On Mon, 2016-07-11 at 09:50 -0700, Linus Torvalds wrote: > On Mon, Jul 11, 2016 at 8:01 AM, Andreas Gruenbacher > <agruenba@xxxxxxxxxx> wrote: > > From: Jeff Layton <jlayton@xxxxxxxxxx> > > > > Currently the two are unioned together, but I don't think that's > > safe. > > Ack. That does look fishy. Almost anything else can probably be > unioned together with the rcu list, but *not* the refcount that we > might want to look at during the rcu grace period. > I looked at unioning the rcu_head it with something else, but there's nothing else that doesn't get peeked at in an rcu-critical section. So I think this is the only real fix, unfortunately. > Al, this is your code (from long long ago). Want to take it through > the vfs tree, or should I just apply direectly? I'd like to have your > ack regardless, but the fix looks ObviouslyCorrect(tm) to me. > I think this is a v4.7 regression. We've freed these things via RCU for a long time, but monkeying with the refcount under the rcu_read_lock is new with the patch in the "Fixes:" line, AFAICT. -- Jeff Layton <jlayton@xxxxxxxxxx> -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
