> -----Original Message-----
> From: David Howells [mailto:dhowe...@redhat.com]
> Sent: Wednesday, February 22, 2017 6:32 PM
> To: Reshetova, Elena <elena.reshet...@intel.com>
> Cc: dhowe...@redhat.com; linux-kernel@vger.kernel.org; linux-
> a...@lists.infradead.org; pet...@infradead.org; gre...@linuxfoundation.org;
> Hans Liljestrand <ishkam...@gmail.com>; Kees Cook
> <keesc...@chromium.org>; David Windsor <dwind...@gmail.com>
> Subject: Re: [PATCH 3/4] fs, afs: convert afs_server.usage from atomic_t to
> refcount_t
>
> Elena Reshetova <elena.reshet...@intel.com> wrote:
>
> > refcount_t type and corresponding API should be
> > used instead of atomic_t when the variable is used as
> > a reference counter. This allows to avoid accidental
> > refcounter overflows that might lead to use-after-free
> > situations.
>
> Although I don't see an assertion for this (the window is too small), it is
> possible for a dead server record to get resurrected. Take a look at
> afs_put_server() and note there's a check around the move to the graveyard.
This one is harder to transform correctly using +1 scheme. As far as I
understand, I would need to take
an additional lock in afs_put_server() to accommodate this section:
@@ -230,12 +230,17 @@ void afs_put_server(struct afs_server *server)
_debug("PUT SERVER %d", refcount_read(&server->usage));
- ASSERTCMP(refcount_read(&server->usage), >, 0);
+ ASSERTCMP(refcount_read(&server->usage), >, 1);
- if (likely(!refcount_dec_and_test(&server->usage))) {
+ /* need a write lock for &server the following section */
+ refcount_dec(&server->usage);
+
+ if (likely(refcount_read(&server->usage) > 1)) {
+ spin_unlock(&old_server->fs_lock);
_leave("");
return;
}
+ /* end of lock section */
And smth tells me that you don't want to have any locks here... Right?
Same story is with afs_vlocation_lookup().
Best Regards,
Elena.
>
> So, please hold this patch also.
>
> David
