Return-path: Received: from e5.ny.us.ibm.com ([32.97.182.145]:36734 "EHLO e5.ny.us.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751321AbZFXBoV (ORCPT ); Tue, 23 Jun 2009 21:44:21 -0400 Date: Tue, 23 Jun 2009 18:44:21 -0700 From: "Paul E. McKenney" To: Jesper Dangaard Brouer Cc: "David S. Miller" , netdev@vger.kernel.org, linux-kernel@vger.kernel.org, dougthompson@xmission.com, bluesmoke-devel@lists.sourceforge.net, axboe@kernel.dk, Patrick McHardy , christine.caulfield@googlemail.com, Trond.Myklebust@netapp.com, linux-wireless@vger.kernel.org, johannes@sipsolutions.net, yoshfuji@linux-ipv6.org, shemminger@linux-foundation.org, linux-nfs@vger.kernel.org, bfields@fieldses.org, neilb@suse.de, linux-ext4@vger.kernel.org, tytso@mit.edu, adilger@sun.com, netfilter-devel@vger.kernel.org Subject: Re: [PATCH 00/10] We must use rcu_barrier() on module unload Message-ID: <20090624014421.GN6748@linux.vnet.ibm.com> Reply-To: paulmck@linux.vnet.ibm.com References: <20090623150330.22490.87327.stgit@localhost> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <20090623150330.22490.87327.stgit@localhost> Sender: linux-wireless-owner@vger.kernel.org List-ID: On Tue, Jun 23, 2009 at 05:03:53PM +0200, Jesper Dangaard Brouer wrote: > This patch series is an attempt to cleanup the entire tree, for > potential oops'es during module unload, due to outstanding RCU > callbacks. (My last rcu_barrier patch series only addressed net/). > > If an unloadable module uses RCU callbacks, it need to use > rcu_barrier() so that the module may be safely unloaded. > > For documentation see: > > Paul E. McKenney's Blog > http://paulmck.livejournal.com/7314.html > > http://lwn.net/Articles/217484/ > > Documentation/RCU/rcubarrier.txt > > > Looking through the Linux kernel for call_rcu() users and unloadable > modules I found 10 modules that didn't behave correctly. These look good from an RCU viewpoint, but I am in no better position than is Jesper to analyze the individual modules. From an RCU viewpoint: Acked-by: Paul E. McKenney > Please: MAINTAINERS needs to verify that the module exit code prevent > any new RCU callbacks from being posted (before rcu_barrier() is > called). (I have tried to do this verification, but most of these > module are simply too large and complex for me to verify this within > reasonable time) > > [Overview description, following patch ordering] > > The modules ext4, bridge, mac80211, sunrpc, nfs and ipv6 are fairly > straight forward (maintainers still needs to check for prevent of new > RCU callbacks). > > The module decnet, has disabled its module_exit() (since ^1da177e) but > it still seems relevant to keep the code updated. > > The modules edac_core and cfq-iosched, has implemented their own > open-coded wait_for_completion() scheme, in order to wait for > call_rcu() calls. Maintainers needs to look into removing this code > and using rcu_barrier() instead. > > The module nf_conntrack, has embedded some comments that I would like > Patrick McHardy to look at. As I'm not sure which is are most optimal > place to call rcu_barrier(). The patch probably calls rcu_barrier() > too much, but its a better safe than sorry approach. > > > I have made a patch for each individual module, so objections can be > made on a per module basis. I have Cc'ed all of the patches to the > maintainers of each module (according to the MAINTAINERS file). > > > The patchset is made on top of Linus Torvalds tree (starting on top of > commit f234012f52a3). > > Who wants to pickup these patches? (I usually go through DaveM, but > this also touches subsystems that are not (yet?) under DaveM's > maintainer ship) > > > --- > Jesper Dangaard Brouer (10): > nf_conntrack: Use rcu_barrier(). > cfq-iosched: Uses its own open-coded rcu_barrier. > edac_core: Uses call_rcu() and its own wait_for_completion scheme. > decnet: Use rcu_barrier() on module unload. > ipv6: Use rcu_barrier() on module unload. > nfs: Use rcu_barrier() on module unload. > sunrpc: Use rcu_barrier() on unload. > mac80211: Use rcu_barrier() on unload. > bridge: Use rcu_barrier() instead of syncronize_net() on unload. > ext4: Use rcu_barrier() on module unload. > > > block/cfq-iosched.c | 6 ++++++ > drivers/edac/edac_device.c | 5 +++++ > drivers/edac/edac_mc.c | 5 +++++ > drivers/edac/edac_pci.c | 5 +++++ > fs/ext4/mballoc.c | 4 +++- > fs/nfs/inode.c | 1 + > net/bridge/br.c | 2 +- > net/decnet/af_decnet.c | 6 ++++++ > net/ipv6/af_inet6.c | 2 ++ > net/mac80211/main.c | 2 ++ > net/netfilter/nf_conntrack_core.c | 5 +++++ > net/netfilter/nf_conntrack_standalone.c | 2 ++ > net/sunrpc/sunrpc_syms.c | 1 + > 13 files changed, 44 insertions(+), 2 deletions(-) > > > -- > Best regards, > Jesper Brouer > ComX Networks A/S > Linux Network developer > Cand. Scient Datalog / MSc. > Author of http://adsl-optimizer.dk > LinkedIn: http://www.linkedin.com/in/brouer >