[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <8a616c28-c886-4ae7-82d0-cc8e37a9d175@mojatatu.com>
Date: Wed, 7 Feb 2024 18:44:41 -0300
From: Victor Nogueira <victor@...atatu.com>
To: Eric Dumazet <edumazet@...gle.com>
Cc: jhs@...atatu.com, xiyou.wangcong@...il.com, jiri@...nulli.us,
davem@...emloft.net, kuba@...nel.org, pabeni@...hat.com,
netdev@...r.kernel.org, kernel@...atatu.com, pctammela@...atatu.com
Subject: Re: [PATCH net] net/sched: act_mirred: Don't zero blockid when netns
is going down
On 07/02/2024 18:32, Eric Dumazet wrote:
> On Wed, Feb 7, 2024 at 9:54 PM Victor Nogueira <victor@...atatu.com> wrote:
>>
>> While testing tdc with parallel tests for mirred to block we caught an
>> intermittent bug. The blockid was being zeroed out when a parallel net
>> namespace was going down and, thus, giving us an incorrect blockid value(0)
>> whenever we tried to dump the mirred action. Since we don't increment the
>> block refcount in the control path (and only use the ID), we don't need to
>> zero the blockid field whenever the net namespace is going down.
>>
>
> You mention netns being removed, but the issue is more about unregistering
> a net device ?
> Ie the following would also trigger the bug ?
>
> ip link add dev dummy type dummy
> ip link del dummy
Indeed you are right.
I focused on the net namespace part because this is what triggered the
bug in our testing. I'll send a v2 with a better commit message.
cheers,
Victor
Powered by blists - more mailing lists