netdev - Re: [PATCH net 1/3] net/sched: act_pedit: fix action bind logic

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 for Android: free password hash cracker in your pocket

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

Message-ID: <a15d21c6-8a88-6c9a-ca7e-77a31ecfbe28@mojatatu.com>
Date:   Sat, 25 Feb 2023 10:38:31 -0300
From:   Pedro Tammela <pctammela@...atatu.com>
To:     Simon Horman <simon.horman@...igine.com>
Cc:     netdev@...r.kernel.org, jhs@...atatu.com, xiyou.wangcong@...il.com,
        jiri@...nulli.us, davem@...emloft.net, edumazet@...gle.com,
        kuba@...nel.org, pabeni@...hat.com, amir@...ai.me,
        dcaratti@...hat.com, willemb@...gle.com, ozsh@...dia.com,
        paulb@...dia.com
Subject: Re: [PATCH net 1/3] net/sched: act_pedit: fix action bind logic

On 25/02/2023 10:08, Simon Horman wrote:
> On Fri, Feb 24, 2023 at 12:00:56PM -0300, Pedro Tammela wrote:
>> The TC architecture allows filters and actions to be created independently.
>> In filters the user can reference action objects using:
>> tc action add action pedit ... index 1
>> tc filter add ... action pedit index 1
>>
>> In the current code for act_pedit this is broken as it checks netlink
>> attributes for create/update before actually checking if we are binding to an
>> existing action.
>>
>> tdc results:
>> 1..69
> 
> ...
> 
> Hi Pedro,
> 
> Thanks for running the tests :)
> 
> I think this patch looks good - though I am still digesting it.
> But I do wonder if you considered adding a test for this condition.

Yes, they are in my backlog to post when net-next reopens.

> 
> Also, what is the failure mode?

When referencing actions via its indexes on filters there would be three 
outcomes:
1 - Action binds to filter (expected)
2 - Action fails netlink parsing in kernel
3 - Action fails parsing in iproute2

I also posted complementary iproute2 patches.

> 
> If it is that user's can't bind actions to filters,  but the kernel behaves
> correctly with configurations it accepts. If so, then perhaps this is more
> of a feature than a fix.

I would argue it's a fix...

> OTOH, perhaps it's a regression wrt the oldest of
> the two patches references below.

...because filters and actions are completely separate TC objects.
There shouldn't be actions that can be created independently but can't 
be really used.

> 
> I've haven't looked at the other patches in this series yet.
> But I expect my comments apply to them too.
> 
>> Fixes: 71d0ed7079df ("net/act_pedit: Support using offset relative to the conventional network headers")
>> Fixes: f67169fef8db ("net/sched: act_pedit: fix WARN() in the traffic path")
>> Reviewed-by: Jamal Hadi Salim <jhs@...atatu.com>
>> Signed-off-by: Pedro Tammela <pctammela@...atatu.com>
> 
> ...