Re: BUG_ON in skb_segment, after bpf_skb_change_proto was applied

[Shmulik Ladkani <shmulik@metanetworks.com>]

On Tue, 3 Sep 2019 12:23:54 -0400
Willem de Bruijn <willemdebruijn.kernel@gmail.com> wrote:

> This is a lot more code change. Especially for stable fixes that need
> to be backported, a smaller patch is preferable.

Indeed. Thanks for the feedback.

> My suggestion only tested the first frag_skb length. If a list can be
> created where the first frag_skb is head_frag but a later one is not,
> it will fail short. I kind of doubt that.
> 
> By default skb_gro_receive builds GSO skbs that can be segmented
> along the original gso_size boundaries. We have so far only observed
> this issue when messing with gso_size.

The rationale was based on inputs specified in 43170c4e0ba7, where a GRO
skb has a fraglist with different amounts of payloads.

> We can easily refine the test to fall back on to copying only if
> skb_headlen(list_skb) != mss.

I'm concerned this is too generic; innocent skbs may fall victim to our
skb copy fallback. Probably those mentioned in 43170c4e0ba7.

> Alternatively, only on SKB_GSO_DODGY is fine, too.
> 
> I suggest we stick with the two-liner.

OK.
So lets refine your original codition, testing only the first
frag_skb, but also ensuring SKB_GSO_DODGY *and* 'skb_headlen(list_skb) != mss'
(we know existing code DOES work OK for unchanged gso_size, even if frags
have linear, non head_frag, data).

This hits the known, reproducable case of the mentioned BUG_ON, and is
tightly scoped to that case.

If that's agreed, I'll submit a proper patch.

Best,
Shmulik