Re: syzcaller patch postings...

From: Daniel Axtens <dja@axtens.net>
To: Dmitry Vyukov <dvyukov@google.com>, David Miller <davem@davemloft.net>
Cc: Paolo Abeni <pabeni@redhat.com>,
	syzkaller-bugs@googlegroups.com, netdev <netdev@vger.kernel.org>,
	Florian Westphal <fw@strlen.de>
Subject: Re: syzcaller patch postings...
Date: Fri, 23 Feb 2018 02:09:29 +1100	[thread overview]
Message-ID: <87d10xhxhy.fsf@linkitivity.dja.id.au> (raw)
In-Reply-To: <CACT4Y+Y3hw4Mc4Q1fjv4xjsu-7P+_2tcqTSBgSbEbnEpBd8nag@mail.gmail.com>

Dmitry Vyukov <dvyukov@google.com> writes:

> On Thu, Feb 22, 2018 at 3:35 PM, David Miller <davem@davemloft.net> wrote:
>> From: Dmitry Vyukov <dvyukov@google.com>
>> Date: Thu, 22 Feb 2018 10:58:07 +0100
>>
>>> Do I understand it correctly that if syzbot replies to the CC list
>>> that was in the testing request, it will resolve the problem? So if
>>> netdev wasn't in CC, it will not be added to CC.
>>>
>>> I will go and fix it now.
>>
>> I don't want syzbot to send the patch to netdev, even if it
>> was in the CC: list.
>>
>> And again this goes for netfilter-devel and linux-wireless as
>> well.
>>
>> There is no reason whatsoever for syzbot to ever post an already
>> posted patch back to the list again, even if it was on the CC:
>> list.
>>
>> In fact netdev will be on that CC: list most of the time.
>
> But if the list on CC the first time, then the patch is already on
> Patchwork, right? When syzbot replies it will add In-Reply-To, so this
> should be treated as a comment to the existing patch? Will it still
> cause problems?

You would think that it would treat it as a comment, but it doesn't.  We
treat something as a comment if and only if the subject begins with some
variant of Re:.  An I-R-T is not enough: I think the reasoning was that
people sometimes post their v2 series as replies to their v1 series,
which is bad; but we erred on the side of not losing patches.

There's not really a solid conceptual framework for this - in part
because parsing the sheer variety of mail people post is really, really
hard. Suggestions on a better algorithm are of course welcome.

> How does Patchwork understand that an email contains a patch? Perhaps
> if we make these emails appear as they don't contain a patch, it will
> solve the problem.

We have a pretty sophisticated parser that looks for things that look
like the output of diff. [FWIW, the algorithm is at
https://github.com/getpatchwork/patchwork/blob/master/patchwork/parser.py#L688 ]

You *could* try to trick it: looking at the code, probably the simplest
way would be to replace '---' in '--- a/foo/bar.c' with something else,
but I'm really quite uncomfortable with that.

Regards,
Daniel