linux-kernel - Re: [PATCH 0/9] net: bridge: reduce multicast checks in fast path

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <fb8b827e-9aee-4aea-8291-30e5e4c539dc@blackwall.org>
Date: Wed, 3 Sep 2025 13:11:50 +0300
From: Nikolay Aleksandrov <razor@...ckwall.org>
To: Linus Lüssing <linus.luessing@...3.blue>
Cc: Jakub Kicinski <kuba@...nel.org>, bridge@...ts.linux.dev,
 netdev@...r.kernel.org, linux-kernel@...r.kernel.org,
 Ido Schimmel <idosch@...dia.com>, Andrew Lunn <andrew+netdev@...n.ch>,
 Simon Horman <horms@...nel.org>, Paolo Abeni <pabeni@...hat.com>,
 Eric Dumazet <edumazet@...gle.com>, "David S . Miller"
 <davem@...emloft.net>, Kuniyuki Iwashima <kuniyu@...gle.com>,
 Stanislav Fomichev <sdf@...ichev.me>, Xiao Liang <shaw.leon@...il.com>
Subject: Re: [PATCH 0/9] net: bridge: reduce multicast checks in fast path

On 9/3/25 02:00, Linus Lüssing wrote:
> On Tue, Sep 02, 2025 at 10:16:04PM +0200, Linus Lüssing wrote:
>> On the other hand, moving the spinlock out of / around
>> __br_multicast_stop() would lead to a sleeping-while-atomic bug
>> when calling timer_delete_sync() in there. And if I were to change
>> these to a timer_delete() I guess there is no way to do the sync
>> part after unlocking? There is no equivalent to something like the
>> flush_workqueue() / drain_workqueue() for workqueues, but for
>> simple timers instead, right?
> 
> I'm wondering if it would be sufficient to use timer_del() on
> .ndo_stop->br_dev_stop()->br_multicast_stop().
> 
> And use timer_del_sync() only on
> .ndo_uninit->br_dev_uninit()-> br_multicast_dev_del()->
> br_multicast_ctx_deinit() and
> br_vlan_put_master()->br_multicast_ctx_deinit().
> 
> 
> So basically only sync / wait for timer callbacks to finish if
> their context is about to be free'd, on bridge or VLAN destruction?

You should make sure the state is sane if the callbacks can be executed after
br_multicast_stop(). There are many timers that can affect different aspects,
it might be doable but needs careful inspection and code ordering.