From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751853Ab3FYW6H (ORCPT ); Tue, 25 Jun 2013 18:58:07 -0400 Received: from e32.co.us.ibm.com ([32.97.110.150]:47500 "EHLO e32.co.us.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751244Ab3FYW6F (ORCPT ); Tue, 25 Jun 2013 18:58:05 -0400 Date: Tue, 25 Jun 2013 15:57:59 -0700 From: "Paul E. McKenney" To: Tejun Heo Cc: Dipankar Sarma , Fengguang Wu , "David S. Miller" , Li Zefan , Patrick McHardy , linux-kernel@vger.kernel.org Subject: Re: [PATCH] rculist: list_first_or_null_rcu() should use list_entry_rcu() Message-ID: <20130625225759.GI3828@linux.vnet.ibm.com> Reply-To: paulmck@linux.vnet.ibm.com References: <20130621003244.GD11837@mtj.dyndns.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20130621003244.GD11837@mtj.dyndns.org> User-Agent: Mutt/1.5.21 (2010-09-15) X-TM-AS-MML: No X-Content-Scanned: Fidelis XPS MAILER x-cbid: 13062522-5406-0000-0000-000009DB7842 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Jun 20, 2013 at 05:32:44PM -0700, Tejun Heo wrote: > list_first_or_null() should test whether the list is empty and return > pointer to the first entry if not in a RCU safe manner. It's broken > in two ways. > > * It compares __kernel @__ptr with __rcu @__next triggering the > following sparse warning. > > net/core/dev.c:4331:17: error: incompatible types in comparison expression (different address spaces) > > * It doesn't perform rcu_dereference*() and computes the entry address > using container_of() directly from the __rcu pointer which is > inconsitent with other rculist interface. As a result, all three > in-kernel users - net/core/dev.c, macvlan, cgroup - are buggy. They > dereference the pointer w/o going through read barrier. > > Fix it by making list_first_or_null_rcu() dereference ->next directly > and then use list_entry_rcu() on it like other rculist accessors. > > Signed-off-by: Tejun Heo > Reported-by: Fengguang Wu > Cc: Dipankar Sarma > Cc: "Paul E. McKenney" > Cc: "David S. Miller" > Cc: Li Zefan > Cc: Patrick McHardy > Cc: stable@vger.kernel.org > --- > include/linux/rculist.h | 5 +++-- > 1 file changed, 3 insertions(+), 2 deletions(-) > > --- a/include/linux/rculist.h > +++ b/include/linux/rculist.h > @@ -267,8 +267,9 @@ static inline void list_splice_init_rcu( > */ > #define list_first_or_null_rcu(ptr, type, member) \ > ({struct list_head *__ptr = (ptr); \ > - struct list_head __rcu *__next = list_next_rcu(__ptr); \ > - likely(__ptr != __next) ? container_of(__next, type, member) : NULL; \ > + struct list_head *__next = __ptr->next; \ > + likely(__ptr != __next) ? \ > + list_entry_rcu(__next, type, member) : NULL; \ > }) > > /** I am a bit uneasy with this, and would feel better if the volatile cast was on the very first fetch of the ->next pointer. Is there some reason why my unease is ill-founded? Thanx, Paul