Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.37 as permitted sender) client-ip=23.128.96.37;
Message-ID: <2c985516-88a3-9fee-dbd1-134aecd323e5@blackwall.org>
Date:   Thu, 28 Sep 2023 15:35:53 +0300
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.5.0
Subject: Re: [PATCH] bridge: MTU auto tuning ignores IFLA_MTU on NEWLINK
Content-Language: en-US
From:   Nikolay Aleksandrov <razor@blackwall.org>
To:     Trent Lloyd <trent.lloyd@canonical.com>,
        Roopa Prabhu <roopa@nvidia.com>,
        "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>,
        Paolo Abeni <pabeni@redhat.com>, Shuah Khan <shuah@kernel.org>
Cc:     linux-kernel@vger.kernel.org, bridge@lists.linux-foundation.org,
        netdev@vger.kernel.org, linux-kselftest@vger.kernel.org
References: <20230927075713.1253681-1-trent.lloyd@canonical.com>
 <3dccacd8-4249-87f8-690c-6083374dc9d1@blackwall.org>
In-Reply-To: <3dccacd8-4249-87f8-690c-6083374dc9d1@blackwall.org>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 9/27/23 11:10, Nikolay Aleksandrov wrote:
> On 9/27/23 10:57, Trent Lloyd wrote:
>> Commit 804b854d374e ("net: bridge: disable bridge MTU auto tuning if it
>> was set manually") disabled auto-tuning of the bridge MTU when the MTU
>> was explicitly set by the user, however that would only happen when the
>> MTU was set after creation. This commit ensures auto-tuning is also
>> disabled when the MTU is set during bridge creation.
>>
>> Currently when the br_netdev_ops br_change_mtu function is called, the
>> flag BROPT_MTU_SET_BY_USER is set. However this function is only called
>> when the MTU is changed after interface creation and is not called if
>> the MTU is specified during creation with IFLA_MTU (br_dev_newlink).
>>
>> br_change_mtu also does not get called if the MTU is set to the same
>> value it currently has, which makes it difficult to work around this
>> issue (especially for the default MTU of 1500) as you have to first
>> change the MTU to some other value and then back to the desired value.
>>
> 
> Yep, I think I also described this in the commit message of my patch.
> 
>> Add new selftests to ensure the bridge MTU is handled correctly:
>>   - Bridge created with user-specified MTU (1500)
>>   - Bridge created with user-specified MTU (2000)
>>   - Bridge created without user-specified MTU
>>   - Bridge created with user-specified MTU set after creation (2000)
>>
>> Regression risk: Any workload which erroneously specified an MTU during
>> creation but accidentally relied upon auto-tuning to a different value
>> may be broken by this change.
>>
> 
> Hmm, you're right. There's a risk of regression. Also it acts 
> differently when set to 1500 as you've mentioned. I think they should 
> act the same, also bridge's fake rtable RTAX_MTU is not set.
> 
>> Link: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2034099
>> Fixes: 804b854d374e ("net: bridge: disable bridge MTU auto tuning if 
>> it was set manually")
>> Signed-off-by: Trent Lloyd <trent.lloyd@canonical.com>
>> ---
> 

So I've been thinking about this and to elaborate - my concerns are two
first is inconsistency between setting MTU at create vs later when it's 
the default (i.e. this way disables auto-tuning, while later it doesn't)
and second is potential breakage as some network managers always set the 
mtu when creating devices. That would suddenly start disabling 
auto-tuning and that will surprise some people which expect it to work.

I think if you make them both act the same and leave out that case with 
a big comment why, this would be good for -net. To fully solve the 
problem without breaking anyone I think we should add control over the 
autotuning flag so it can be turned on/off by the users. That would be 
explicit and will work for all cases, but that is net-next material.

Thanks,
  Nik