Re: [REGRESSION][4.13.y][4.14.y][v4.15.y] net: reduce skb_warn_bad_offload() noise

From: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
To: Joseph Salisbury <joseph.salisbury@canonical.com>
Cc: Eric Dumazet <edumazet@google.com>,
	Dmitry Vyukov <dvyukov@google.com>,
	Willem de Bruijn <willemb@google.com>,
	David Miller <davem@davemloft.net>,
	Daniel Borkmann <daniel@iogearbox.net>,
	jakub.kicinski@netronome.com, linux@rasmusvillemoes.dk,
	John Fastabend <john.fastabend@gmail.com>,
	me@tobin.cc, idosch@mellanox.com,
	Network Development <netdev@vger.kernel.org>,
	LKML <linux-kernel@vger.kernel.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	stable@vger.kernel.org, 1715609@bugs.launchpad.net
Subject: Re: [REGRESSION][4.13.y][4.14.y][v4.15.y] net: reduce skb_warn_bad_offload() noise
Date: Mon, 11 Dec 2017 16:25:26 -0500	[thread overview]
Message-ID: <CAF=yD-L_S312Ryyg5fE3p6kqFus4SKQcgL2zXKiofq-ohiK7bg@mail.gmail.com> (raw)
In-Reply-To: <0a352e7b-0404-2f91-206f-099e2376ab9a@canonical.com>

On Mon, Dec 11, 2017 at 3:35 PM, Joseph Salisbury
<joseph.salisbury@canonical.com> wrote:
> Hi Eric,
>
> A kernel bug report was opened against Ubuntu [0].  It was found that
> reverting the following commit resolved this bug:

The recorded trace in that bug is against 4.10.0 with some backports.
Given that commit b2504a5dbef3 ("net: reduce skb_warn_bad_offload()
noise") is implicated, I guess that that was backported from 4.11-rc1.

The WARN shows

  e1000e: caps=(0x00000030002149a9, 0x0000000000000000)
  len=1701 data_len=1659 gso_size=1480 gso_type=2 ip_summed=0

The numbering changed in 4.14, but for this kernel

  SKB_GSO_UDP = 1 << 1,

so this is a UFO packet with CHECKSUM_NONE. The stack shows

kernel: [570943.494549] skb_warn_bad_offload+0xd1/0x120
kernel: [570943.494550] __skb_gso_segment+0x17d/0x190
kernel: [570943.494564] validate_xmit_skb+0x14f/0x2a0
kernel: [570943.494565] validate_xmit_skb_list+0x43/0x70

so if that patch has been backported, then this must trigger in
__skb_gso_segment on the return path from skb_mac_gso_segment.

Did you backport

commit 8d63bee643f1fb53e472f0e135cae4eb99d62d19
Author: Willem de Bruijn <willemb@google.com>
Date:   Tue Aug 8 14:22:55 2017 -0400

    net: avoid skb_warn_bad_offload false positives on UFO

    skb_warn_bad_offload triggers a warning when an skb enters the GSO
    stack at __skb_gso_segment that does not have CHECKSUM_PARTIAL
    checksum offload set.

    Commit b2504a5dbef3 ("net: reduce skb_warn_bad_offload() noise")
    observed that SKB_GSO_DODGY producers can trigger the check and
    that passing those packets through the GSO handlers will fix it
    up. But, the software UFO handler will set ip_summed to
    CHECKSUM_NONE.

    When __skb_gso_segment is called from the receive path, this
    triggers the warning again.

    Make UFO set CHECKSUM_UNNECESSARY instead of CHECKSUM_NONE. On
    Tx these two are equivalent. On Rx, this better matches the
    skb state (checksum computed), as CHECKSUM_NONE here means no
    checksum computed.

    See also this thread for context:
    http://patchwork.ozlabs.org/patch/799015/

    Fixes: b2504a5dbef3 ("net: reduce skb_warn_bad_offload() noise")
    Signed-off-by: Willem de Bruijn <willemb@google.com>
    Signed-off-by: David S. Miller <davem@davemloft.net>

Note that UFO was removed in 4.14 and that skb_warn_bad_offload
can happen for various types of packets, so there may be multiple
independent bug reports. I'm investigating two other non-UFO reports
just now.