On Tue, May 11, 2021 at 01:42:46PM -0400, Willem de Bruijn wrote:
> On Tue, May 11, 2021 at 2:51 AM Dongseok Yi <dseok.yi@xxxxxxxxxxx> wrote:
> >
> > In the forwarding path GRO -> BPF 6 to 4 -> GSO for TCP traffic, the
> > coalesced packet payload can be > MSS, but < MSS + 20.
> > bpf_skb_proto_6_to_4 will increase the MSS and it can be > the payload
> > length. After then tcp_gso_segment checks for the payload length if it
> > is <= MSS. The condition is causing the packet to be dropped.
> >
> > tcp_gso_segment():
> > [...]
> > mss = skb_shinfo(skb)->gso_size;
> > if (unlikely(skb->len <= mss))
> > goto out;
> > [...]
> >
> > Allow to increase MSS when BPF_F_ADJ_ROOM_FIXED_GSO is not set.
> >
> > Fixes: 6578171a7ff0 (bpf: add bpf_skb_change_proto helper)
> > Signed-off-by: Dongseok Yi <dseok.yi@xxxxxxxxxxx>
> >
> > ---
>
> Thanks. Note that this feature does not preclude the alternatives
> discussed, of converting the packet to non-TSO (by clearing gso_size)
> or optionally modifying MSS (but that should get okay from TCP
> experts).
>
> I would target this for bpf-next and drop the Fixes. But that is
> admittedly debatable.
No problem. We can make a better decision under bpf-next.
>
> > net/core/filter.c | 13 +++++++------
> > 1 file changed, 7 insertions(+), 6 deletions(-)
> >
> > v2:
> > per Willem de Bruijn request,
> > checked the flag instead of a generic approach.
> >
> > diff --git a/net/core/filter.c b/net/core/filter.c
> > index cae56d0..a98b28d 100644
> > --- a/net/core/filter.c
> > +++ b/net/core/filter.c
> > @@ -3276,7 +3276,7 @@ static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
> > return 0;
> > }
> >
> > -static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
> > +static int bpf_skb_proto_6_to_4(struct sk_buff *skb, u64 flags)
> > {
> > const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
> > u32 off = skb_mac_header_len(skb);
> > @@ -3305,7 +3305,8 @@ static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
> > }
> >
> > /* Due to IPv4 header, MSS can be upgraded. */
> > - skb_increase_gso_size(shinfo, len_diff);
> > + if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
> > + skb_increase_gso_size(shinfo, len_diff);
> > /* Header must be checked, and gso_segs recomputed. */
> > shinfo->gso_type |= SKB_GSO_DODGY;
> > shinfo->gso_segs = 0;
> > @@ -3317,7 +3318,7 @@ static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
> > return 0;
> > }
> >
> > -static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
> > +static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto, u64 flags)
> > {
> > __be16 from_proto = skb->protocol;
> >
> > @@ -3327,7 +3328,7 @@ static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
> >
> > if (from_proto == htons(ETH_P_IPV6) &&
> > to_proto == htons(ETH_P_IP))
> > - return bpf_skb_proto_6_to_4(skb);
> > + return bpf_skb_proto_6_to_4(skb, flags);
> >
> > return -ENOTSUPP;
> > }
> > @@ -3337,7 +3338,7 @@ BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
> > {
> > int ret;
> >
> > - if (unlikely(flags))
> > + if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO)))
> > return -EINVAL;
>
> Once allowing this flag, please immediately support it for both
> bpf_skb_proto_6_to_4 and bpf_skb_4_to_6.
>
> We cannot do that later if we ignore the second case now.
I will make v3 for both 6_to_4 and 4_to_6.
>
>
> > /* General idea is that this helper does the basic groundwork
> > @@ -3357,7 +3358,7 @@ BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
> > * that. For offloads, we mark packet as dodgy, so that headers
> > * need to be verified first.
> > */
> > - ret = bpf_skb_proto_xlat(skb, proto);
> > + ret = bpf_skb_proto_xlat(skb, proto, flags);
> > bpf_compute_data_pointers(skb);
> > return ret;
> > }
> > --
> > 2.7.4
> >
