From mboxrd@z Thu Jan 1 00:00:00 1970 From: Neil Horman Subject: Re: [PATCH] netpoll: protect napi_poll and poll_controller during dev_[open|close] Date: Wed, 30 Jan 2013 16:25:53 -0500 Message-ID: <20130130212553.GA3937@hmsreliant.think-freely.org> References: <1359578665-29248-1-git-send-email-nhorman@tuxdriver.com> <1359580042.2789.53.camel@bwh-desktop.uk.solarflarecom.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: netdev@vger.kernel.org, Ivan Vecera , "David S. Miller" To: Ben Hutchings Return-path: Received: from charlotte.tuxdriver.com ([70.61.120.58]:51838 "EHLO smtp.tuxdriver.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754355Ab3A3VZ6 (ORCPT ); Wed, 30 Jan 2013 16:25:58 -0500 Content-Disposition: inline In-Reply-To: <1359580042.2789.53.camel@bwh-desktop.uk.solarflarecom.com> Sender: netdev-owner@vger.kernel.org List-ID: On Wed, Jan 30, 2013 at 09:07:22PM +0000, Ben Hutchings wrote: > On Wed, 2013-01-30 at 15:44 -0500, Neil Horman wrote: > > Ivan Vercera was recently backporting commit > > 9c13cb8bb477a83b9a3c9e5a5478a4e21294a760 to a RHEL kernel, and I noticed that, > > while this patch protects the tg3 driver from having its ndo_poll_controller > > routine called during device initalization, it does nothing for the driver > > during shutdown. I.e. it would be entirely possible to have the > > ndo_poll_controller method (or subsequently the ndo_poll) routine called for a > > driver in the netpoll path on CPU A while in parallel on CPU B, the ndo_close or > > ndo_open routine could be called. Given that the two latter routines tend to > > initizlize and free many data structures that the former two rely on, the result > > can easily be data corruption or various other crashes. Furthermore, it seems > > that this is potentially a problem with all net drivers that support netpoll, > > and so this should ideally be fixed in a common path. > > > > Fix it by creating a spinlock in the netpoll_info structure, and holding it on > > netpoll_poll_dev, and in dev_close and dev_open. That will prevent the driver > > from getting torn down while we're using it in the netpoll path > > > > I've done some testing on this, flooding a netconsole enabled system with > > messages and ifup/downing the interface. No problems observed > > > > Signed-off-by: Neil Horman > > CC: Ivan Vecera > > CC: "David S. Miller" > > --- > > include/linux/netpoll.h | 1 + > > net/core/dev.c | 16 ++++++++++++++++ > > net/core/netpoll.c | 3 +++ > > 3 files changed, 20 insertions(+) > > > > diff --git a/include/linux/netpoll.h b/include/linux/netpoll.h > > index f54c3bb..bb1d364 100644 > > --- a/include/linux/netpoll.h > > +++ b/include/linux/netpoll.h > > @@ -40,6 +40,7 @@ struct netpoll_info { > > > > int rx_flags; > > spinlock_t rx_lock; > > + spinlock_t napi_lock; > > struct list_head rx_np; /* netpolls that registered an rx_hook */ > > > > struct sk_buff_head neigh_tx; /* list of neigh requests to reply to */ > > diff --git a/net/core/dev.c b/net/core/dev.c > > index a87bc74..18f85e1 100644 > > --- a/net/core/dev.c > > +++ b/net/core/dev.c > > @@ -1307,11 +1307,19 @@ static int __dev_open(struct net_device *dev) > > int dev_open(struct net_device *dev) > > { > > int ret; > > + struct netpoll_info *ni; > > > > if (dev->flags & IFF_UP) > > return 0; > > > > + rcu_read_lock(); > > + ni = rcu_dereference(dev->npinfo); > > + if (ni) > > + spin_lock(&ni->napi_lock); > > ret = __dev_open(dev); > > + if (ni) > > + spin_unlock(&ni->napi_lock); > [...] > > No, you can't call ndo_open and ndo_stop in atomic context. > Crap, you're right. I might deadlock too if we print something while going down via netconsole. I'll change this to a flag to skip the poll_controller routine if we're going down. New patch in the AM. Thanks Neil > Ben. > > -- > Ben Hutchings, Staff Engineer, Solarflare > Not speaking for my employer; that's the marketing department's job. > They asked us to note that Solarflare product names are trademarked. > >