David Miller <da...@davemloft.net> writes:
> From: Toke Høiland-Jørgensen <t...@redhat.com>
> Date: Thu, 25 Jun 2020 13:55:03 +0200
>
>> From: Ilya Ponetayev <i.ponet...@ndmsystems.com>
>>
>> CAKE was using the return value of tc_skb_protocol() and expecting it to be
>> the IP protocol type. This can fail in the presence of QinQ VLAN tags,
>> making CAKE unable to handle ECN marking and diffserv parsing in this case.
>> Fix this by implementing our own version of tc_skb_protocol(), which will
>> use skb->protocol directly, but also parse and skip over any VLAN tags and
>> return the inner protocol number instead.
>>
>> Also fix CE marking by implementing a version of INET_ECN_set_ce() that
>> uses the same parsing routine.
>>
>> Fixes: ea82511518f4 ("sch_cake: Add NAT awareness to packet classifier")
>> Fixes: b2100cc56fca ("sch_cake: Use tc_skb_protocol() helper for getting
>> packet protocol")
>> Fixes: 046f6fd5daef ("sched: Add Common Applications Kept Enhanced (cake)
>> qdisc")
>> Signed-off-by: Ilya Ponetayev <i.ponet...@ndmsystems.com>
>> [ squash original two patches, rewrite commit message ]
>> Signed-off-by: Toke Høiland-Jørgensen <t...@redhat.com>
>
> First, this is a bug fix and should probably be steered to 'net'.
>
> Also, other users of tc_skb_protocol() are almost certainly hitting a
> similar problem aren't they? Maybe fix this generically.
I think it depends a little on the use case; some callers actually care
about the VLAN tags themselves and handle that specially (e.g.,
act_csum). Whereas others (e.g., sch_dsmark) probably will have the same
issue. I guess I can trying going through them all and figuring out if
there's a more generic solution.
I'll split out the diffserv parsing fixes and send those for your net
tree straight away, then circle back to this one...
-Toke
