[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <166fe7950809101614y20b60745t652183e32703ccc9@mail.gmail.com>
Date: Wed, 10 Sep 2008 16:14:42 -0700
From: "Ranjit Manomohan" <ranjitm@...gle.com>
To: "David Miller" <davem@...emloft.net>
Cc: tgraf@...g.ch, akpm@...ux-foundation.org, kaber@...sh.net,
lizf@...fujitsu.com, menage@...gle.com,
linux-kernel@...r.kernel.org, netdev@...r.kernel.org
Subject: Re: [PATCH 1/2] Traffic control cgroups subsystem
On Wed, Sep 10, 2008 at 4:00 PM, David Miller <davem@...emloft.net> wrote:
> From: "Ranjit Manomohan" <ranjitm@...gle.com>
> Date: Wed, 10 Sep 2008 15:56:55 -0700
>
>> That is correct for ingress, for egress the sk is already available in
>> the skb so should be fine.
>
> That is not something you can rely upon, even for egress, %100 of the time.
>
> Some forms of reallocation and mangling might decide to orphan the SKB
> and thus drop the skb->sk reference before you see the packet. And they
> are absolutely free to do this.
>
> Just grep for skb_orphan().
>
> Therefore, it is absolutely something you should not rely upon for
> correct operation.
Thats fine and we do not rely on this. Those packets would just not be
classified, the cgroup classid is only a hint and used when available
(which is most of the time).
>
> Like I said from the beginning, Thomas's approach is the superior one.
It would leave a lot of packets (like acks) unaccounted for and these
do take up a significant portion of network packets transmitted on a
high speed link). I am ok with Thomas' simpler approach too, just
pointing out that it is not as accurate as the proposed alternative.
-Thanks,
Ranjit
>
--
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
Powered by blists - more mailing lists