Re: Calls to vfs_setlease() from NFSD code cause unnecessary CAP_LEASE security checks

[Ondrej Mosnacek <omosnace@redhat.com>]

On Fri, Feb 2, 2024 at 5:08 PM Jeff Layton <jlayton@kernel.org> wrote:
>
> On Fri, 2024-02-02 at 16:31 +0100, Ondrej Mosnacek wrote:
> > Hello,
> >
> > In [1] a user reports seeing SELinux denials from NFSD when it writes
> > into /proc/fs/nfsd/threads with the following kernel backtrace:
> >  => trace_event_raw_event_selinux_audited
> >  => avc_audit_post_callback
> >  => common_lsm_audit
> >  => slow_avc_audit
> >  => cred_has_capability.isra.0
> >  => security_capable
> >  => capable
> >  => generic_setlease
> >  => destroy_unhashed_deleg
> >  => __destroy_client
> >  => nfs4_state_shutdown_net
> >  => nfsd_shutdown_net
> >  => nfsd_last_thread
> >  => nfsd_svc
> >  => write_threads
> >  => nfsctl_transaction_write
> >  => vfs_write
> >  => ksys_write
> >  => do_syscall_64
> >  => entry_SYSCALL_64_after_hwframe
> >
> > It seems to me that the security checks in generic_setlease() should
> > be skipped (at least) when called through this codepath, since the
> > userspace process merely writes into /proc/fs/nfsd/threads and it's
> > just the kernel's internal code that releases the lease as a side
> > effect. For example, for vfs_write() there is kernel_write(), which
> > provides a no-security-check equivalent. Should there be something
> > similar for vfs_setlease() that could be utilized for this purpose?
> >
> > [1] https://bugzilla.redhat.com/show_bug.cgi?id=2248830
> >
>
> Thanks for the bug report!
>
> Am I correct that we only want to do this check when someone from
> userland tries to set a lease via fcntl? The rest of the callers are all
> in-kernel callers and I don't think we need to check for any of them. It
> may be simpler to just push this check into the appropriate callers of
> generic_setlease instead.
>
> Hmm now that I look too...it looks like we aren't checking CAP_LEASE on
> filesystems that have their own ->setlease operation. I'll have a look
> at that soon too.

I did briefly check this while analyzing the issue and all of the
setlease fops implementations seemed to be either simple_nosetlease()
or some wrappers around generic_setlease(), which should both be OK.
But it can't hurt to double-check :)

-- 
Ondrej Mosnacek
Senior Software Engineer, Linux Security - SELinux kernel
Red Hat, Inc.