the uid and euid fields with a pointer to the current
> > credentials?  This would enables LSMs to not copy the same kind of
> > credential informations and save some memory, simplify credential
> > management, and improve consistency.
> 
> To clarify: These two paragraphs are supposed to be two alternative
> options, right? One option is to call security_file_set_fowner() with
> the lock held, the other option is to completely rip out the
> security_file_set_fowner() hook and instead let the VFS provide LSMs
> with the creds they need for the file_send_sigiotask hook?

I think it would be fine to stash the credentials in struct fown_struct
same as we do for struct file itself. Calling security_file_set_fowner()
with the irq lock held seems suboptimal to me. Plus, this also means one
less LSM hook and that seems like a win too.