Support PMTU discovery with bridged UDP tunnels

It's currently possible to bridge Ethernet tunnels carrying IP packets directly to external interfaces without assigning them addresses and routes on the bridged network itself: this is the case for UDP tunnels bridged with a standard bridge or by Open vSwitch.

PMTU discovery is currently broken with those configurations, because the encapsulation effectively decreases the MTU of the link, and while we are able to account for this using PMTU discovery on the lower layer, we don't have a way to relay ICMP or ICMPv6 messages needed by the sender, because we don't have valid routes to it.

On the other hand, as a tunnel endpoint, we can't fragment packets as a general approach: this is for instance clearly forbidden for VXLAN by RFC 7348, section 4.3:

VTEPs MUST NOT fragment VXLAN packets. Intermediate routers may fragment encapsulated VXLAN packets due to the larger frame size. The destination VTEP MAY silently discard such VXLAN fragments.

The same paragraph recommends that the MTU over the physical network accomodates for encapsulations, but this isn't a practical option for complex topologies, especially for typical Open vSwitch use cases.

Further, it states that:

Other techniques like Path MTU discovery (see [RFC1191] and [RFC1981]) MAY be used to address this requirement as well.

Now, PMTU discovery already works for routed interfaces, we get route exceptions created by the encapsulation device as they receive ICMP Fragmentation Needed and ICMPv6 Packet Too Big messages, and we already rebuild those messages with the appropriate MTU and route them back to the sender.

Add the missing bits for bridged cases:

- checks in skb_tunnel_check_pmtu() to understand if it's appropriate to trigger a reply according to RFC 1122 section 3.2.2 for ICMP and RFC 4443 section 2.4 for ICMPv6. This function is already called by UDP tunnels

- a new function generating those ICMP or ICMPv6 replies. We can't reuse icmp_send() and icmp6_send() as we don't see the sender as a valid destination. This doesn't need to be generic, as we don't cover any other type of ICMP errors given that we only provide an encapsulation function to the sender

While at it, make the MTU check in skb_tunnel_check_pmtu() accurate:
we might receive GSO buffers here, and the passed headroom already includes the inner MAC length, so we don't have to account for it a second time (that would imply three MAC headers on the wire, but there are just two).

This issue became visible while bridging IPv6 packets with 4500 bytes of payload over GENEVE using IPv4 with a PMTU of 4000. Given the 50 bytes of encapsulation headroom, we would advertise MTU as 3950, and we would reject fragmented IPv6 datagrams of 3958 bytes size on the wire. We're exclusively dealing with network MTU here, though, so we could get Ethernet frames up to 3964 octets in that case.

v2:
- moved skb_tunnel_check_pmtu() to ip_tunnel_core.c (David Ahern)
- split IPv4/IPv6 functions (David Ahern)

Signed-off-by Stefano Brivio <[email protected]>

---
 drivers/net/bareudp.c     |   5 +-
 drivers/net/geneve.c      |   5 +-
 drivers/net/vxlan.c       |   4 +-
 include/net/dst.h         |  10 --
 include/net/ip_tunnels.h  |   2 +
 net/ipv4/ip_tunnel_core.c | 244 ++++++++++++++++++++++++++++++++++++++
 6 files changed, 254 insertions(+), 16 deletions(-)

Currently, processes sending traffic to a local bridge with an encapsulation device as a port don't get ICMP errors if they exceed the PMTU of the encapsulated link.

David Ahern suggested this as a hack, but it actually looks like the correct solution: when we update the PMTU for a given destination by means of updating or creating a route exception, the encapsulation might trigger this because of PMTU discovery happening either on the encapsulation device itself, or its lower layer. This happens on bridged encapsulations only.

The output interface shouldn't matter, because we already have a valid destination. Drop the output interface restriction from the associated route lookup.

For UDP tunnels, we will now have a route exception created for the encapsulation itself, with a MTU value reflecting its headroom, which allows a bridge forwarding IP packets originated locally to deliver errors back to the sending socket.

The behaviour is now consistent with IPv6 and verified with selftests pmtu_ipv{4,6}_br_{geneve,vxlan}{4,6}_exception introduced later in this series.

v2:
- reset output interface only for bridge ports (David Ahern)
- add and use netif_is_any_bridge_port() helper (David Ahern)

Suggested-by David Ahern <[email protected]>

Signed-off-by Stefano Brivio <[email protected]>

---
 include/linux/netdevice.h | 5 +++++
 net/ipv4/route.c          | 5 +++++
 2 files changed, 10 insertions(+)

tools/testing/selftests/net/pmtu.sh +4 -2

@@ -497,12 +529,19 @@ setup_vxlan_or_geneve() {
 	run_cmd ${ns_a} ip link add ${type}_a type ${type} id 1 ${opts_a} remote ${b_addr} ${opts} || return 1
 	run_cmd ${ns_b} ip link add ${type}_b type ${type} id 1 ${opts_b} remote ${a_addr} ${opts}
 
-	run_cmd ${ns_a} ip addr add ${tunnel4_a_addr}/${tunnel4_mask} dev ${type}_a
-	run_cmd ${ns_b} ip addr add ${tunnel4_b_addr}/${tunnel4_mask} dev ${type}_b
+	if [ -n "${br_if_a}" ]; then
+		run_cmd ${ns_a} ip addr add ${tunnel4_a_addr}/${tunnel4_mask} dev ${br_if_a}
+		run_cmd ${ns_a} ip addr add ${tunnel6_a_addr}/${tunnel6_mask} dev ${br_if_a}
+		run_cmd ${ns_a} ip link set ${type}_a master ${br_if_a}

Quoted 15 lines · Stefano Brivio

tools/testing/selftests/net/pmtu.sh +8 -2

@@ -497,12 +529,19 @@ setup_vxlan_or_geneve() {
 	run_cmd ${ns_a} ip link add ${type}_a type ${type} id 1 ${opts_a} remote ${b_addr} ${opts} || return 1
 	run_cmd ${ns_b} ip link add ${type}_b type ${type} id 1 ${opts_b} remote ${a_addr} ${opts}
 
-	run_cmd ${ns_a} ip addr add ${tunnel4_a_addr}/${tunnel4_mask} dev ${type}_a
-	run_cmd ${ns_b} ip addr add ${tunnel4_b_addr}/${tunnel4_mask} dev ${type}_b
+	if [ -n "${br_if_a}" ]; then
+		run_cmd ${ns_a} ip addr add ${tunnel4_a_addr}/${tunnel4_mask} dev ${br_if_a}
+		run_cmd ${ns_a} ip addr add ${tunnel6_a_addr}/${tunnel6_mask} dev ${br_if_a}
+		run_cmd ${ns_a} ip link set ${type}_a master ${br_if_a}
+	else
+		run_cmd ${ns_a} ip addr add ${tunnel4_a_addr}/${tunnel4_mask} dev ${type}_a
+		run_cmd ${ns_a} ip addr add ${tunnel6_a_addr}/${tunnel6_mask} dev ${type}_a
+	fi
 

tools/testing/selftests/net/pmtu.sh setup_vxlan_or_geneve() { · line 544

Quoted 5 lines

tools/testing/selftests/net/pmtu.sh +2 -1

-	run_cmd ${ns_a} ip addr add ${tunnel6_a_addr}/${tunnel6_mask} dev ${type}_a
+	run_cmd ${ns_b} ip addr add ${tunnel4_b_addr}/${tunnel4_mask} dev ${type}_b
 	run_cmd ${ns_b} ip addr add ${tunnel6_b_addr}/${tunnel6_mask} dev ${type}_b
 
+  

extra newline snuck in

other than that:

Reviewed-by David Ahern <[email protected]>

Currently, PMTU discovery for UDP tunnels only works if packets are routed to the encapsulating interfaces, not bridged.

This results from the fact that we generally don't have valid routes to the senders we can use to relay ICMP and ICMPv6 errors, and makes PMTU discovery completely non-functional for VXLAN and GENEVE ports of both regular bridges and Open vSwitch instances.

If the sender is local, and packets are forwarded to the port by a regular bridge, all it takes is to generate a corresponding route exception on the encapsulating device. The bridge then finds the route exception carrying the PMTU value estimate as it forwards frames, and relays ICMP messages back to the socket of the local sender. Patch 1/6 fixes this case.

If the sender resides on another node, we actually need to reply to IP and IPv6 packets ourselves and send these ICMP or ICMPv6 errors back, using the same encapsulating device. Patch 2/6, based on an original idea by Florian Westphal, adds the needed functionality, while patches 3/6 and 4/6 add matching support for VXLAN and GENEVE.

Finally, 5/6 and 6/6 introduce selftests for all combinations of inner and outer IP versions, covering both VXLAN and GENEVE, with both regular bridges and Open vSwitch instances.

v2: Add helper to check for any bridge port, skip oif check for PMTU
routes for bridge ports only, split IPv4 and IPv6 helpers and functions (all suggested by David Ahern)

It's currently possible to bridge Ethernet tunnels carrying IP packets directly to external interfaces without assigning them addresses and routes on the bridged network itself: this is the case for UDP tunnels bridged with a standard bridge or by Open vSwitch.

PMTU discovery is currently broken with those configurations, because the encapsulation effectively decreases the MTU of the link, and while we are able to account for this using PMTU discovery on the lower layer, we don't have a way to relay ICMP or ICMPv6 messages needed by the sender, because we don't have valid routes to it.

On the other hand, as a tunnel endpoint, we can't fragment packets as a general approach: this is for instance clearly forbidden for VXLAN by RFC 7348, section 4.3:

VTEPs MUST NOT fragment VXLAN packets. Intermediate routers may fragment encapsulated VXLAN packets due to the larger frame size. The destination VTEP MAY silently discard such VXLAN fragments.

The same paragraph recommends that the MTU over the physical network accomodates for encapsulations, but this isn't a practical option for complex topologies, especially for typical Open vSwitch use cases.

Further, it states that:

Other techniques like Path MTU discovery (see [RFC1191] and [RFC1981]) MAY be used to address this requirement as well.

Now, PMTU discovery already works for routed interfaces, we get route exceptions created by the encapsulation device as they receive ICMP Fragmentation Needed and ICMPv6 Packet Too Big messages, and we already rebuild those messages with the appropriate MTU and route them back to the sender.

Add the missing bits for bridged cases:

- checks in skb_tunnel_check_pmtu() to understand if it's appropriate to trigger a reply according to RFC 1122 section 3.2.2 for ICMP and RFC 4443 section 2.4 for ICMPv6. This function is already called by UDP tunnels

- a new function generating those ICMP or ICMPv6 replies. We can't reuse icmp_send() and icmp6_send() as we don't see the sender as a valid destination. This doesn't need to be generic, as we don't cover any other type of ICMP errors given that we only provide an encapsulation function to the sender

While at it, make the MTU check in skb_tunnel_check_pmtu() accurate:
we might receive GSO buffers here, and the passed headroom already includes the inner MAC length, so we don't have to account for it a second time (that would imply three MAC headers on the wire, but there are just two).

This issue became visible while bridging IPv6 packets with 4500 bytes of payload over GENEVE using IPv4 with a PMTU of 4000. Given the 50 bytes of encapsulation headroom, we would advertise MTU as 3950, and we would reject fragmented IPv6 datagrams of 3958 bytes size on the wire. We're exclusively dealing with network MTU here, though, so we could get Ethernet frames up to 3964 octets in that case.

v2:
- moved skb_tunnel_check_pmtu() to ip_tunnel_core.c (David Ahern)
- split IPv4/IPv6 functions (David Ahern)

Signed-off-by Stefano Brivio <[email protected]>

---
 drivers/net/bareudp.c     |   5 +-
 drivers/net/geneve.c      |   5 +-
 drivers/net/vxlan.c       |   4 +-
 include/net/dst.h         |  10 --
 include/net/ip_tunnels.h  |   2 +
 net/ipv4/ip_tunnel_core.c | 244 ++++++++++++++++++++++++++++++++++++++
 6 files changed, 254 insertions(+), 16 deletions(-)

drivers/net/bareudp.c +3 -2

index 3b6664c7e73c..841910f1db65 100644
@@ -308,7 +308,7 @@ static int bareudp_xmit_skb(struct sk_buff *skb, struct net_device *dev,
                return PTR_ERR(rt);
 
        skb_tunnel_check_pmtu(skb, &rt->dst,
-                             BAREUDP_IPV4_HLEN + info->options_len);
+                             BAREUDP_IPV4_HLEN + info->options_len, false);
 
        sport = udp_flow_src_port(bareudp->net, skb,
                                  bareudp->sport_min, USHRT_MAX,

@@ -369,7 +369,8 @@ static int bareudp6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
        if (IS_ERR(dst))
                return PTR_ERR(dst);
 
-       skb_tunnel_check_pmtu(skb, dst, BAREUDP_IPV6_HLEN + info->options_len);
+       skb_tunnel_check_pmtu(skb, dst, BAREUDP_IPV6_HLEN + info->options_len,
+                             false);
 
        sport = udp_flow_src_port(bareudp->net, skb,
                                  bareudp->sport_min, USHRT_MAX,

drivers/net/geneve.c +3 -2

index 017c13acc911..de86b6d82132 100644
@@ -894,7 +894,7 @@ static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
                return PTR_ERR(rt);
 
        skb_tunnel_check_pmtu(skb, &rt->dst,
-                             GENEVE_IPV4_HLEN + info->options_len);
+                             GENEVE_IPV4_HLEN + info->options_len, false);
 
        sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
        if (geneve->cfg.collect_md) {

@@ -955,7 +955,8 @@ static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
        if (IS_ERR(dst))
                return PTR_ERR(dst);
 
-       skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len);
+       skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len,
+                             false);
 
        sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
        if (geneve->cfg.collect_md) {

drivers/net/vxlan.c +2 -2

index 77658425db8a..1432544da507 100644
@@ -2720,7 +2720,7 @@ static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
                }
 
                ndst = &rt->dst;
-               skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM);
+               skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM, false);
 
                tos = ip_tunnel_ecn_encap(RT_TOS(tos), old_iph, skb);
                ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);

@@ -2760,7 +2760,7 @@ static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
                                goto out_unlock;
                }
 
-               skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM);
+               skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM, false);
 
                tos = ip_tunnel_ecn_encap(RT_TOS(tos), old_iph, skb);
                ttl = ttl ? : ip6_dst_hoplimit(ndst);

include/net/dst.h +0 -10

index 852d8fb36ab7..6ae2e625050d 100644
@@ -535,14 +535,4 @@ static inline void skb_dst_update_pmtu_no_confirm(struct sk_buff *skb, u32 mtu)
                dst->ops->update_pmtu(dst, NULL, skb, mtu, false);
 }
 
-static inline void skb_tunnel_check_pmtu(struct sk_buff *skb,
-                                        struct dst_entry *encap_dst,
-                                        int headroom)
-{
-       u32 encap_mtu = dst_mtu(encap_dst);
-
-       if (skb->len > encap_mtu - headroom)
-               skb_dst_update_pmtu_no_confirm(skb, encap_mtu - headroom);
-}
-
 #endif /* _NET_DST_H */

include/net/ip_tunnels.h +2 -0

index 36025dea7612..02ccd32542d0 100644
@@ -420,6 +420,8 @@ void iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb,
                   u8 tos, u8 ttl, __be16 df, bool xnet);
 struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md,
                                             gfp_t flags);
+int skb_tunnel_check_pmtu(struct sk_buff *skb, struct dst_entry *encap_dst,
+                         int headroom, bool reply);
 
 int iptunnel_handle_offloads(struct sk_buff *skb, int gso_type_mask);
 

net/ipv4/ip_tunnel_core.c +140 -0

index f8b419e2475c..9ddee2a0c66d 100644
@@ -184,6 +184,250 @@ int iptunnel_handle_offloads(struct sk_buff *skb,
 }
 EXPORT_SYMBOL_GPL(iptunnel_handle_offloads);
 
+/**
+ * iptunnel_pmtud_build_icmp() - Build ICMP error message for PMTUD
+ * @skb:       Original packet with L2 header
+ * @mtu:       MTU value for ICMP error
+ *
+ * Return: length on success, negative error code if message couldn't be built.
+ */
+static int iptunnel_pmtud_build_icmp(struct sk_buff *skb, int mtu)
+{
+       const struct iphdr *iph = ip_hdr(skb);
+       struct icmphdr *icmph;
+       struct iphdr *niph;
+       struct ethhdr eh;
+       int len, err;
+
+       if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr)))
+               return -EINVAL;
+
+       skb_copy_bits(skb, skb_mac_offset(skb), &eh, ETH_HLEN);
+       pskb_pull(skb, ETH_HLEN);
+       skb_reset_network_header(skb);
+
+       err = pskb_trim(skb, 576 - sizeof(*niph) - sizeof(*icmph));
+       if (err)
+               return err;
+
+       len = skb->len + sizeof(*icmph);
+       err = skb_cow(skb, sizeof(*niph) + sizeof(*icmph) + ETH_HLEN);
+       if (err)
+               return err;
+
+       icmph = skb_push(skb, sizeof(*icmph));
+       *icmph = (struct icmphdr) {
+               .type                   = ICMP_DEST_UNREACH,
+               .code                   = ICMP_FRAG_NEEDED,
+               .checksum               = 0,
+               .un.frag.__unused       = 0,
+               .un.frag.mtu            = ntohs(mtu),
+       };
+       icmph->checksum = ip_compute_csum(icmph, len);
+       skb_reset_transport_header(skb);
+
+       niph = skb_push(skb, sizeof(*niph));
+       *niph = (struct iphdr) {
+               .ihl                    = sizeof(*niph) / 4u,
+               .version                = 4,
+               .tos                    = 0,
+               .tot_len                = htons(len + sizeof(*niph)),
+               .id                     = 0,
+               .frag_off               = htons(IP_DF),
+               .ttl                    = iph->ttl,
+               .protocol               = IPPROTO_ICMP,
+               .saddr                  = iph->daddr,
+               .daddr                  = iph->saddr,
+       };
+       ip_send_check(niph);
+       skb_reset_network_header(skb);
+
+       skb->ip_summed = CHECKSUM_NONE;
+
+       eth_header(skb, skb->dev, htons(eh.h_proto), eh.h_source, eh.h_dest, 0);
+       skb_reset_mac_header(skb);
+
+       return skb->len;
+}
+
+/**
+ * iptunnel_pmtud_check_icmp() - Trigger ICMP reply if needed and allowed
+ * @skb:       Buffer being sent by encapsulation, L2 headers expected
+ * @mtu:       Network MTU for path
+ *
+ * Return: 0 for no ICMP reply, length if built, negative value on error.
+ */
+static int iptunnel_pmtud_check_icmp(struct sk_buff *skb, int mtu)
+{
+       const struct icmphdr *icmph = icmp_hdr(skb);
+       const struct iphdr *iph = ip_hdr(skb);
+
+       if (mtu <= 576 || iph->frag_off != htons(IP_DF))
+               return 0;
+
+       if (ipv4_is_lbcast(iph->daddr)  || ipv4_is_multicast(iph->daddr) ||
+           ipv4_is_zeronet(iph->saddr) || ipv4_is_loopback(iph->saddr)  ||
+           ipv4_is_lbcast(iph->saddr)  || ipv4_is_multicast(iph->saddr))
+               return 0;
+
+       if (iph->protocol == IPPROTO_ICMP && icmp_is_err(icmph->type))
+               return 0;
+
+       return iptunnel_pmtud_build_icmp(skb, mtu);
+}
+
+#if IS_ENABLED(CONFIG_IPV6)
+/**
+ * iptunnel_pmtud_build_icmpv6() - Build ICMPv6 error message for PMTUD
+ * @skb:       Original packet with L2 header
+ * @mtu:       MTU value for ICMPv6 error
+ *
+ * Return: length on success, negative error code if message couldn't be built.
+ */
+static int iptunnel_pmtud_build_icmpv6(struct sk_buff *skb, int mtu)
+{
+       const struct ipv6hdr *ip6h = ipv6_hdr(skb);
+       struct icmp6hdr *icmp6h;
+       struct ipv6hdr *nip6h;
+       struct ethhdr eh;
+       int len, err;
+       __wsum csum;
+
+       if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr)))
+               return -EINVAL;
+
+       skb_copy_bits(skb, skb_mac_offset(skb), &eh, ETH_HLEN);
+       pskb_pull(skb, ETH_HLEN);
+       skb_reset_network_header(skb);
+
+       err = pskb_trim(skb, IPV6_MIN_MTU - sizeof(*nip6h) - sizeof(*icmp6h));
+       if (err)
+               return err;
+
+       len = skb->len + sizeof(*icmp6h);
+       err = skb_cow(skb, sizeof(*nip6h) + sizeof(*icmp6h) + ETH_HLEN);
+       if (err)
+               return err;
+
+       icmp6h = skb_push(skb, sizeof(*icmp6h));
+       *icmp6h = (struct icmp6hdr) {
+               .icmp6_type             = ICMPV6_PKT_TOOBIG,
+               .icmp6_code             = 0,
+               .icmp6_cksum            = 0,
+               .icmp6_mtu              = htonl(mtu),
+       };
+       skb_reset_transport_header(skb);
+
+       nip6h = skb_push(skb, sizeof(*nip6h));
+       *nip6h = (struct ipv6hdr) {
+               .priority               = 0,
+               .version                = 6,
+               .flow_lbl               = { 0 },
+               .payload_len            = htons(len),

net/ipv4/ip_tunnel_core.c int iptunnel_handle_offloads(struct sk_buff *skb, · line 336

Quoted 3 lines · Stefano Brivio

net/ipv4/ip_tunnel_core.c +2 -0

+       icmp6h->icmp6_cksum = csum_ipv6_magic(&nip6h->saddr, &nip6h->daddr, len,
+                                             IPPROTO_ICMPV6, csum);

Linux next build breaks for riscv architecture defconfig build.