netdev - Re: [PATCH net] bonding: fix arp monitoring with vlan slaves

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]

Date:	Fri, 02 Aug 2013 16:59:07 -0700
From:	Jay Vosburgh <fubar@...ibm.com>
To:	Nikolay Aleksandrov <nikolay@...hat.com>
cc:	Eric Dumazet <eric.dumazet@...il.com>,
	David Miller <davem@...emloft.net>, netdev@...r.kernel.org,
	andy@...yhouse.net
Subject: Re: [PATCH net] bonding: fix arp monitoring with vlan slaves

Nikolay Aleksandrov <nikolay@...hat.com> wrote:

>On 08/03/2013 01:17 AM, Eric Dumazet wrote:
>> On Fri, 2013-08-02 at 15:32 -0700, David Miller wrote:
>> 
>>> Handling this specially in bonding isn't really ideal.
>>>
>>> Please either hide this detail in dev_trans_start(), or (preferrably)
>>> have vlan_dev_hard_start_xmit() set the trans_start timestamp
>>> properly thus making this just work for everything.
>> 
>> vlan is LLTX, so setting the timestamp would incur false sharing on
>> multiqueue.
>> 
>Yeah, this statement actually explains a lot for me, thanks :-)
>I knew it was because of the LLTX, but I was wondering about the possible
>reasons for the xmit_lock_owner check.
>So basically the arp monitoring (or any dev_trans_start code) won't work
>with LLTX devices because they don't get their trans_start updated (not the
>txq trans_start nor the dev->trans_start), is this correct ?

	I think what Eric means is that it'll be a performance problem,
because the cache line with the trans_start field will be thrashed
between CPUs as updates compete with each other or inspection from
dev_trans_start.  I suspect it will work from a functional point of view
(unless I'm missing something).

	As a practical matter, it looks like bonding is at present the
only caller of dev_trans_start that passes either (a) a VLAN device, or
(b) a device other than itself (most drivers seem to use it for transmit
timeout detection on their own device).

	Having software devices set trans_start seems unnecessary
(bonding doesn't set it, either), since they'll generally have a
hardware device underneath with a real trans_start value.  It might be
hard to hunt down for tun, though.

>But what if the txqs get bound to a particular CPU, then the txq
>trans_start is okay to be updated I suppose.
>
>Nik
>> But it's true we need a helper, because many callers do
>> 
>> if (dev->priv_flags & IFF_802_1Q_VLAN)
>>        dev = vlan_dev_real_dev(dev);
>> 
>> or the slighly better
>> 
>> if (is_vlan_dev(dev))
>>    dev = vlan_dev_real_dev(dev);

	For just the bonding dev_trans_start() for a VLAN case, one of
the above in dev_trans_start() ought to be sufficient.

	-J

---
	-Jay Vosburgh, IBM Linux Technology Center, fubar@...ibm.com

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html