On Tue, Jul 26, 2011 at 01:04:16PM -0700, Hugh Dickins wrote:
> On Thu, 21 Jul 2011, Linus Torvalds wrote:
>
> > Oh, this didn't get marked for stable.
> >
> > It's commit 59430262401b, "vfs: fix race in rcu lookup of pruned
> > dentry", and it applies to 39 too.
> >
> > It turns out 2.6.38 got it right. The bug was introduced by
> > 62a7375e5d77 ("vfs - check non-mountpoint dentry might block in
> > __follow_mount_rcu()").
>
> Mike Waychison observes that although that patch applies to 39
> in principle, it doesn't quite apply in practice, because of the
> "reverse_transit" business in 2.6.39's __follow_mount_rcu().
>
> Below is the obvious backport to 2.6.39-stable, which we believe
> to be correct; but notice how follow_dotdot_rcu() additionally
> calls __follow_mount_rcu() in that release, and it is also
> updating inode without the seqcount care which __d_lookup_rcu()
> exercised.
>
> I think that's okay, that there's pinning implicit in the dotdot
> lookup, which makes it safe. But please don't take my word for it,
> without an ack from Al.
An ACK you have...
_______________________________________________
stable mailing list
stable@linux.kernel.org
http://linux.kernel.org/mailman/listinfo/stable
