[net] net: skbuff: Fix incorrect skb->mac_len adjustment in skb_vlan_push()

In case 'skb_vlan_push' is called on an skb with a hw-accel vlan tag
already present, the existing hw-accel tag is inserted into payload, and
the new given tag is placed as new hw-accel tag.

After the insertion:
 - 'mac_header' is adjusted to point to the new start of the vlan_ethhdr
 - 'data' is adjusted to point to the vlan_hdr portion
   (since packet's payload is the inner 802.1q)

However, 'mac_len' is incorrectly incremented with additional VLAN_HLEN
bytes, resulting in a total value of 18 bytes.

Meaning, when issuing 'skb_push(skb, skb->mac_len)' the data points
to random content, 4 bytes PRIOR the ethhdr location.

This is problematic, as many constructs in the stack are issuing
'skb_push(skb, skb->mac_len)' prior xmit to a device (e.g tcf_mirred,
tcf_bpf, nf_dup_netdev_egress), resulting in bogus frames being
xmitted (having random 4 bytes at start of frame).

For example:

 # ip l add dev d0 type dummy
 # tc filter add dev eth0 parent ffff: pref 1 basic \
     action vlan push protocol 802.1ad id 5 pipe \
     action mirred egress redirect dev d0

Any 802.1q (hw-accel) tagged frames arriving on eth0 are xmitted as
bogus frames on d0; whereas the expected behavior is having QinQ frames.

Fix, removing the unnecessary VLAN_HLEN adjustment of mac_len.

Fixes: 93515d53b1 ("net: move vlan pop/push functions into common code")
Signed-off-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>
Cc: Pravin Shelar <pshelar@ovn.org>
Cc: Jiri Pirko <jiri@mellanox.com>
---

 - David, if patch ok, suggest this goes to -stable

 - Pravin, original push_vlan() code in openvswitch/actions.c prior Jiri
   has moved it into skbuff.c had the following comment:
        /* Update mac_len for subsequent MPLS actions */
        skb->mac_len += VLAN_HLEN;
   Can you please acknowlegde OvS code is also ok with suggested change?

 net/core/skbuff.c | 1 -
 1 file changed, 1 deletion(-)

On Tue, Sep 27, 2016 at 10:31 AM, Shmulik Ladkani
<shmulik.ladkani@gmail.com> wrote:
> In case 'skb_vlan_push' is called on an skb with a hw-accel vlan tag
> already present, the existing hw-accel tag is inserted into payload, and
> the new given tag is placed as new hw-accel tag.
>
> After the insertion:
>  - 'mac_header' is adjusted to point to the new start of the vlan_ethhdr
>  - 'data' is adjusted to point to the vlan_hdr portion
>    (since packet's payload is the inner 802.1q)
>
> However, 'mac_len' is incorrectly incremented with additional VLAN_HLEN
> bytes, resulting in a total value of 18 bytes.
>
> Meaning, when issuing 'skb_push(skb, skb->mac_len)' the data points
> to random content, 4 bytes PRIOR the ethhdr location.
>
> This is problematic, as many constructs in the stack are issuing
> 'skb_push(skb, skb->mac_len)' prior xmit to a device (e.g tcf_mirred,
> tcf_bpf, nf_dup_netdev_egress), resulting in bogus frames being
> xmitted (having random 4 bytes at start of frame).
>
> For example:
>
>  # ip l add dev d0 type dummy
>  # tc filter add dev eth0 parent ffff: pref 1 basic \
>      action vlan push protocol 802.1ad id 5 pipe \
>      action mirred egress redirect dev d0
>
> Any 802.1q (hw-accel) tagged frames arriving on eth0 are xmitted as
> bogus frames on d0; whereas the expected behavior is having QinQ frames.
>
> Fix, removing the unnecessary VLAN_HLEN adjustment of mac_len.
>
> Fixes: 93515d53b1 ("net: move vlan pop/push functions into common code")
> Signed-off-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>
> Cc: Pravin Shelar <pshelar@ovn.org>
> Cc: Jiri Pirko <jiri@mellanox.com>
> ---
>
>  - David, if patch ok, suggest this goes to -stable
>
>  - Pravin, original push_vlan() code in openvswitch/actions.c prior Jiri
>    has moved it into skbuff.c had the following comment:
>         /* Update mac_len for subsequent MPLS actions */
>         skb->mac_len += VLAN_HLEN;
>    Can you please acknowlegde OvS code is also ok with suggested change?
>

OVS MPLS does depends on mac-len to track MPLS header (ref
skb_mpls_header()). Therefore vlan header changes needs to update
mac-len. But after commit 48d2ab609b6bb ("net: mpls: Fixups for GSO")
it does not need to use mac-len to track MPLS header.
To keep simple fix for stable, we can move the skb mac-len adjustment
to OVS module. Later on patch can remove this MPLS dependency on
mac-len for net-next branch.

Hi,

On Tue, 27 Sep 2016 13:04:35 -0700, pshelar@ovn.org wrote:
> 
> OVS MPLS does depends on mac-len to track MPLS header (ref
> skb_mpls_header()). Therefore vlan header changes needs to update
> mac-len.

Thanks for the explanation.
I'm working on a better fix.

Regards,
Shmulik