From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Ahern Subject: Re: [patch net-next 11/17] ipv6: fib: Allow non-FIB users to take reference on route Date: Wed, 19 Jul 2017 09:49:37 -0600 Message-ID: <606f31c5-c5f9-193a-9527-6e47f827dd75@gmail.com> References: <20170719070232.28457-1-jiri@resnulli.us> <20170719070232.28457-12-jiri@resnulli.us> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 7bit Cc: davem@davemloft.net, idosch@mellanox.com, mlxsw@mellanox.com, roopa@cumulusnetworks.com, nikolay@cumulusnetworks.com, kafai@fb.com, hannes@stressinduktion.org, yoshfuji@linux-ipv6.org, edumazet@google.com, yanhaishuang@cmss.chinamobile.com To: Jiri Pirko , netdev@vger.kernel.org Return-path: Received: from mail-pf0-f194.google.com ([209.85.192.194]:36545 "EHLO mail-pf0-f194.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755319AbdGSPtj (ORCPT ); Wed, 19 Jul 2017 11:49:39 -0400 Received: by mail-pf0-f194.google.com with SMTP id a11so312618pfj.3 for ; Wed, 19 Jul 2017 08:49:39 -0700 (PDT) In-Reply-To: <20170719070232.28457-12-jiri@resnulli.us> Sender: netdev-owner@vger.kernel.org List-ID: On 7/19/17 1:02 AM, Jiri Pirko wrote: > From: Ido Schimmel > > Listeners of the FIB notification chain are expected to be able to take > and release a reference on notified IPv6 routes. This is needed in the > case of drivers capable of offloading these routes to a capable device. > > Since notifications are sent in an atomic context, these drivers need to > take a reference on the route, prepare a work item to offload the route > and release the reference at the end of the work. > > Currently, rt6i_ref is used to indicate in how many FIB nodes a route > appears. Different code paths rely on rt6i_ref being 0 to indicate the > route is no longer used by the FIB. > > For example, whenever a route is deleted or replaced, fib6_purge_rt() is > run to make sure the route is no longer present in intermediate nodes. A > BUG_ON() at the end of the function is executed in case the reference > count isn't 1, as it's only supposed to appear in the non-intermediate > node from which it's going to be deleted. > > Instead of changing the semantics of rt6i_ref, a new reference count is > added, so that external users could also take a reference on routes > without modifying rt6i_ref. > > To make sure external users don't release routes used by the FIB, the > reference count is set to 1 upon creation of a route and decremented by > the FIB upon rt6_release(). > > The reference count is atomic, as it's not protected by any locks and > placed in the 40 bytes hole after the existing rt6i_ref. I'd rather not add another reference counter. Debugging reference leaks is a huge PITA now; adding another counter just makes it worse. Why can't the BUG_ON in fib6_purge_rt be removed since there are other reference holders now?