| Message ID | 1468582987-18990-1-git-send-email-shmulik.ladkani@gmail.com |
|---|---|
| State | Superseded, archived |
| Delegated to: | David Miller |
| Headers | show |
On 15.07.2016 13:43, Shmulik Ladkani wrote: > Given: > - tap0 and vxlan0 are bridged > - vxlan0 stacked on eth0, eth0 having small mtu (e.g. 1400) > > Assume GSO skbs arriving from tap0 having a gso_size as determined by > user-provided virtio_net_hdr (e.g. 1460 corresponding to VM mtu of 1500). > > After encapsulation these skbs have skb_gso_network_seglen that exceed > eth0's ip_skb_dst_mtu. > > These skbs are accidentally passed to ip_finish_output2 AS IS. > Alas, each final segment (segmented either by validate_xmit_skb or by > hardware UFO) would be larger than eth0 mtu. > As a result, those above-mtu segments get dropped on certain networks. > > This behavior is not aligned with the NON-GSO case: > Assume a non-gso 1500-sized IP packet arrives from tap0. After > encapsulation, the vxlan datagram is fragmented normally at the > ip_finish_output-->ip_fragment code path. > > The expected behavior for the GSO case would be segmenting the > "gso-oversized" skb first, then fragmenting each segment according to > dst mtu, and finally passing the resulting fragments to ip_finish_output2. > > 'ip_finish_output_gso' already supports this "Slowpath" behavior, > but it is only considered if IPSKB_FORWARDED is set (which is not set in > the bridged case). > > In order to support the bridged case, we'll mark skbs arriving from an > ingress interface that get udp-encaspulated as "allowed to be fragmented". > > This mark (as well as the original IPSKB_FORWARDED mark) gets tested in > 'ip_finish_output_gso', in order to determine whether validating the > network seglen is needed. > > Note the TUNNEL_DONT_FRAGMENT tun_flag is still honoured (both in the > gso and non-gso cases), which serves users wishing to forbid > fragmentation at the udp tunnel endpoint. > > Cc: Hannes Frederic Sowa <hannes@stressinduktion.org> > Cc: Florian Westphal <fw@strlen.de> > Signed-off-by: Shmulik Ladkani <shmulik.ladkani@gmail.com> I think this is reasonable, thanks! Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
On Mon, 18 Jul 2016 12:06:00 +0200, hannes@stressinduktion.org wrote: > > Signed-off-by: Shmulik Ladkani <shmulik.ladkani@gmail.com> > > I think this is reasonable, thanks! > > Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Thanks for the feedback and assistance. I'll spin a v3 with a tiny coding change. Regards, Shmulik
On 18.07.2016 12:11, Shmulik Ladkani wrote: > On Mon, 18 Jul 2016 12:06:00 +0200, hannes@stressinduktion.org wrote: >>> Signed-off-by: Shmulik Ladkani <shmulik.ladkani@gmail.com> >> >> I think this is reasonable, thanks! >> >> Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org> > > Thanks for the feedback and assistance. > > I'll spin a v3 with a tiny coding change. Ah, then you can make allow_frag bool, please. Thanks, Hannes
diff --git a/include/net/ip.h b/include/net/ip.h index 08f36cd2b8..9742b92dc9 100644 --- a/include/net/ip.h +++ b/include/net/ip.h @@ -47,6 +47,7 @@ struct inet_skb_parm { #define IPSKB_REROUTED BIT(4) #define IPSKB_DOREDIRECT BIT(5) #define IPSKB_FRAG_PMTU BIT(6) +#define IPSKB_FRAG_SEGS BIT(7) u16 frag_max_size; }; diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c index e23f141c9b..18bb7639dd 100644 --- a/net/ipv4/ip_output.c +++ b/net/ipv4/ip_output.c @@ -221,11 +221,15 @@ static int ip_finish_output_gso(struct net *net, struct sock *sk, { netdev_features_t features; struct sk_buff *segs; + int allow_frag; int ret = 0; - /* common case: locally created skb or seglen is <= mtu */ - if (((IPCB(skb)->flags & IPSKB_FORWARDED) == 0) || - skb_gso_validate_mtu(skb, mtu)) + allow_frag = IPCB(skb)->flags & (IPSKB_FORWARDED | IPSKB_FRAG_SEGS); + + /* common case: locally created skb and fragmentation of segments is + * not allowed, or seglen is <= mtu + */ + if (!allow_frag || skb_gso_validate_mtu(skb, mtu)) return ip_finish_output2(net, sk, skb); /* Slowpath - GSO segment length is exceeding the dst MTU. diff --git a/net/ipv4/ip_tunnel_core.c b/net/ipv4/ip_tunnel_core.c index afd6b5968c..9d847c3025 100644 --- a/net/ipv4/ip_tunnel_core.c +++ b/net/ipv4/ip_tunnel_core.c @@ -63,6 +63,7 @@ void iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb, int pkt_len = skb->len - skb_inner_network_offset(skb); struct net *net = dev_net(rt->dst.dev); struct net_device *dev = skb->dev; + int skb_iif = skb->skb_iif; struct iphdr *iph; int err; @@ -72,6 +73,14 @@ void iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb, skb_dst_set(skb, &rt->dst); memset(IPCB(skb), 0, sizeof(*IPCB(skb))); + if (skb_iif && proto == IPPROTO_UDP) { + /* Arrived from an ingress interface and got udp encapuslated. + * The encapsulated network segment length may exceed dst mtu. + * Allow IP Fragmentation of segments. + */ + IPCB(skb)->flags |= IPSKB_FRAG_SEGS; + } + /* Push down and install the IP header. */ skb_push(skb, sizeof(struct iphdr)); skb_reset_network_header(skb);
Given: - tap0 and vxlan0 are bridged - vxlan0 stacked on eth0, eth0 having small mtu (e.g. 1400) Assume GSO skbs arriving from tap0 having a gso_size as determined by user-provided virtio_net_hdr (e.g. 1460 corresponding to VM mtu of 1500). After encapsulation these skbs have skb_gso_network_seglen that exceed eth0's ip_skb_dst_mtu. These skbs are accidentally passed to ip_finish_output2 AS IS. Alas, each final segment (segmented either by validate_xmit_skb or by hardware UFO) would be larger than eth0 mtu. As a result, those above-mtu segments get dropped on certain networks. This behavior is not aligned with the NON-GSO case: Assume a non-gso 1500-sized IP packet arrives from tap0. After encapsulation, the vxlan datagram is fragmented normally at the ip_finish_output-->ip_fragment code path. The expected behavior for the GSO case would be segmenting the "gso-oversized" skb first, then fragmenting each segment according to dst mtu, and finally passing the resulting fragments to ip_finish_output2. 'ip_finish_output_gso' already supports this "Slowpath" behavior, but it is only considered if IPSKB_FORWARDED is set (which is not set in the bridged case). In order to support the bridged case, we'll mark skbs arriving from an ingress interface that get udp-encaspulated as "allowed to be fragmented". This mark (as well as the original IPSKB_FORWARDED mark) gets tested in 'ip_finish_output_gso', in order to determine whether validating the network seglen is needed. Note the TUNNEL_DONT_FRAGMENT tun_flag is still honoured (both in the gso and non-gso cases), which serves users wishing to forbid fragmentation at the udp tunnel endpoint. Cc: Hannes Frederic Sowa <hannes@stressinduktion.org> Cc: Florian Westphal <fw@strlen.de> Signed-off-by: Shmulik Ladkani <shmulik.ladkani@gmail.com> --- v2: Instead of completely removing the IPSKB_FORWARDED condition of 'ip_finish_output_gso' (forcing an expensive 'skb_gso_validate_mtu' on all local traffic), augment the condition to the tunneled usecase, as suggested by Florian and Hannes. include/net/ip.h | 1 + net/ipv4/ip_output.c | 10 +++++++--- net/ipv4/ip_tunnel_core.c | 9 +++++++++ 3 files changed, 17 insertions(+), 3 deletions(-)