Re: [PATCH] net: skbuff: Fix length validation in skb_vlan_pop()

[Shmulik Ladkani <shmulik.ladkani@gmail.com>]

On Mon, 19 Sep 2016 14:22:57 +0200, daniel@iogearbox.net wrote:
> On 09/19/2016 08:15 AM, Shmulik Ladkani wrote:
> > On Sun, 18 Sep 2016 13:26:30 -0700, pshelar@ovn.org wrote:
> >> On Sun, Sep 18, 2016 at 3:09 AM, Shmulik Ladkani
> >> <shmulik.ladkani@gmail.com> wrote:
> >>> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> >>> index 1e329d4112..cc2c004838 100644
> >>> --- a/net/core/skbuff.c
> >>> +++ b/net/core/skbuff.c
> >>> @@ -4537,7 +4537,7 @@ int skb_vlan_pop(struct sk_buff *skb)
> >>>          } else {
> >>>                  if (unlikely((skb->protocol != htons(ETH_P_8021Q) &&
> >>>                                skb->protocol != htons(ETH_P_8021AD)) ||
> >>> -                            skb->len < VLAN_ETH_HLEN))
> >>> +                            skb->mac_len < VLAN_ETH_HLEN))
> >>
> >> There is already check in __skb_vlan_pop() to validate skb for a vlan
> >> header. So it is safe to drop this check entirely.
> >
> > Seems validation in '__skb_vlan_pop' has slightly different semantics:
> >
> > 	unsigned int offset = skb->data - skb_mac_header(skb);
> >
> > 	__skb_push(skb, offset);
> > 	err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
> >
> > this pushes 'data' back to mac_header, then makes sure there's sufficient
> > place in skb to _store_ VLAN_ETH_HLEN bytes (by pulling into linear part
> > if needed, or erroring if skb is too small).
> 
> Yes, but this skb_ensure_writable() is needed for doing the memmove anyway.

Had no intention dropping the skb_ensure_writable from __skb_vlan_pop :)

> > There's no guarantee the original mac header was sized VLAN_ETH_HLEN.
> 
> I'm wondering, what happens when you'd call this on tx path, when you'd
> change that to suggested skb->mac_len? Isn't that 0 in such case, thus
> such setups could fail then?

Thanks, I think you're right.

Seems __dev_queue_xmit needs a 'skb_reset_mac_len' right after call to
'skb_reset_mac_header' (or maybe call 'skb_reset_mac_len' from within
skb_reset_mac_header?).

Also, I'm okay with removing the excess 'skb->mac_len < VLAN_ETH_HLEN'
condition if it is agreed that the "tag exists but insufficient to pop"
semantic is no longer wanted.

Regards,
Shmulik