netdev - Re: [PATCH net 1/1] bonding: fix PACKET_ORIGDEV regression on bonding masters

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

Message-ID: <87va2z45zn.fsf@bernat.ch>
Date:   Tue, 08 Jan 2019 08:46:20 -0500
From:   Vincent Bernat <vincent@...nat.ch>
To:     David Miller <davem@...emloft.net>
Cc:     soltys@....info, netdev@...r.kernel.org,
        jay.vosburgh@...onical.com, maheshb@...gle.com,
        chonggangli@...gle.com
Subject: Re: [PATCH net 1/1] bonding: fix PACKET_ORIGDEV regression on bonding masters

 ❦  7 janvier 2019 09:12 -08, David Miller <davem@...emloft.net>:

>> The patch reverts to pre-b89f04c61efe state, so:
>> 
>> - both master and original (via PACKET_ORIGDEV) devices are available
>>   when listening on the master
>> - original device is available when listening directly on one of its
>>   slaves
>> 
>> Reported-by: Vincent Bernat <vincent@...nat.ch>
>> Signed-off-by: Michal Soltys <soltys@....info>

Maybe add:

Fixes: b89f04c61efe ("bonding: deliver link-local packets with skb->dev set to link that packets arrived on")

> It is tiring and disappointing to see the behavior sway back and forth
> multiple times like this.  We are breaking someone, then breaking them
> again if they adjusted to the new behavior.
>
> Pretty much unacceptable.
>
> I'm not applying anything until someone can convince me the full scope
> of the situation and why this time it's "right."

For me, the problem was not visible until the fix introduced for 4.19,
hence the late report of the problem. Before 4.12 (and from 2.6.27 for
non-multicast packets), the behaviour was:

 - ability to receive link-local packets on physical interfaces by
   listening directly to them,
 - ability to receive link-local packets on bonded devices and get the
   original interface using PACKET_ORIGDEV.

The patch introduced in 4.12 doesn't change the first ability and breaks
the second one as the packets are not received on bonded devices
anymore. 

With the fix introduced in 4.19 enables to receive link-local packets on
bonded devices again, but it's still not possible to get the original
interface with PACKET_ORIGDEV.

I don't understand what b89f04c61efe purpose is. My first report was on
November 30th <https://marc.info/?l=linux-netdev&m=154361371507825&w=2>
but I didn't get any additional info, notably how the original interface
is retrieved when listening to a bond device. Without an example of
usage, it's difficult to understand if a revert would break anything.
However, a revert would unbreak users of PACKET_ORIGDEV.
-- 
The human race has one really effective weapon, and that is laughter.
		-- Mark Twain