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>
39 #include <net/rtnetlink.h>
40 #include <net/route.h>
41 #include <net/dsfield.h>
42 #include <net/inet_ecn.h>
43 #include <net/net_namespace.h>
44 #include <net/netns/generic.h>
46 #define VXLAN_VERSION "0.1"
48 #define VNI_HASH_BITS 10
49 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
50 #define FDB_HASH_BITS 8
51 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
52 #define FDB_AGE_DEFAULT 300 /* 5 min */
53 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
55 #define VXLAN_N_VID (1u << 24)
56 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
57 /* IP header + UDP + VXLAN + Ethernet header */
58 #define VXLAN_HEADROOM (20 + 8 + 8 + 14)
60 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
62 /* VXLAN protocol header */
68 /* UDP port for VXLAN traffic. */
69 static unsigned int vxlan_port __read_mostly = 8472;
70 module_param_named(udp_port, vxlan_port, uint, 0444);
71 MODULE_PARM_DESC(udp_port, "Destination UDP port");
73 static bool log_ecn_error = true;
74 module_param(log_ecn_error, bool, 0644);
75 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
77 /* per-net private data for this module */
78 static unsigned int vxlan_net_id;
80 struct socket *sock; /* UDP encap socket */
81 struct hlist_head vni_list[VNI_HASH_SIZE];
84 /* Forwarding table entry */
86 struct hlist_node hlist; /* linked list of entries */
88 unsigned long updated; /* jiffies */
91 u16 state; /* see ndm_state */
92 u8 eth_addr[ETH_ALEN];
95 /* Per-cpu network traffic stats */
101 struct u64_stats_sync syncp;
104 /* Pseudo network device */
106 struct hlist_node hlist;
107 struct net_device *dev;
108 struct vxlan_stats __percpu *stats;
109 __u32 vni; /* virtual network id */
110 __be32 gaddr; /* multicast group */
111 __be32 saddr; /* source address */
112 unsigned int link; /* link to multicast over */
113 __u16 port_min; /* source port range */
115 __u8 tos; /* TOS override */
117 u32 flags; /* VXLAN_F_* below */
119 unsigned long age_interval;
120 struct timer_list age_timer;
121 spinlock_t hash_lock;
122 unsigned int addrcnt;
123 unsigned int addrmax;
125 struct hlist_head fdb_head[FDB_HASH_SIZE];
128 #define VXLAN_F_LEARN 0x01
129 #define VXLAN_F_PROXY 0x02
130 #define VXLAN_F_RSC 0x04
131 #define VXLAN_F_L2MISS 0x08
132 #define VXLAN_F_L3MISS 0x10
134 /* salt for hash table */
135 static u32 vxlan_salt __read_mostly;
137 static inline struct hlist_head *vni_head(struct net *net, u32 id)
139 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
141 return &vn->vni_list[hash_32(id, VNI_HASH_BITS)];
144 /* Look up VNI in a per net namespace table */
145 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
147 struct vxlan_dev *vxlan;
148 struct hlist_node *node;
150 hlist_for_each_entry_rcu(vxlan, node, vni_head(net, id), hlist) {
151 if (vxlan->vni == id)
158 /* Fill in neighbour message in skbuff. */
159 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
160 const struct vxlan_fdb *fdb,
161 u32 portid, u32 seq, int type, unsigned int flags)
163 unsigned long now = jiffies;
164 struct nda_cacheinfo ci;
165 struct nlmsghdr *nlh;
167 bool send_ip, send_eth;
169 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
173 ndm = nlmsg_data(nlh);
174 memset(ndm, 0, sizeof(*ndm));
176 send_eth = send_ip = true;
178 if (type == RTM_GETNEIGH) {
179 ndm->ndm_family = AF_INET;
180 send_ip = fdb->remote_ip != 0;
181 send_eth = !is_zero_ether_addr(fdb->eth_addr);
183 ndm->ndm_family = AF_BRIDGE;
184 ndm->ndm_state = fdb->state;
185 ndm->ndm_ifindex = vxlan->dev->ifindex;
186 ndm->ndm_flags = NTF_SELF;
187 ndm->ndm_type = NDA_DST;
189 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
190 goto nla_put_failure;
192 if (send_ip && nla_put_be32(skb, NDA_DST, fdb->remote_ip))
193 goto nla_put_failure;
195 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
196 ci.ndm_confirmed = 0;
197 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
200 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
201 goto nla_put_failure;
203 return nlmsg_end(skb, nlh);
206 nlmsg_cancel(skb, nlh);
210 static inline size_t vxlan_nlmsg_size(void)
212 return NLMSG_ALIGN(sizeof(struct ndmsg))
213 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
214 + nla_total_size(sizeof(__be32)) /* NDA_DST */
215 + nla_total_size(sizeof(struct nda_cacheinfo));
218 static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
219 const struct vxlan_fdb *fdb, int type)
221 struct net *net = dev_net(vxlan->dev);
225 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
229 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0);
231 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
232 WARN_ON(err == -EMSGSIZE);
237 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
241 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
244 static void vxlan_ip_miss(struct net_device *dev, __be32 ipa)
246 struct vxlan_dev *vxlan = netdev_priv(dev);
249 memset(&f, 0, sizeof f);
251 f.remote_ip = ipa; /* goes to NDA_DST */
253 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
256 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
260 memset(&f, 0, sizeof f);
262 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
264 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
267 /* Hash Ethernet address */
268 static u32 eth_hash(const unsigned char *addr)
270 u64 value = get_unaligned((u64 *)addr);
272 /* only want 6 bytes */
278 return hash_64(value, FDB_HASH_BITS);
281 /* Hash chain to use given mac address */
282 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
285 return &vxlan->fdb_head[eth_hash(mac)];
288 /* Look up Ethernet address in forwarding table */
289 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
293 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
295 struct hlist_node *node;
297 hlist_for_each_entry_rcu(f, node, head, hlist) {
298 if (compare_ether_addr(mac, f->eth_addr) == 0)
305 /* Add new entry to forwarding table -- assumes lock held */
306 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
307 const u8 *mac, __be32 ip,
308 __u16 state, __u16 flags)
313 f = vxlan_find_mac(vxlan, mac);
315 if (flags & NLM_F_EXCL) {
316 netdev_dbg(vxlan->dev,
317 "lost race to create %pM\n", mac);
320 if (f->state != state) {
322 f->updated = jiffies;
326 if (!(flags & NLM_F_CREATE))
329 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
332 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
333 f = kmalloc(sizeof(*f), GFP_ATOMIC);
340 f->updated = f->used = jiffies;
341 memcpy(f->eth_addr, mac, ETH_ALEN);
344 hlist_add_head_rcu(&f->hlist,
345 vxlan_fdb_head(vxlan, mac));
349 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
354 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
356 netdev_dbg(vxlan->dev,
357 "delete %pM\n", f->eth_addr);
360 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
362 hlist_del_rcu(&f->hlist);
366 /* Add static entry (via netlink) */
367 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
368 struct net_device *dev,
369 const unsigned char *addr, u16 flags)
371 struct vxlan_dev *vxlan = netdev_priv(dev);
375 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
376 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
381 if (tb[NDA_DST] == NULL)
384 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
385 return -EAFNOSUPPORT;
387 ip = nla_get_be32(tb[NDA_DST]);
389 spin_lock_bh(&vxlan->hash_lock);
390 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags);
391 spin_unlock_bh(&vxlan->hash_lock);
396 /* Delete entry (via netlink) */
397 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
398 struct net_device *dev,
399 const unsigned char *addr)
401 struct vxlan_dev *vxlan = netdev_priv(dev);
405 spin_lock_bh(&vxlan->hash_lock);
406 f = vxlan_find_mac(vxlan, addr);
408 vxlan_fdb_destroy(vxlan, f);
411 spin_unlock_bh(&vxlan->hash_lock);
416 /* Dump forwarding table */
417 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
418 struct net_device *dev, int idx)
420 struct vxlan_dev *vxlan = netdev_priv(dev);
423 for (h = 0; h < FDB_HASH_SIZE; ++h) {
425 struct hlist_node *n;
428 hlist_for_each_entry_rcu(f, n, &vxlan->fdb_head[h], hlist) {
429 if (idx < cb->args[0])
432 err = vxlan_fdb_info(skb, vxlan, f,
433 NETLINK_CB(cb->skb).portid,
447 /* Watch incoming packets to learn mapping between Ethernet address
448 * and Tunnel endpoint.
450 static void vxlan_snoop(struct net_device *dev,
451 __be32 src_ip, const u8 *src_mac)
453 struct vxlan_dev *vxlan = netdev_priv(dev);
457 f = vxlan_find_mac(vxlan, src_mac);
460 if (likely(f->remote_ip == src_ip))
465 "%pM migrated from %pI4 to %pI4\n",
466 src_mac, &f->remote_ip, &src_ip);
468 f->remote_ip = src_ip;
469 f->updated = jiffies;
471 /* learned new entry */
472 spin_lock(&vxlan->hash_lock);
473 err = vxlan_fdb_create(vxlan, src_mac, src_ip,
475 NLM_F_EXCL|NLM_F_CREATE);
476 spin_unlock(&vxlan->hash_lock);
481 /* See if multicast group is already in use by other ID */
482 static bool vxlan_group_used(struct vxlan_net *vn,
483 const struct vxlan_dev *this)
485 const struct vxlan_dev *vxlan;
486 struct hlist_node *node;
489 for (h = 0; h < VNI_HASH_SIZE; ++h)
490 hlist_for_each_entry(vxlan, node, &vn->vni_list[h], hlist) {
494 if (!netif_running(vxlan->dev))
497 if (vxlan->gaddr == this->gaddr)
504 /* kernel equivalent to IP_ADD_MEMBERSHIP */
505 static int vxlan_join_group(struct net_device *dev)
507 struct vxlan_dev *vxlan = netdev_priv(dev);
508 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
509 struct sock *sk = vn->sock->sk;
510 struct ip_mreqn mreq = {
511 .imr_multiaddr.s_addr = vxlan->gaddr,
512 .imr_ifindex = vxlan->link,
516 /* Already a member of group */
517 if (vxlan_group_used(vn, vxlan))
520 /* Need to drop RTNL to call multicast join */
523 err = ip_mc_join_group(sk, &mreq);
531 /* kernel equivalent to IP_DROP_MEMBERSHIP */
532 static int vxlan_leave_group(struct net_device *dev)
534 struct vxlan_dev *vxlan = netdev_priv(dev);
535 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
537 struct sock *sk = vn->sock->sk;
538 struct ip_mreqn mreq = {
539 .imr_multiaddr.s_addr = vxlan->gaddr,
540 .imr_ifindex = vxlan->link,
543 /* Only leave group when last vxlan is done. */
544 if (vxlan_group_used(vn, vxlan))
547 /* Need to drop RTNL to call multicast leave */
550 err = ip_mc_leave_group(sk, &mreq);
557 /* Callback from net/ipv4/udp.c to receive packets */
558 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
561 struct vxlanhdr *vxh;
562 struct vxlan_dev *vxlan;
563 struct vxlan_stats *stats;
567 /* pop off outer UDP header */
568 __skb_pull(skb, sizeof(struct udphdr));
570 /* Need Vxlan and inner Ethernet header to be present */
571 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
574 /* Drop packets with reserved bits set */
575 vxh = (struct vxlanhdr *) skb->data;
576 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
577 (vxh->vx_vni & htonl(0xff))) {
578 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
579 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
583 __skb_pull(skb, sizeof(struct vxlanhdr));
585 /* Is this VNI defined? */
586 vni = ntohl(vxh->vx_vni) >> 8;
587 vxlan = vxlan_find_vni(sock_net(sk), vni);
589 netdev_dbg(skb->dev, "unknown vni %d\n", vni);
593 if (!pskb_may_pull(skb, ETH_HLEN)) {
594 vxlan->dev->stats.rx_length_errors++;
595 vxlan->dev->stats.rx_errors++;
599 skb_reset_mac_header(skb);
601 /* Re-examine inner Ethernet packet */
603 skb->protocol = eth_type_trans(skb, vxlan->dev);
605 /* Ignore packet loops (and multicast echo) */
606 if (compare_ether_addr(eth_hdr(skb)->h_source,
607 vxlan->dev->dev_addr) == 0)
610 if (vxlan->flags & VXLAN_F_LEARN)
611 vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source);
613 __skb_tunnel_rx(skb, vxlan->dev);
614 skb_reset_network_header(skb);
616 /* If the NIC driver gave us an encapsulated packet with
617 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
618 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
619 * for us. Otherwise force the upper layers to verify it.
621 if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
622 !(vxlan->dev->features & NETIF_F_RXCSUM))
623 skb->ip_summed = CHECKSUM_NONE;
625 skb->encapsulation = 0;
627 err = IP_ECN_decapsulate(oip, skb);
630 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
631 &oip->saddr, oip->tos);
633 ++vxlan->dev->stats.rx_frame_errors;
634 ++vxlan->dev->stats.rx_errors;
639 stats = this_cpu_ptr(vxlan->stats);
640 u64_stats_update_begin(&stats->syncp);
642 stats->rx_bytes += skb->len;
643 u64_stats_update_end(&stats->syncp);
649 /* Put UDP header back */
650 __skb_push(skb, sizeof(struct udphdr));
654 /* Consume bad packet */
659 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
661 struct vxlan_dev *vxlan = netdev_priv(dev);
667 if (dev->flags & IFF_NOARP)
670 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
671 dev->stats.tx_dropped++;
676 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
677 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
678 parp->ar_pro != htons(ETH_P_IP) ||
679 parp->ar_op != htons(ARPOP_REQUEST) ||
680 parp->ar_hln != dev->addr_len ||
683 arpptr = (u8 *)parp + sizeof(struct arphdr);
685 arpptr += dev->addr_len; /* sha */
686 memcpy(&sip, arpptr, sizeof(sip));
687 arpptr += sizeof(sip);
688 arpptr += dev->addr_len; /* tha */
689 memcpy(&tip, arpptr, sizeof(tip));
691 if (ipv4_is_loopback(tip) ||
692 ipv4_is_multicast(tip))
695 n = neigh_lookup(&arp_tbl, &tip, dev);
698 struct vxlan_dev *vxlan = netdev_priv(dev);
700 struct sk_buff *reply;
702 if (!(n->nud_state & NUD_CONNECTED)) {
707 f = vxlan_find_mac(vxlan, n->ha);
708 if (f && f->remote_ip == 0) {
709 /* bridge-local neighbor */
714 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
719 skb_reset_mac_header(reply);
720 __skb_pull(reply, skb_network_offset(reply));
721 reply->ip_summed = CHECKSUM_UNNECESSARY;
722 reply->pkt_type = PACKET_HOST;
724 if (netif_rx_ni(reply) == NET_RX_DROP)
725 dev->stats.rx_dropped++;
726 } else if (vxlan->flags & VXLAN_F_L3MISS)
727 vxlan_ip_miss(dev, tip);
733 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
735 struct vxlan_dev *vxlan = netdev_priv(dev);
739 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
743 switch (ntohs(eth_hdr(skb)->h_proto)) {
745 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
748 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
757 diff = compare_ether_addr(eth_hdr(skb)->h_dest, n->ha) != 0;
759 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
761 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
765 } else if (vxlan->flags & VXLAN_F_L3MISS)
766 vxlan_ip_miss(dev, pip->daddr);
770 /* Extract dsfield from inner protocol */
771 static inline u8 vxlan_get_dsfield(const struct iphdr *iph,
772 const struct sk_buff *skb)
774 if (skb->protocol == htons(ETH_P_IP))
776 else if (skb->protocol == htons(ETH_P_IPV6))
777 return ipv6_get_dsfield((const struct ipv6hdr *)iph);
782 /* Propogate ECN bits out */
783 static inline u8 vxlan_ecn_encap(u8 tos,
784 const struct iphdr *iph,
785 const struct sk_buff *skb)
787 u8 inner = vxlan_get_dsfield(iph, skb);
789 return INET_ECN_encapsulate(tos, inner);
792 static void vxlan_sock_free(struct sk_buff *skb)
797 /* On transmit, associate with the tunnel socket */
798 static void vxlan_set_owner(struct net_device *dev, struct sk_buff *skb)
800 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
801 struct sock *sk = vn->sock->sk;
806 skb->destructor = vxlan_sock_free;
809 /* Compute source port for outgoing packet
810 * first choice to use L4 flow hash since it will spread
811 * better and maybe available from hardware
812 * secondary choice is to use jhash on the Ethernet header
814 static u16 vxlan_src_port(const struct vxlan_dev *vxlan, struct sk_buff *skb)
816 unsigned int range = (vxlan->port_max - vxlan->port_min) + 1;
819 hash = skb_get_rxhash(skb);
821 hash = jhash(skb->data, 2 * ETH_ALEN,
822 (__force u32) skb->protocol);
824 return (((u64) hash * range) >> 32) + vxlan->port_min;
827 /* Transmit local packets over Vxlan
829 * Outer IP header inherits ECN and DF from inner header.
830 * Outer UDP destination is the VXLAN assigned port.
831 * source port is based on hash of flow
833 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
835 struct vxlan_dev *vxlan = netdev_priv(dev);
837 const struct iphdr *old_iph;
840 struct vxlanhdr *vxh;
843 unsigned int pkt_len = skb->len;
849 bool did_rsc = false;
850 const struct vxlan_fdb *f;
852 skb_reset_mac_header(skb);
855 if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP)
856 return arp_reduce(dev, skb);
857 else if ((vxlan->flags&VXLAN_F_RSC) && ntohs(eth->h_proto) == ETH_P_IP)
858 did_rsc = route_shortcircuit(dev, skb);
860 f = vxlan_find_mac(vxlan, eth->h_dest);
864 if (!dst && (vxlan->flags & VXLAN_F_L2MISS) &&
865 !is_multicast_ether_addr(eth->h_dest))
866 vxlan_fdb_miss(vxlan, eth->h_dest);
872 __skb_pull(skb, skb_network_offset(skb));
873 skb->ip_summed = CHECKSUM_NONE;
874 skb->pkt_type = PACKET_HOST;
876 /* short-circuited back to local bridge */
877 if (netif_rx(skb) == NET_RX_SUCCESS) {
878 struct vxlan_stats *stats =
879 this_cpu_ptr(vxlan->stats);
881 u64_stats_update_begin(&stats->syncp);
883 stats->tx_bytes += pkt_len;
884 u64_stats_update_end(&stats->syncp);
886 dev->stats.tx_errors++;
887 dev->stats.tx_aborted_errors++;
894 if (!skb->encapsulation) {
895 skb_reset_inner_headers(skb);
896 skb->encapsulation = 1;
899 /* Need space for new headers (invalidates iph ptr) */
900 if (skb_cow_head(skb, VXLAN_HEADROOM))
903 old_iph = ip_hdr(skb);
906 if (!ttl && IN_MULTICAST(ntohl(dst)))
911 tos = vxlan_get_dsfield(old_iph, skb);
913 src_port = vxlan_src_port(vxlan, skb);
915 memset(&fl4, 0, sizeof(fl4));
916 fl4.flowi4_oif = vxlan->link;
917 fl4.flowi4_tos = RT_TOS(tos);
919 fl4.saddr = vxlan->saddr;
921 rt = ip_route_output_key(dev_net(dev), &fl4);
923 netdev_dbg(dev, "no route to %pI4\n", &dst);
924 dev->stats.tx_carrier_errors++;
928 if (rt->dst.dev == dev) {
929 netdev_dbg(dev, "circular route to %pI4\n", &dst);
931 dev->stats.collisions++;
935 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
936 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
939 skb_dst_set(skb, &rt->dst);
941 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
942 vxh->vx_flags = htonl(VXLAN_FLAGS);
943 vxh->vx_vni = htonl(vxlan->vni << 8);
945 __skb_push(skb, sizeof(*uh));
946 skb_reset_transport_header(skb);
949 uh->dest = htons(vxlan_port);
950 uh->source = htons(src_port);
952 uh->len = htons(skb->len);
955 __skb_push(skb, sizeof(*iph));
956 skb_reset_network_header(skb);
959 iph->ihl = sizeof(struct iphdr) >> 2;
961 iph->protocol = IPPROTO_UDP;
962 iph->tos = vxlan_ecn_encap(tos, old_iph, skb);
964 iph->saddr = fl4.saddr;
965 iph->ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
966 tunnel_ip_select_ident(skb, old_iph, &rt->dst);
968 vxlan_set_owner(dev, skb);
970 /* See iptunnel_xmit() */
971 if (skb->ip_summed != CHECKSUM_PARTIAL)
972 skb->ip_summed = CHECKSUM_NONE;
974 err = ip_local_out(skb);
975 if (likely(net_xmit_eval(err) == 0)) {
976 struct vxlan_stats *stats = this_cpu_ptr(vxlan->stats);
978 u64_stats_update_begin(&stats->syncp);
980 stats->tx_bytes += pkt_len;
981 u64_stats_update_end(&stats->syncp);
983 dev->stats.tx_errors++;
984 dev->stats.tx_aborted_errors++;
989 dev->stats.tx_dropped++;
993 dev->stats.tx_errors++;
999 /* Walk the forwarding table and purge stale entries */
1000 static void vxlan_cleanup(unsigned long arg)
1002 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1003 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1006 if (!netif_running(vxlan->dev))
1009 spin_lock_bh(&vxlan->hash_lock);
1010 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1011 struct hlist_node *p, *n;
1012 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1014 = container_of(p, struct vxlan_fdb, hlist);
1015 unsigned long timeout;
1017 if (f->state & NUD_PERMANENT)
1020 timeout = f->used + vxlan->age_interval * HZ;
1021 if (time_before_eq(timeout, jiffies)) {
1022 netdev_dbg(vxlan->dev,
1023 "garbage collect %pM\n",
1025 f->state = NUD_STALE;
1026 vxlan_fdb_destroy(vxlan, f);
1027 } else if (time_before(timeout, next_timer))
1028 next_timer = timeout;
1031 spin_unlock_bh(&vxlan->hash_lock);
1033 mod_timer(&vxlan->age_timer, next_timer);
1036 /* Setup stats when device is created */
1037 static int vxlan_init(struct net_device *dev)
1039 struct vxlan_dev *vxlan = netdev_priv(dev);
1041 vxlan->stats = alloc_percpu(struct vxlan_stats);
1048 /* Start ageing timer and join group when device is brought up */
1049 static int vxlan_open(struct net_device *dev)
1051 struct vxlan_dev *vxlan = netdev_priv(dev);
1055 err = vxlan_join_group(dev);
1060 if (vxlan->age_interval)
1061 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
1066 /* Purge the forwarding table */
1067 static void vxlan_flush(struct vxlan_dev *vxlan)
1071 spin_lock_bh(&vxlan->hash_lock);
1072 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1073 struct hlist_node *p, *n;
1074 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1076 = container_of(p, struct vxlan_fdb, hlist);
1077 vxlan_fdb_destroy(vxlan, f);
1080 spin_unlock_bh(&vxlan->hash_lock);
1083 /* Cleanup timer and forwarding table on shutdown */
1084 static int vxlan_stop(struct net_device *dev)
1086 struct vxlan_dev *vxlan = netdev_priv(dev);
1089 vxlan_leave_group(dev);
1091 del_timer_sync(&vxlan->age_timer);
1098 /* Merge per-cpu statistics */
1099 static struct rtnl_link_stats64 *vxlan_stats64(struct net_device *dev,
1100 struct rtnl_link_stats64 *stats)
1102 struct vxlan_dev *vxlan = netdev_priv(dev);
1103 struct vxlan_stats tmp, sum = { 0 };
1106 for_each_possible_cpu(cpu) {
1108 const struct vxlan_stats *stats
1109 = per_cpu_ptr(vxlan->stats, cpu);
1112 start = u64_stats_fetch_begin_bh(&stats->syncp);
1113 memcpy(&tmp, stats, sizeof(tmp));
1114 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1116 sum.tx_bytes += tmp.tx_bytes;
1117 sum.tx_packets += tmp.tx_packets;
1118 sum.rx_bytes += tmp.rx_bytes;
1119 sum.rx_packets += tmp.rx_packets;
1122 stats->tx_bytes = sum.tx_bytes;
1123 stats->tx_packets = sum.tx_packets;
1124 stats->rx_bytes = sum.rx_bytes;
1125 stats->rx_packets = sum.rx_packets;
1127 stats->multicast = dev->stats.multicast;
1128 stats->rx_length_errors = dev->stats.rx_length_errors;
1129 stats->rx_frame_errors = dev->stats.rx_frame_errors;
1130 stats->rx_errors = dev->stats.rx_errors;
1132 stats->tx_dropped = dev->stats.tx_dropped;
1133 stats->tx_carrier_errors = dev->stats.tx_carrier_errors;
1134 stats->tx_aborted_errors = dev->stats.tx_aborted_errors;
1135 stats->collisions = dev->stats.collisions;
1136 stats->tx_errors = dev->stats.tx_errors;
1141 /* Stub, nothing needs to be done. */
1142 static void vxlan_set_multicast_list(struct net_device *dev)
1146 static const struct net_device_ops vxlan_netdev_ops = {
1147 .ndo_init = vxlan_init,
1148 .ndo_open = vxlan_open,
1149 .ndo_stop = vxlan_stop,
1150 .ndo_start_xmit = vxlan_xmit,
1151 .ndo_get_stats64 = vxlan_stats64,
1152 .ndo_set_rx_mode = vxlan_set_multicast_list,
1153 .ndo_change_mtu = eth_change_mtu,
1154 .ndo_validate_addr = eth_validate_addr,
1155 .ndo_set_mac_address = eth_mac_addr,
1156 .ndo_fdb_add = vxlan_fdb_add,
1157 .ndo_fdb_del = vxlan_fdb_delete,
1158 .ndo_fdb_dump = vxlan_fdb_dump,
1161 /* Info for udev, that this is a virtual tunnel endpoint */
1162 static struct device_type vxlan_type = {
1166 static void vxlan_free(struct net_device *dev)
1168 struct vxlan_dev *vxlan = netdev_priv(dev);
1170 free_percpu(vxlan->stats);
1174 /* Initialize the device structure. */
1175 static void vxlan_setup(struct net_device *dev)
1177 struct vxlan_dev *vxlan = netdev_priv(dev);
1181 eth_hw_addr_random(dev);
1183 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
1185 dev->netdev_ops = &vxlan_netdev_ops;
1186 dev->destructor = vxlan_free;
1187 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
1189 dev->tx_queue_len = 0;
1190 dev->features |= NETIF_F_LLTX;
1191 dev->features |= NETIF_F_NETNS_LOCAL;
1192 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1193 dev->features |= NETIF_F_RXCSUM;
1195 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1196 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1197 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1199 spin_lock_init(&vxlan->hash_lock);
1201 init_timer_deferrable(&vxlan->age_timer);
1202 vxlan->age_timer.function = vxlan_cleanup;
1203 vxlan->age_timer.data = (unsigned long) vxlan;
1205 inet_get_local_port_range(&low, &high);
1206 vxlan->port_min = low;
1207 vxlan->port_max = high;
1211 for (h = 0; h < FDB_HASH_SIZE; ++h)
1212 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
1215 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
1216 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
1217 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1218 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
1219 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1220 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
1221 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
1222 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
1223 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
1224 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
1225 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
1226 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
1227 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
1228 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
1229 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
1232 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
1234 if (tb[IFLA_ADDRESS]) {
1235 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1236 pr_debug("invalid link address (not ethernet)\n");
1240 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1241 pr_debug("invalid all zero ethernet address\n");
1242 return -EADDRNOTAVAIL;
1249 if (data[IFLA_VXLAN_ID]) {
1250 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
1251 if (id >= VXLAN_VID_MASK)
1255 if (data[IFLA_VXLAN_GROUP]) {
1256 __be32 gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1257 if (!IN_MULTICAST(ntohl(gaddr))) {
1258 pr_debug("group address is not IPv4 multicast\n");
1259 return -EADDRNOTAVAIL;
1263 if (data[IFLA_VXLAN_PORT_RANGE]) {
1264 const struct ifla_vxlan_port_range *p
1265 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1267 if (ntohs(p->high) < ntohs(p->low)) {
1268 pr_debug("port range %u .. %u not valid\n",
1269 ntohs(p->low), ntohs(p->high));
1277 static void vxlan_get_drvinfo(struct net_device *netdev,
1278 struct ethtool_drvinfo *drvinfo)
1280 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
1281 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
1284 static const struct ethtool_ops vxlan_ethtool_ops = {
1285 .get_drvinfo = vxlan_get_drvinfo,
1286 .get_link = ethtool_op_get_link,
1289 static int vxlan_newlink(struct net *net, struct net_device *dev,
1290 struct nlattr *tb[], struct nlattr *data[])
1292 struct vxlan_dev *vxlan = netdev_priv(dev);
1296 if (!data[IFLA_VXLAN_ID])
1299 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1300 if (vxlan_find_vni(net, vni)) {
1301 pr_info("duplicate VNI %u\n", vni);
1306 if (data[IFLA_VXLAN_GROUP])
1307 vxlan->gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1309 if (data[IFLA_VXLAN_LOCAL])
1310 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1312 if (data[IFLA_VXLAN_LINK] &&
1313 (vxlan->link = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
1314 struct net_device *lowerdev
1315 = __dev_get_by_index(net, vxlan->link);
1318 pr_info("ifindex %d does not exist\n", vxlan->link);
1323 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
1325 /* update header length based on lower device */
1326 dev->hard_header_len = lowerdev->hard_header_len +
1330 if (data[IFLA_VXLAN_TOS])
1331 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1333 if (data[IFLA_VXLAN_TTL])
1334 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
1336 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
1337 vxlan->flags |= VXLAN_F_LEARN;
1339 if (data[IFLA_VXLAN_AGEING])
1340 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1342 vxlan->age_interval = FDB_AGE_DEFAULT;
1344 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
1345 vxlan->flags |= VXLAN_F_PROXY;
1347 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
1348 vxlan->flags |= VXLAN_F_RSC;
1350 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
1351 vxlan->flags |= VXLAN_F_L2MISS;
1353 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
1354 vxlan->flags |= VXLAN_F_L3MISS;
1356 if (data[IFLA_VXLAN_LIMIT])
1357 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1359 if (data[IFLA_VXLAN_PORT_RANGE]) {
1360 const struct ifla_vxlan_port_range *p
1361 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1362 vxlan->port_min = ntohs(p->low);
1363 vxlan->port_max = ntohs(p->high);
1366 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
1368 err = register_netdevice(dev);
1370 hlist_add_head_rcu(&vxlan->hlist, vni_head(net, vxlan->vni));
1375 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1377 struct vxlan_dev *vxlan = netdev_priv(dev);
1379 hlist_del_rcu(&vxlan->hlist);
1381 unregister_netdevice_queue(dev, head);
1384 static size_t vxlan_get_size(const struct net_device *dev)
1387 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1388 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1389 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1390 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1391 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1392 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1393 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
1394 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
1395 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
1396 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
1397 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
1398 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1399 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
1400 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
1404 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1406 const struct vxlan_dev *vxlan = netdev_priv(dev);
1407 struct ifla_vxlan_port_range ports = {
1408 .low = htons(vxlan->port_min),
1409 .high = htons(vxlan->port_max),
1412 if (nla_put_u32(skb, IFLA_VXLAN_ID, vxlan->vni))
1413 goto nla_put_failure;
1415 if (vxlan->gaddr && nla_put_be32(skb, IFLA_VXLAN_GROUP, vxlan->gaddr))
1416 goto nla_put_failure;
1418 if (vxlan->link && nla_put_u32(skb, IFLA_VXLAN_LINK, vxlan->link))
1419 goto nla_put_failure;
1421 if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
1422 goto nla_put_failure;
1424 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1425 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
1426 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
1427 !!(vxlan->flags & VXLAN_F_LEARN)) ||
1428 nla_put_u8(skb, IFLA_VXLAN_PROXY,
1429 !!(vxlan->flags & VXLAN_F_PROXY)) ||
1430 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
1431 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
1432 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
1433 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
1434 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
1435 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1436 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax))
1437 goto nla_put_failure;
1439 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
1440 goto nla_put_failure;
1448 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1450 .maxtype = IFLA_VXLAN_MAX,
1451 .policy = vxlan_policy,
1452 .priv_size = sizeof(struct vxlan_dev),
1453 .setup = vxlan_setup,
1454 .validate = vxlan_validate,
1455 .newlink = vxlan_newlink,
1456 .dellink = vxlan_dellink,
1457 .get_size = vxlan_get_size,
1458 .fill_info = vxlan_fill_info,
1461 static __net_init int vxlan_init_net(struct net *net)
1463 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1465 struct sockaddr_in vxlan_addr = {
1466 .sin_family = AF_INET,
1467 .sin_addr.s_addr = htonl(INADDR_ANY),
1472 /* Create UDP socket for encapsulation receive. */
1473 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vn->sock);
1475 pr_debug("UDP socket create failed\n");
1478 /* Put in proper namespace */
1480 sk_change_net(sk, net);
1482 vxlan_addr.sin_port = htons(vxlan_port);
1484 rc = kernel_bind(vn->sock, (struct sockaddr *) &vxlan_addr,
1485 sizeof(vxlan_addr));
1487 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1488 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
1489 sk_release_kernel(sk);
1494 /* Disable multicast loopback */
1495 inet_sk(sk)->mc_loop = 0;
1497 /* Mark socket as an encapsulation socket. */
1498 udp_sk(sk)->encap_type = 1;
1499 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1502 for (h = 0; h < VNI_HASH_SIZE; ++h)
1503 INIT_HLIST_HEAD(&vn->vni_list[h]);
1508 static __net_exit void vxlan_exit_net(struct net *net)
1510 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1513 sk_release_kernel(vn->sock->sk);
1518 static struct pernet_operations vxlan_net_ops = {
1519 .init = vxlan_init_net,
1520 .exit = vxlan_exit_net,
1521 .id = &vxlan_net_id,
1522 .size = sizeof(struct vxlan_net),
1525 static int __init vxlan_init_module(void)
1529 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1531 rc = register_pernet_device(&vxlan_net_ops);
1535 rc = rtnl_link_register(&vxlan_link_ops);
1542 unregister_pernet_device(&vxlan_net_ops);
1546 module_init(vxlan_init_module);
1548 static void __exit vxlan_cleanup_module(void)
1550 rtnl_link_unregister(&vxlan_link_ops);
1551 unregister_pernet_device(&vxlan_net_ops);
1553 module_exit(vxlan_cleanup_module);
1555 MODULE_LICENSE("GPL");
1556 MODULE_VERSION(VXLAN_VERSION);
1557 MODULE_AUTHOR("Stephen Hemminger <shemminger@vyatta.com>");
1558 MODULE_ALIAS_RTNL_LINK("vxlan");