2 * VXLAN: Virtual eXtensible Local Area Network
4 * Copyright (c) 2012 Vyatta Inc.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
11 * - use IANA UDP port number (when defined)
15 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17 #include <linux/kernel.h>
18 #include <linux/types.h>
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/skbuff.h>
23 #include <linux/rculist.h>
24 #include <linux/netdevice.h>
27 #include <linux/udp.h>
28 #include <linux/igmp.h>
29 #include <linux/etherdevice.h>
30 #include <linux/if_ether.h>
31 #include <linux/hash.h>
32 #include <linux/ethtool.h>
34 #include <net/ndisc.h>
38 #include <net/rtnetlink.h>
39 #include <net/route.h>
40 #include <net/dsfield.h>
41 #include <net/inet_ecn.h>
42 #include <net/net_namespace.h>
43 #include <net/netns/generic.h>
45 #define VXLAN_VERSION "0.1"
47 #define VNI_HASH_BITS 10
48 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
49 #define FDB_HASH_BITS 8
50 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
51 #define FDB_AGE_DEFAULT 300 /* 5 min */
52 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
54 #define VXLAN_N_VID (1u << 24)
55 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
56 /* IP header + UDP + VXLAN + Ethernet header */
57 #define VXLAN_HEADROOM (20 + 8 + 8 + 14)
59 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
61 /* VXLAN protocol header */
67 /* UDP port for VXLAN traffic. */
68 static unsigned int vxlan_port __read_mostly = 8472;
69 module_param_named(udp_port, vxlan_port, uint, 0444);
70 MODULE_PARM_DESC(udp_port, "Destination UDP port");
72 static bool log_ecn_error = true;
73 module_param(log_ecn_error, bool, 0644);
74 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
76 /* per-net private data for this module */
77 static unsigned int vxlan_net_id;
79 struct socket *sock; /* UDP encap socket */
80 struct hlist_head vni_list[VNI_HASH_SIZE];
83 /* Forwarding table entry */
85 struct hlist_node hlist; /* linked list of entries */
87 unsigned long updated; /* jiffies */
90 u16 state; /* see ndm_state */
91 u8 eth_addr[ETH_ALEN];
94 /* Per-cpu network traffic stats */
100 struct u64_stats_sync syncp;
103 /* Pseudo network device */
105 struct hlist_node hlist;
106 struct net_device *dev;
107 struct vxlan_stats __percpu *stats;
108 __u32 vni; /* virtual network id */
109 __be32 gaddr; /* multicast group */
110 __be32 saddr; /* source address */
111 unsigned int link; /* link to multicast over */
112 __u16 port_min; /* source port range */
114 __u8 tos; /* TOS override */
116 u32 flags; /* VXLAN_F_* below */
118 unsigned long age_interval;
119 struct timer_list age_timer;
120 spinlock_t hash_lock;
121 unsigned int addrcnt;
122 unsigned int addrmax;
124 struct hlist_head fdb_head[FDB_HASH_SIZE];
127 #define VXLAN_F_LEARN 0x01
128 #define VXLAN_F_PROXY 0x02
129 #define VXLAN_F_RSC 0x04
130 #define VXLAN_F_L2MISS 0x08
131 #define VXLAN_F_L3MISS 0x10
133 /* salt for hash table */
134 static u32 vxlan_salt __read_mostly;
136 static inline struct hlist_head *vni_head(struct net *net, u32 id)
138 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
140 return &vn->vni_list[hash_32(id, VNI_HASH_BITS)];
143 /* Look up VNI in a per net namespace table */
144 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
146 struct vxlan_dev *vxlan;
147 struct hlist_node *node;
149 hlist_for_each_entry_rcu(vxlan, node, vni_head(net, id), hlist) {
150 if (vxlan->vni == id)
157 /* Fill in neighbour message in skbuff. */
158 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
159 const struct vxlan_fdb *fdb,
160 u32 portid, u32 seq, int type, unsigned int flags)
162 unsigned long now = jiffies;
163 struct nda_cacheinfo ci;
164 struct nlmsghdr *nlh;
166 bool send_ip, send_eth;
168 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
172 ndm = nlmsg_data(nlh);
173 memset(ndm, 0, sizeof(*ndm));
175 send_eth = send_ip = true;
177 if (type == RTM_GETNEIGH) {
178 ndm->ndm_family = AF_INET;
179 send_ip = fdb->remote_ip != 0;
180 send_eth = !is_zero_ether_addr(fdb->eth_addr);
182 ndm->ndm_family = AF_BRIDGE;
183 ndm->ndm_state = fdb->state;
184 ndm->ndm_ifindex = vxlan->dev->ifindex;
185 ndm->ndm_flags = NTF_SELF;
186 ndm->ndm_type = NDA_DST;
188 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
189 goto nla_put_failure;
191 if (send_ip && nla_put_be32(skb, NDA_DST, fdb->remote_ip))
192 goto nla_put_failure;
194 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
195 ci.ndm_confirmed = 0;
196 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
199 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
200 goto nla_put_failure;
202 return nlmsg_end(skb, nlh);
205 nlmsg_cancel(skb, nlh);
209 static inline size_t vxlan_nlmsg_size(void)
211 return NLMSG_ALIGN(sizeof(struct ndmsg))
212 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
213 + nla_total_size(sizeof(__be32)) /* NDA_DST */
214 + nla_total_size(sizeof(struct nda_cacheinfo));
217 static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
218 const struct vxlan_fdb *fdb, int type)
220 struct net *net = dev_net(vxlan->dev);
224 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
228 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0);
230 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
231 WARN_ON(err == -EMSGSIZE);
236 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
240 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
243 static void vxlan_ip_miss(struct net_device *dev, __be32 ipa)
245 struct vxlan_dev *vxlan = netdev_priv(dev);
248 memset(&f, 0, sizeof f);
250 f.remote_ip = ipa; /* goes to NDA_DST */
252 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
255 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
259 memset(&f, 0, sizeof f);
261 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
263 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
266 /* Hash Ethernet address */
267 static u32 eth_hash(const unsigned char *addr)
269 u64 value = get_unaligned((u64 *)addr);
271 /* only want 6 bytes */
277 return hash_64(value, FDB_HASH_BITS);
280 /* Hash chain to use given mac address */
281 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
284 return &vxlan->fdb_head[eth_hash(mac)];
287 /* Look up Ethernet address in forwarding table */
288 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
292 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
294 struct hlist_node *node;
296 hlist_for_each_entry_rcu(f, node, head, hlist) {
297 if (compare_ether_addr(mac, f->eth_addr) == 0)
304 /* Add new entry to forwarding table -- assumes lock held */
305 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
306 const u8 *mac, __be32 ip,
307 __u16 state, __u16 flags)
312 f = vxlan_find_mac(vxlan, mac);
314 if (flags & NLM_F_EXCL) {
315 netdev_dbg(vxlan->dev,
316 "lost race to create %pM\n", mac);
319 if (f->state != state) {
321 f->updated = jiffies;
325 if (!(flags & NLM_F_CREATE))
328 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
331 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
332 f = kmalloc(sizeof(*f), GFP_ATOMIC);
339 f->updated = f->used = jiffies;
340 memcpy(f->eth_addr, mac, ETH_ALEN);
343 hlist_add_head_rcu(&f->hlist,
344 vxlan_fdb_head(vxlan, mac));
348 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
353 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
355 netdev_dbg(vxlan->dev,
356 "delete %pM\n", f->eth_addr);
359 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
361 hlist_del_rcu(&f->hlist);
365 /* Add static entry (via netlink) */
366 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
367 struct net_device *dev,
368 const unsigned char *addr, u16 flags)
370 struct vxlan_dev *vxlan = netdev_priv(dev);
374 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
375 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
380 if (tb[NDA_DST] == NULL)
383 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
384 return -EAFNOSUPPORT;
386 ip = nla_get_be32(tb[NDA_DST]);
388 spin_lock_bh(&vxlan->hash_lock);
389 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags);
390 spin_unlock_bh(&vxlan->hash_lock);
395 /* Delete entry (via netlink) */
396 static int vxlan_fdb_delete(struct ndmsg *ndm, struct net_device *dev,
397 const unsigned char *addr)
399 struct vxlan_dev *vxlan = netdev_priv(dev);
403 spin_lock_bh(&vxlan->hash_lock);
404 f = vxlan_find_mac(vxlan, addr);
406 vxlan_fdb_destroy(vxlan, f);
409 spin_unlock_bh(&vxlan->hash_lock);
414 /* Dump forwarding table */
415 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
416 struct net_device *dev, int idx)
418 struct vxlan_dev *vxlan = netdev_priv(dev);
421 for (h = 0; h < FDB_HASH_SIZE; ++h) {
423 struct hlist_node *n;
426 hlist_for_each_entry_rcu(f, n, &vxlan->fdb_head[h], hlist) {
427 if (idx < cb->args[0])
430 err = vxlan_fdb_info(skb, vxlan, f,
431 NETLINK_CB(cb->skb).portid,
445 /* Watch incoming packets to learn mapping between Ethernet address
446 * and Tunnel endpoint.
448 static void vxlan_snoop(struct net_device *dev,
449 __be32 src_ip, const u8 *src_mac)
451 struct vxlan_dev *vxlan = netdev_priv(dev);
455 f = vxlan_find_mac(vxlan, src_mac);
458 if (likely(f->remote_ip == src_ip))
463 "%pM migrated from %pI4 to %pI4\n",
464 src_mac, &f->remote_ip, &src_ip);
466 f->remote_ip = src_ip;
467 f->updated = jiffies;
469 /* learned new entry */
470 spin_lock(&vxlan->hash_lock);
471 err = vxlan_fdb_create(vxlan, src_mac, src_ip,
473 NLM_F_EXCL|NLM_F_CREATE);
474 spin_unlock(&vxlan->hash_lock);
479 /* See if multicast group is already in use by other ID */
480 static bool vxlan_group_used(struct vxlan_net *vn,
481 const struct vxlan_dev *this)
483 const struct vxlan_dev *vxlan;
484 struct hlist_node *node;
487 for (h = 0; h < VNI_HASH_SIZE; ++h)
488 hlist_for_each_entry(vxlan, node, &vn->vni_list[h], hlist) {
492 if (!netif_running(vxlan->dev))
495 if (vxlan->gaddr == this->gaddr)
502 /* kernel equivalent to IP_ADD_MEMBERSHIP */
503 static int vxlan_join_group(struct net_device *dev)
505 struct vxlan_dev *vxlan = netdev_priv(dev);
506 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
507 struct sock *sk = vn->sock->sk;
508 struct ip_mreqn mreq = {
509 .imr_multiaddr.s_addr = vxlan->gaddr,
510 .imr_ifindex = vxlan->link,
514 /* Already a member of group */
515 if (vxlan_group_used(vn, vxlan))
518 /* Need to drop RTNL to call multicast join */
521 err = ip_mc_join_group(sk, &mreq);
529 /* kernel equivalent to IP_DROP_MEMBERSHIP */
530 static int vxlan_leave_group(struct net_device *dev)
532 struct vxlan_dev *vxlan = netdev_priv(dev);
533 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
535 struct sock *sk = vn->sock->sk;
536 struct ip_mreqn mreq = {
537 .imr_multiaddr.s_addr = vxlan->gaddr,
538 .imr_ifindex = vxlan->link,
541 /* Only leave group when last vxlan is done. */
542 if (vxlan_group_used(vn, vxlan))
545 /* Need to drop RTNL to call multicast leave */
548 err = ip_mc_leave_group(sk, &mreq);
555 /* Callback from net/ipv4/udp.c to receive packets */
556 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
559 struct vxlanhdr *vxh;
560 struct vxlan_dev *vxlan;
561 struct vxlan_stats *stats;
565 /* pop off outer UDP header */
566 __skb_pull(skb, sizeof(struct udphdr));
568 /* Need Vxlan and inner Ethernet header to be present */
569 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
572 /* Drop packets with reserved bits set */
573 vxh = (struct vxlanhdr *) skb->data;
574 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
575 (vxh->vx_vni & htonl(0xff))) {
576 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
577 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
581 __skb_pull(skb, sizeof(struct vxlanhdr));
583 /* Is this VNI defined? */
584 vni = ntohl(vxh->vx_vni) >> 8;
585 vxlan = vxlan_find_vni(sock_net(sk), vni);
587 netdev_dbg(skb->dev, "unknown vni %d\n", vni);
591 if (!pskb_may_pull(skb, ETH_HLEN)) {
592 vxlan->dev->stats.rx_length_errors++;
593 vxlan->dev->stats.rx_errors++;
597 skb_reset_mac_header(skb);
599 /* Re-examine inner Ethernet packet */
601 skb->protocol = eth_type_trans(skb, vxlan->dev);
603 /* Ignore packet loops (and multicast echo) */
604 if (compare_ether_addr(eth_hdr(skb)->h_source,
605 vxlan->dev->dev_addr) == 0)
608 if (vxlan->flags & VXLAN_F_LEARN)
609 vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source);
611 __skb_tunnel_rx(skb, vxlan->dev);
612 skb_reset_network_header(skb);
614 /* If the NIC driver gave us an encapsulated packet with
615 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
616 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
617 * for us. Otherwise force the upper layers to verify it.
619 if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
620 !(vxlan->dev->features & NETIF_F_RXCSUM))
621 skb->ip_summed = CHECKSUM_NONE;
623 skb->encapsulation = 0;
625 err = IP_ECN_decapsulate(oip, skb);
628 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
629 &oip->saddr, oip->tos);
631 ++vxlan->dev->stats.rx_frame_errors;
632 ++vxlan->dev->stats.rx_errors;
637 stats = this_cpu_ptr(vxlan->stats);
638 u64_stats_update_begin(&stats->syncp);
640 stats->rx_bytes += skb->len;
641 u64_stats_update_end(&stats->syncp);
647 /* Put UDP header back */
648 __skb_push(skb, sizeof(struct udphdr));
652 /* Consume bad packet */
657 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
659 struct vxlan_dev *vxlan = netdev_priv(dev);
665 if (dev->flags & IFF_NOARP)
668 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
669 dev->stats.tx_dropped++;
674 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
675 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
676 parp->ar_pro != htons(ETH_P_IP) ||
677 parp->ar_op != htons(ARPOP_REQUEST) ||
678 parp->ar_hln != dev->addr_len ||
681 arpptr = (u8 *)parp + sizeof(struct arphdr);
683 arpptr += dev->addr_len; /* sha */
684 memcpy(&sip, arpptr, sizeof(sip));
685 arpptr += sizeof(sip);
686 arpptr += dev->addr_len; /* tha */
687 memcpy(&tip, arpptr, sizeof(tip));
689 if (ipv4_is_loopback(tip) ||
690 ipv4_is_multicast(tip))
693 n = neigh_lookup(&arp_tbl, &tip, dev);
696 struct vxlan_dev *vxlan = netdev_priv(dev);
698 struct sk_buff *reply;
700 if (!(n->nud_state & NUD_CONNECTED)) {
705 f = vxlan_find_mac(vxlan, n->ha);
706 if (f && f->remote_ip == 0) {
707 /* bridge-local neighbor */
712 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
717 skb_reset_mac_header(reply);
718 __skb_pull(reply, skb_network_offset(reply));
719 reply->ip_summed = CHECKSUM_UNNECESSARY;
720 reply->pkt_type = PACKET_HOST;
722 if (netif_rx_ni(reply) == NET_RX_DROP)
723 dev->stats.rx_dropped++;
724 } else if (vxlan->flags & VXLAN_F_L3MISS)
725 vxlan_ip_miss(dev, tip);
731 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
733 struct vxlan_dev *vxlan = netdev_priv(dev);
737 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
741 switch (ntohs(eth_hdr(skb)->h_proto)) {
743 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
746 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
755 diff = compare_ether_addr(eth_hdr(skb)->h_dest, n->ha) != 0;
757 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
759 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
763 } else if (vxlan->flags & VXLAN_F_L3MISS)
764 vxlan_ip_miss(dev, pip->daddr);
768 /* Extract dsfield from inner protocol */
769 static inline u8 vxlan_get_dsfield(const struct iphdr *iph,
770 const struct sk_buff *skb)
772 if (skb->protocol == htons(ETH_P_IP))
774 else if (skb->protocol == htons(ETH_P_IPV6))
775 return ipv6_get_dsfield((const struct ipv6hdr *)iph);
780 /* Propogate ECN bits out */
781 static inline u8 vxlan_ecn_encap(u8 tos,
782 const struct iphdr *iph,
783 const struct sk_buff *skb)
785 u8 inner = vxlan_get_dsfield(iph, skb);
787 return INET_ECN_encapsulate(tos, inner);
790 static void vxlan_sock_free(struct sk_buff *skb)
795 /* On transmit, associate with the tunnel socket */
796 static void vxlan_set_owner(struct net_device *dev, struct sk_buff *skb)
798 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
799 struct sock *sk = vn->sock->sk;
804 skb->destructor = vxlan_sock_free;
807 /* Compute source port for outgoing packet
808 * first choice to use L4 flow hash since it will spread
809 * better and maybe available from hardware
810 * secondary choice is to use jhash on the Ethernet header
812 static u16 vxlan_src_port(const struct vxlan_dev *vxlan, struct sk_buff *skb)
814 unsigned int range = (vxlan->port_max - vxlan->port_min) + 1;
817 hash = skb_get_rxhash(skb);
819 hash = jhash(skb->data, 2 * ETH_ALEN,
820 (__force u32) skb->protocol);
822 return (((u64) hash * range) >> 32) + vxlan->port_min;
825 /* Transmit local packets over Vxlan
827 * Outer IP header inherits ECN and DF from inner header.
828 * Outer UDP destination is the VXLAN assigned port.
829 * source port is based on hash of flow
831 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
833 struct vxlan_dev *vxlan = netdev_priv(dev);
835 const struct iphdr *old_iph;
838 struct vxlanhdr *vxh;
841 unsigned int pkt_len = skb->len;
847 bool did_rsc = false;
848 const struct vxlan_fdb *f;
850 skb_reset_mac_header(skb);
853 if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP)
854 return arp_reduce(dev, skb);
855 else if ((vxlan->flags&VXLAN_F_RSC) && ntohs(eth->h_proto) == ETH_P_IP)
856 did_rsc = route_shortcircuit(dev, skb);
858 f = vxlan_find_mac(vxlan, eth->h_dest);
862 if (!dst && (vxlan->flags & VXLAN_F_L2MISS) &&
863 !is_multicast_ether_addr(eth->h_dest))
864 vxlan_fdb_miss(vxlan, eth->h_dest);
870 __skb_pull(skb, skb_network_offset(skb));
871 skb->ip_summed = CHECKSUM_NONE;
872 skb->pkt_type = PACKET_HOST;
874 /* short-circuited back to local bridge */
875 if (netif_rx(skb) == NET_RX_SUCCESS) {
876 struct vxlan_stats *stats =
877 this_cpu_ptr(vxlan->stats);
879 u64_stats_update_begin(&stats->syncp);
881 stats->tx_bytes += pkt_len;
882 u64_stats_update_end(&stats->syncp);
884 dev->stats.tx_errors++;
885 dev->stats.tx_aborted_errors++;
892 if (!skb->encapsulation) {
893 skb_reset_inner_headers(skb);
894 skb->encapsulation = 1;
897 /* Need space for new headers (invalidates iph ptr) */
898 if (skb_cow_head(skb, VXLAN_HEADROOM))
901 old_iph = ip_hdr(skb);
904 if (!ttl && IN_MULTICAST(ntohl(dst)))
909 tos = vxlan_get_dsfield(old_iph, skb);
911 src_port = vxlan_src_port(vxlan, skb);
913 memset(&fl4, 0, sizeof(fl4));
914 fl4.flowi4_oif = vxlan->link;
915 fl4.flowi4_tos = RT_TOS(tos);
917 fl4.saddr = vxlan->saddr;
919 rt = ip_route_output_key(dev_net(dev), &fl4);
921 netdev_dbg(dev, "no route to %pI4\n", &dst);
922 dev->stats.tx_carrier_errors++;
926 if (rt->dst.dev == dev) {
927 netdev_dbg(dev, "circular route to %pI4\n", &dst);
929 dev->stats.collisions++;
933 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
934 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
937 skb_dst_set(skb, &rt->dst);
939 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
940 vxh->vx_flags = htonl(VXLAN_FLAGS);
941 vxh->vx_vni = htonl(vxlan->vni << 8);
943 __skb_push(skb, sizeof(*uh));
944 skb_reset_transport_header(skb);
947 uh->dest = htons(vxlan_port);
948 uh->source = htons(src_port);
950 uh->len = htons(skb->len);
953 __skb_push(skb, sizeof(*iph));
954 skb_reset_network_header(skb);
957 iph->ihl = sizeof(struct iphdr) >> 2;
959 iph->protocol = IPPROTO_UDP;
960 iph->tos = vxlan_ecn_encap(tos, old_iph, skb);
962 iph->saddr = fl4.saddr;
963 iph->ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
965 vxlan_set_owner(dev, skb);
967 /* See iptunnel_xmit() */
968 if (skb->ip_summed != CHECKSUM_PARTIAL)
969 skb->ip_summed = CHECKSUM_NONE;
970 ip_select_ident(iph, &rt->dst, NULL);
972 err = ip_local_out(skb);
973 if (likely(net_xmit_eval(err) == 0)) {
974 struct vxlan_stats *stats = this_cpu_ptr(vxlan->stats);
976 u64_stats_update_begin(&stats->syncp);
978 stats->tx_bytes += pkt_len;
979 u64_stats_update_end(&stats->syncp);
981 dev->stats.tx_errors++;
982 dev->stats.tx_aborted_errors++;
987 dev->stats.tx_dropped++;
991 dev->stats.tx_errors++;
997 /* Walk the forwarding table and purge stale entries */
998 static void vxlan_cleanup(unsigned long arg)
1000 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1001 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1004 if (!netif_running(vxlan->dev))
1007 spin_lock_bh(&vxlan->hash_lock);
1008 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1009 struct hlist_node *p, *n;
1010 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1012 = container_of(p, struct vxlan_fdb, hlist);
1013 unsigned long timeout;
1015 if (f->state & NUD_PERMANENT)
1018 timeout = f->used + vxlan->age_interval * HZ;
1019 if (time_before_eq(timeout, jiffies)) {
1020 netdev_dbg(vxlan->dev,
1021 "garbage collect %pM\n",
1023 f->state = NUD_STALE;
1024 vxlan_fdb_destroy(vxlan, f);
1025 } else if (time_before(timeout, next_timer))
1026 next_timer = timeout;
1029 spin_unlock_bh(&vxlan->hash_lock);
1031 mod_timer(&vxlan->age_timer, next_timer);
1034 /* Setup stats when device is created */
1035 static int vxlan_init(struct net_device *dev)
1037 struct vxlan_dev *vxlan = netdev_priv(dev);
1039 vxlan->stats = alloc_percpu(struct vxlan_stats);
1046 /* Start ageing timer and join group when device is brought up */
1047 static int vxlan_open(struct net_device *dev)
1049 struct vxlan_dev *vxlan = netdev_priv(dev);
1053 err = vxlan_join_group(dev);
1058 if (vxlan->age_interval)
1059 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
1064 /* Purge the forwarding table */
1065 static void vxlan_flush(struct vxlan_dev *vxlan)
1069 spin_lock_bh(&vxlan->hash_lock);
1070 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1071 struct hlist_node *p, *n;
1072 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1074 = container_of(p, struct vxlan_fdb, hlist);
1075 vxlan_fdb_destroy(vxlan, f);
1078 spin_unlock_bh(&vxlan->hash_lock);
1081 /* Cleanup timer and forwarding table on shutdown */
1082 static int vxlan_stop(struct net_device *dev)
1084 struct vxlan_dev *vxlan = netdev_priv(dev);
1087 vxlan_leave_group(dev);
1089 del_timer_sync(&vxlan->age_timer);
1096 /* Merge per-cpu statistics */
1097 static struct rtnl_link_stats64 *vxlan_stats64(struct net_device *dev,
1098 struct rtnl_link_stats64 *stats)
1100 struct vxlan_dev *vxlan = netdev_priv(dev);
1101 struct vxlan_stats tmp, sum = { 0 };
1104 for_each_possible_cpu(cpu) {
1106 const struct vxlan_stats *stats
1107 = per_cpu_ptr(vxlan->stats, cpu);
1110 start = u64_stats_fetch_begin_bh(&stats->syncp);
1111 memcpy(&tmp, stats, sizeof(tmp));
1112 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1114 sum.tx_bytes += tmp.tx_bytes;
1115 sum.tx_packets += tmp.tx_packets;
1116 sum.rx_bytes += tmp.rx_bytes;
1117 sum.rx_packets += tmp.rx_packets;
1120 stats->tx_bytes = sum.tx_bytes;
1121 stats->tx_packets = sum.tx_packets;
1122 stats->rx_bytes = sum.rx_bytes;
1123 stats->rx_packets = sum.rx_packets;
1125 stats->multicast = dev->stats.multicast;
1126 stats->rx_length_errors = dev->stats.rx_length_errors;
1127 stats->rx_frame_errors = dev->stats.rx_frame_errors;
1128 stats->rx_errors = dev->stats.rx_errors;
1130 stats->tx_dropped = dev->stats.tx_dropped;
1131 stats->tx_carrier_errors = dev->stats.tx_carrier_errors;
1132 stats->tx_aborted_errors = dev->stats.tx_aborted_errors;
1133 stats->collisions = dev->stats.collisions;
1134 stats->tx_errors = dev->stats.tx_errors;
1139 /* Stub, nothing needs to be done. */
1140 static void vxlan_set_multicast_list(struct net_device *dev)
1144 static const struct net_device_ops vxlan_netdev_ops = {
1145 .ndo_init = vxlan_init,
1146 .ndo_open = vxlan_open,
1147 .ndo_stop = vxlan_stop,
1148 .ndo_start_xmit = vxlan_xmit,
1149 .ndo_get_stats64 = vxlan_stats64,
1150 .ndo_set_rx_mode = vxlan_set_multicast_list,
1151 .ndo_change_mtu = eth_change_mtu,
1152 .ndo_validate_addr = eth_validate_addr,
1153 .ndo_set_mac_address = eth_mac_addr,
1154 .ndo_fdb_add = vxlan_fdb_add,
1155 .ndo_fdb_del = vxlan_fdb_delete,
1156 .ndo_fdb_dump = vxlan_fdb_dump,
1159 /* Info for udev, that this is a virtual tunnel endpoint */
1160 static struct device_type vxlan_type = {
1164 static void vxlan_free(struct net_device *dev)
1166 struct vxlan_dev *vxlan = netdev_priv(dev);
1168 free_percpu(vxlan->stats);
1172 /* Initialize the device structure. */
1173 static void vxlan_setup(struct net_device *dev)
1175 struct vxlan_dev *vxlan = netdev_priv(dev);
1179 eth_hw_addr_random(dev);
1181 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
1183 dev->netdev_ops = &vxlan_netdev_ops;
1184 dev->destructor = vxlan_free;
1185 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
1187 dev->tx_queue_len = 0;
1188 dev->features |= NETIF_F_LLTX;
1189 dev->features |= NETIF_F_NETNS_LOCAL;
1190 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1191 dev->features |= NETIF_F_RXCSUM;
1193 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1194 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1195 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1197 spin_lock_init(&vxlan->hash_lock);
1199 init_timer_deferrable(&vxlan->age_timer);
1200 vxlan->age_timer.function = vxlan_cleanup;
1201 vxlan->age_timer.data = (unsigned long) vxlan;
1203 inet_get_local_port_range(&low, &high);
1204 vxlan->port_min = low;
1205 vxlan->port_max = high;
1209 for (h = 0; h < FDB_HASH_SIZE; ++h)
1210 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
1213 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
1214 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
1215 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1216 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
1217 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1218 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
1219 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
1220 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
1221 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
1222 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
1223 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
1224 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
1225 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
1226 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
1227 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
1230 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
1232 if (tb[IFLA_ADDRESS]) {
1233 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1234 pr_debug("invalid link address (not ethernet)\n");
1238 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1239 pr_debug("invalid all zero ethernet address\n");
1240 return -EADDRNOTAVAIL;
1247 if (data[IFLA_VXLAN_ID]) {
1248 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
1249 if (id >= VXLAN_VID_MASK)
1253 if (data[IFLA_VXLAN_GROUP]) {
1254 __be32 gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1255 if (!IN_MULTICAST(ntohl(gaddr))) {
1256 pr_debug("group address is not IPv4 multicast\n");
1257 return -EADDRNOTAVAIL;
1261 if (data[IFLA_VXLAN_PORT_RANGE]) {
1262 const struct ifla_vxlan_port_range *p
1263 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1265 if (ntohs(p->high) < ntohs(p->low)) {
1266 pr_debug("port range %u .. %u not valid\n",
1267 ntohs(p->low), ntohs(p->high));
1275 static void vxlan_get_drvinfo(struct net_device *netdev,
1276 struct ethtool_drvinfo *drvinfo)
1278 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
1279 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
1282 static const struct ethtool_ops vxlan_ethtool_ops = {
1283 .get_drvinfo = vxlan_get_drvinfo,
1284 .get_link = ethtool_op_get_link,
1287 static int vxlan_newlink(struct net *net, struct net_device *dev,
1288 struct nlattr *tb[], struct nlattr *data[])
1290 struct vxlan_dev *vxlan = netdev_priv(dev);
1294 if (!data[IFLA_VXLAN_ID])
1297 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1298 if (vxlan_find_vni(net, vni)) {
1299 pr_info("duplicate VNI %u\n", vni);
1304 if (data[IFLA_VXLAN_GROUP])
1305 vxlan->gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1307 if (data[IFLA_VXLAN_LOCAL])
1308 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1310 if (data[IFLA_VXLAN_LINK] &&
1311 (vxlan->link = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
1312 struct net_device *lowerdev
1313 = __dev_get_by_index(net, vxlan->link);
1316 pr_info("ifindex %d does not exist\n", vxlan->link);
1321 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
1323 /* update header length based on lower device */
1324 dev->hard_header_len = lowerdev->hard_header_len +
1328 if (data[IFLA_VXLAN_TOS])
1329 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1331 if (data[IFLA_VXLAN_TTL])
1332 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
1334 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
1335 vxlan->flags |= VXLAN_F_LEARN;
1337 if (data[IFLA_VXLAN_AGEING])
1338 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1340 vxlan->age_interval = FDB_AGE_DEFAULT;
1342 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
1343 vxlan->flags |= VXLAN_F_PROXY;
1345 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
1346 vxlan->flags |= VXLAN_F_RSC;
1348 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
1349 vxlan->flags |= VXLAN_F_L2MISS;
1351 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
1352 vxlan->flags |= VXLAN_F_L3MISS;
1354 if (data[IFLA_VXLAN_LIMIT])
1355 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1357 if (data[IFLA_VXLAN_PORT_RANGE]) {
1358 const struct ifla_vxlan_port_range *p
1359 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1360 vxlan->port_min = ntohs(p->low);
1361 vxlan->port_max = ntohs(p->high);
1364 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
1366 err = register_netdevice(dev);
1368 hlist_add_head_rcu(&vxlan->hlist, vni_head(net, vxlan->vni));
1373 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1375 struct vxlan_dev *vxlan = netdev_priv(dev);
1377 hlist_del_rcu(&vxlan->hlist);
1379 unregister_netdevice_queue(dev, head);
1382 static size_t vxlan_get_size(const struct net_device *dev)
1385 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1386 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1387 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1388 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1389 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1390 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1391 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
1392 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
1393 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
1394 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
1395 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
1396 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1397 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
1398 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
1402 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1404 const struct vxlan_dev *vxlan = netdev_priv(dev);
1405 struct ifla_vxlan_port_range ports = {
1406 .low = htons(vxlan->port_min),
1407 .high = htons(vxlan->port_max),
1410 if (nla_put_u32(skb, IFLA_VXLAN_ID, vxlan->vni))
1411 goto nla_put_failure;
1413 if (vxlan->gaddr && nla_put_be32(skb, IFLA_VXLAN_GROUP, vxlan->gaddr))
1414 goto nla_put_failure;
1416 if (vxlan->link && nla_put_u32(skb, IFLA_VXLAN_LINK, vxlan->link))
1417 goto nla_put_failure;
1419 if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
1420 goto nla_put_failure;
1422 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1423 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
1424 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
1425 !!(vxlan->flags & VXLAN_F_LEARN)) ||
1426 nla_put_u8(skb, IFLA_VXLAN_PROXY,
1427 !!(vxlan->flags & VXLAN_F_PROXY)) ||
1428 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
1429 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
1430 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
1431 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
1432 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
1433 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1434 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax))
1435 goto nla_put_failure;
1437 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
1438 goto nla_put_failure;
1446 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1448 .maxtype = IFLA_VXLAN_MAX,
1449 .policy = vxlan_policy,
1450 .priv_size = sizeof(struct vxlan_dev),
1451 .setup = vxlan_setup,
1452 .validate = vxlan_validate,
1453 .newlink = vxlan_newlink,
1454 .dellink = vxlan_dellink,
1455 .get_size = vxlan_get_size,
1456 .fill_info = vxlan_fill_info,
1459 static __net_init int vxlan_init_net(struct net *net)
1461 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1463 struct sockaddr_in vxlan_addr = {
1464 .sin_family = AF_INET,
1465 .sin_addr.s_addr = htonl(INADDR_ANY),
1470 /* Create UDP socket for encapsulation receive. */
1471 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vn->sock);
1473 pr_debug("UDP socket create failed\n");
1476 /* Put in proper namespace */
1478 sk_change_net(sk, net);
1480 vxlan_addr.sin_port = htons(vxlan_port);
1482 rc = kernel_bind(vn->sock, (struct sockaddr *) &vxlan_addr,
1483 sizeof(vxlan_addr));
1485 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1486 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
1487 sk_release_kernel(sk);
1492 /* Disable multicast loopback */
1493 inet_sk(sk)->mc_loop = 0;
1495 /* Mark socket as an encapsulation socket. */
1496 udp_sk(sk)->encap_type = 1;
1497 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1500 for (h = 0; h < VNI_HASH_SIZE; ++h)
1501 INIT_HLIST_HEAD(&vn->vni_list[h]);
1506 static __net_exit void vxlan_exit_net(struct net *net)
1508 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1511 sk_release_kernel(vn->sock->sk);
1516 static struct pernet_operations vxlan_net_ops = {
1517 .init = vxlan_init_net,
1518 .exit = vxlan_exit_net,
1519 .id = &vxlan_net_id,
1520 .size = sizeof(struct vxlan_net),
1523 static int __init vxlan_init_module(void)
1527 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1529 rc = register_pernet_device(&vxlan_net_ops);
1533 rc = rtnl_link_register(&vxlan_link_ops);
1540 unregister_pernet_device(&vxlan_net_ops);
1544 module_init(vxlan_init_module);
1546 static void __exit vxlan_cleanup_module(void)
1548 rtnl_link_unregister(&vxlan_link_ops);
1549 unregister_pernet_device(&vxlan_net_ops);
1551 module_exit(vxlan_cleanup_module);
1553 MODULE_LICENSE("GPL");
1554 MODULE_VERSION(VXLAN_VERSION);
1555 MODULE_AUTHOR("Stephen Hemminger <shemminger@vyatta.com>");
1556 MODULE_ALIAS_RTNL_LINK("vxlan");