netdev - Re: [PATCH RFC net-next] net: vlan/macvlan: count packets properly with gso

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] [day] [month] [year] [list]

Message-ID: <CAM_iQpX3Lre-3gWDK_Grs-pFHcEe5t2EyqfxqFeSVbbAq474-A@mail.gmail.com>
Date:   Thu, 29 Nov 2018 18:07:34 -0800
From:   Cong Wang <xiyou.wangcong@...il.com>
To:     Eric Dumazet <eric.dumazet@...il.com>
Cc:     Alexander Duyck <alexander.h.duyck@...el.com>, dbanerje@...mai.com,
        David Miller <davem@...emloft.net>,
        Jeff Kirsher <jeffrey.t.kirsher@...el.com>,
        Linux Kernel Network Developers <netdev@...r.kernel.org>
Subject: Re: [PATCH RFC net-next] net: vlan/macvlan: count packets properly
 with gso

On Thu, Nov 29, 2018 at 1:46 PM Eric Dumazet <eric.dumazet@...il.com> wrote:
>
>
>
> On 11/29/2018 01:13 PM, Duyck, Alexander H wrote:
>
> > Instead of just checking for the max it might make more sense to do a
> > check using skb_is_gso, and then if true use gso_segs, otherwise just
> > default to 1.
> >
> > Also your bytes are going to be totally messed up as well since the
> > headers are trimmed in the GSO frames. It might be worthwhile to just
> > have a branch based on skb_is_gso that sets the packets and bytes based
> > on the GSO values, and one that sets it for the default case.
> >
> Note that __dev_queue_xmit() is already doing that, no need
> to re-implement in each driver.
>
> It calls qdisc_pkt_len_init(), meaning that drivers can use
> qdisc_skb_cb(skb)->pkt_len instead of skb->len
>
> (Qdisc layers should not modify qdisc_skb_cb(skb)->pkt_len )

It should not modify, but, on the other hand, it is neither supposed
to be used for non-qdisc layer. At very least, you have to audit
each ->ndo_start_xmit() to see if it uses its own skb->cb
first.