2 * VXLAN: Virtual eXtensible Local Area Network
4 * Copyright (c) 2012-2013 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.
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16 #include <linux/kernel.h>
17 #include <linux/types.h>
18 #include <linux/module.h>
19 #include <linux/errno.h>
20 #include <linux/slab.h>
21 #include <linux/skbuff.h>
22 #include <linux/rculist.h>
23 #include <linux/netdevice.h>
26 #include <linux/udp.h>
27 #include <linux/igmp.h>
28 #include <linux/etherdevice.h>
29 #include <linux/if_ether.h>
30 #include <linux/hash.h>
31 #include <linux/ethtool.h>
33 #include <net/ndisc.h>
35 #include <net/ip_tunnels.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 PORT_HASH_BITS 8
48 #define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
49 #define VNI_HASH_BITS 10
50 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
51 #define FDB_HASH_BITS 8
52 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
53 #define FDB_AGE_DEFAULT 300 /* 5 min */
54 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
56 #define VXLAN_N_VID (1u << 24)
57 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
58 /* IP header + UDP + VXLAN + Ethernet header */
59 #define VXLAN_HEADROOM (20 + 8 + 8 + 14)
61 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
63 /* VXLAN protocol header */
69 /* UDP port for VXLAN traffic.
70 * The IANA assigned port is 4789, but the Linux default is 8472
71 * for compatibility with early adopters.
73 static unsigned short vxlan_port __read_mostly = 8472;
74 module_param_named(udp_port, vxlan_port, ushort, 0444);
75 MODULE_PARM_DESC(udp_port, "Destination UDP port");
77 static bool log_ecn_error = true;
78 module_param(log_ecn_error, bool, 0644);
79 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
81 static int vxlan_net_id;
83 static const u8 all_zeros_mac[ETH_ALEN];
85 /* per UDP socket information */
87 struct hlist_node hlist;
89 struct work_struct del_work;
92 struct hlist_head vni_list[VNI_HASH_SIZE];
95 /* per-network namespace private data for this module */
97 struct list_head vxlan_list;
98 struct hlist_head sock_list[PORT_HASH_SIZE];
107 struct list_head list;
111 /* Forwarding table entry */
113 struct hlist_node hlist; /* linked list of entries */
115 unsigned long updated; /* jiffies */
117 struct list_head remotes;
118 u16 state; /* see ndm_state */
119 u8 flags; /* see ndm_flags */
120 u8 eth_addr[ETH_ALEN];
123 /* Pseudo network device */
125 struct hlist_node hlist; /* vni hash table */
126 struct list_head next; /* vxlan's per namespace list */
127 struct vxlan_sock *vn_sock; /* listening socket */
128 struct net_device *dev;
129 struct vxlan_rdst default_dst; /* default destination */
130 __be32 saddr; /* source address */
132 __u16 port_min; /* source port range */
134 __u8 tos; /* TOS override */
136 u32 flags; /* VXLAN_F_* below */
138 struct work_struct sock_work;
139 struct work_struct igmp_work;
141 unsigned long age_interval;
142 struct timer_list age_timer;
143 spinlock_t hash_lock;
144 unsigned int addrcnt;
145 unsigned int addrmax;
147 struct hlist_head fdb_head[FDB_HASH_SIZE];
150 #define VXLAN_F_LEARN 0x01
151 #define VXLAN_F_PROXY 0x02
152 #define VXLAN_F_RSC 0x04
153 #define VXLAN_F_L2MISS 0x08
154 #define VXLAN_F_L3MISS 0x10
156 /* salt for hash table */
157 static u32 vxlan_salt __read_mostly;
158 static struct workqueue_struct *vxlan_wq;
160 static void vxlan_sock_work(struct work_struct *work);
162 /* Virtual Network hash table head */
163 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
165 return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
168 /* Socket hash table head */
169 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
171 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
173 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
176 /* First remote destination for a forwarding entry.
177 * Guaranteed to be non-NULL because remotes are never deleted.
179 static inline struct vxlan_rdst *first_remote(struct vxlan_fdb *fdb)
181 return list_first_or_null_rcu(&fdb->remotes, struct vxlan_rdst, list);
184 /* Find VXLAN socket based on network namespace and UDP port */
185 static struct vxlan_sock *vxlan_find_port(struct net *net, __be16 port)
187 struct vxlan_sock *vs;
189 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
190 if (inet_sk(vs->sock->sk)->inet_sport == port)
196 /* Look up VNI in a per net namespace table */
197 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id, __be16 port)
199 struct vxlan_sock *vs;
200 struct vxlan_dev *vxlan;
202 vs = vxlan_find_port(net, port);
206 hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
207 if (vxlan->default_dst.remote_vni == id)
214 /* Fill in neighbour message in skbuff. */
215 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
216 const struct vxlan_fdb *fdb,
217 u32 portid, u32 seq, int type, unsigned int flags,
218 const struct vxlan_rdst *rdst)
220 unsigned long now = jiffies;
221 struct nda_cacheinfo ci;
222 struct nlmsghdr *nlh;
224 bool send_ip, send_eth;
226 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
230 ndm = nlmsg_data(nlh);
231 memset(ndm, 0, sizeof(*ndm));
233 send_eth = send_ip = true;
235 if (type == RTM_GETNEIGH) {
236 ndm->ndm_family = AF_INET;
237 send_ip = rdst->remote_ip != htonl(INADDR_ANY);
238 send_eth = !is_zero_ether_addr(fdb->eth_addr);
240 ndm->ndm_family = AF_BRIDGE;
241 ndm->ndm_state = fdb->state;
242 ndm->ndm_ifindex = vxlan->dev->ifindex;
243 ndm->ndm_flags = fdb->flags;
244 ndm->ndm_type = NDA_DST;
246 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
247 goto nla_put_failure;
249 if (send_ip && nla_put_be32(skb, NDA_DST, rdst->remote_ip))
250 goto nla_put_failure;
252 if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
253 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
254 goto nla_put_failure;
255 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
256 nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
257 goto nla_put_failure;
258 if (rdst->remote_ifindex &&
259 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
260 goto nla_put_failure;
262 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
263 ci.ndm_confirmed = 0;
264 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
267 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
268 goto nla_put_failure;
270 return nlmsg_end(skb, nlh);
273 nlmsg_cancel(skb, nlh);
277 static inline size_t vxlan_nlmsg_size(void)
279 return NLMSG_ALIGN(sizeof(struct ndmsg))
280 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
281 + nla_total_size(sizeof(__be32)) /* NDA_DST */
282 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
283 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
284 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
285 + nla_total_size(sizeof(struct nda_cacheinfo));
288 static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
289 struct vxlan_fdb *fdb, int type)
291 struct net *net = dev_net(vxlan->dev);
295 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
299 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, first_remote(fdb));
301 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
302 WARN_ON(err == -EMSGSIZE);
307 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
311 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
314 static void vxlan_ip_miss(struct net_device *dev, __be32 ipa)
316 struct vxlan_dev *vxlan = netdev_priv(dev);
317 struct vxlan_fdb f = {
320 struct vxlan_rdst remote = {
321 .remote_ip = ipa, /* goes to NDA_DST */
322 .remote_vni = VXLAN_N_VID,
325 INIT_LIST_HEAD(&f.remotes);
326 list_add_rcu(&remote.list, &f.remotes);
328 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
331 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
333 struct vxlan_fdb f = {
337 INIT_LIST_HEAD(&f.remotes);
338 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
340 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
343 /* Hash Ethernet address */
344 static u32 eth_hash(const unsigned char *addr)
346 u64 value = get_unaligned((u64 *)addr);
348 /* only want 6 bytes */
354 return hash_64(value, FDB_HASH_BITS);
357 /* Hash chain to use given mac address */
358 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
361 return &vxlan->fdb_head[eth_hash(mac)];
364 /* Look up Ethernet address in forwarding table */
365 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
369 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
372 hlist_for_each_entry_rcu(f, head, hlist) {
373 if (compare_ether_addr(mac, f->eth_addr) == 0)
380 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
385 f = __vxlan_find_mac(vxlan, mac);
392 /* caller should hold vxlan->hash_lock */
393 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
394 __be32 ip, __be16 port,
395 __u32 vni, __u32 ifindex)
397 struct vxlan_rdst *rd;
399 list_for_each_entry(rd, &f->remotes, list) {
400 if (rd->remote_ip == ip &&
401 rd->remote_port == port &&
402 rd->remote_vni == vni &&
403 rd->remote_ifindex == ifindex)
410 /* Add/update destinations for multicast */
411 static int vxlan_fdb_append(struct vxlan_fdb *f,
412 __be32 ip, __be16 port, __u32 vni, __u32 ifindex)
414 struct vxlan_rdst *rd;
416 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
420 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
424 rd->remote_port = port;
425 rd->remote_vni = vni;
426 rd->remote_ifindex = ifindex;
428 list_add_tail_rcu(&rd->list, &f->remotes);
433 /* Add new entry to forwarding table -- assumes lock held */
434 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
435 const u8 *mac, __be32 ip,
436 __u16 state, __u16 flags,
437 __be16 port, __u32 vni, __u32 ifindex,
443 f = __vxlan_find_mac(vxlan, mac);
445 if (flags & NLM_F_EXCL) {
446 netdev_dbg(vxlan->dev,
447 "lost race to create %pM\n", mac);
450 if (f->state != state) {
452 f->updated = jiffies;
455 if (f->flags != ndm_flags) {
456 f->flags = ndm_flags;
457 f->updated = jiffies;
460 if ((flags & NLM_F_APPEND) &&
461 is_multicast_ether_addr(f->eth_addr)) {
462 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex);
469 if (!(flags & NLM_F_CREATE))
472 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
475 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
476 f = kmalloc(sizeof(*f), GFP_ATOMIC);
482 f->flags = ndm_flags;
483 f->updated = f->used = jiffies;
484 INIT_LIST_HEAD(&f->remotes);
485 memcpy(f->eth_addr, mac, ETH_ALEN);
487 vxlan_fdb_append(f, ip, port, vni, ifindex);
490 hlist_add_head_rcu(&f->hlist,
491 vxlan_fdb_head(vxlan, mac));
495 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
500 static void vxlan_fdb_free_rdst(struct rcu_head *head)
502 struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
506 static void vxlan_fdb_free(struct rcu_head *head)
508 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
509 struct vxlan_rdst *rd, *nd;
511 list_for_each_entry_safe(rd, nd, &f->remotes, list)
516 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
518 netdev_dbg(vxlan->dev,
519 "delete %pM\n", f->eth_addr);
522 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
524 hlist_del_rcu(&f->hlist);
525 call_rcu(&f->rcu, vxlan_fdb_free);
528 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
529 __be32 *ip, __be16 *port, u32 *vni, u32 *ifindex)
531 struct net *net = dev_net(vxlan->dev);
534 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
535 return -EAFNOSUPPORT;
537 *ip = nla_get_be32(tb[NDA_DST]);
539 *ip = htonl(INADDR_ANY);
543 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
545 *port = nla_get_be16(tb[NDA_PORT]);
547 *port = vxlan->dst_port;
551 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
553 *vni = nla_get_u32(tb[NDA_VNI]);
555 *vni = vxlan->default_dst.remote_vni;
558 if (tb[NDA_IFINDEX]) {
559 struct net_device *tdev;
561 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
563 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
564 tdev = dev_get_by_index(net, *ifindex);
566 return -EADDRNOTAVAIL;
575 /* Add static entry (via netlink) */
576 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
577 struct net_device *dev,
578 const unsigned char *addr, u16 flags)
580 struct vxlan_dev *vxlan = netdev_priv(dev);
581 /* struct net *net = dev_net(vxlan->dev); */
587 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
588 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
593 if (tb[NDA_DST] == NULL)
596 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
600 spin_lock_bh(&vxlan->hash_lock);
601 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags,
602 port, vni, ifindex, ndm->ndm_flags);
603 spin_unlock_bh(&vxlan->hash_lock);
608 /* Delete entry (via netlink) */
609 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
610 struct net_device *dev,
611 const unsigned char *addr)
613 struct vxlan_dev *vxlan = netdev_priv(dev);
615 struct vxlan_rdst *rd = NULL;
621 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
627 spin_lock_bh(&vxlan->hash_lock);
628 f = vxlan_find_mac(vxlan, addr);
632 if (ip != htonl(INADDR_ANY)) {
633 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
640 /* remove a destination if it's not the only one on the list,
641 * otherwise destroy the fdb entry
643 if (rd && !list_is_singular(&f->remotes)) {
644 list_del_rcu(&rd->list);
645 call_rcu(&rd->rcu, vxlan_fdb_free_rdst);
649 vxlan_fdb_destroy(vxlan, f);
652 spin_unlock_bh(&vxlan->hash_lock);
657 /* Dump forwarding table */
658 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
659 struct net_device *dev, int idx)
661 struct vxlan_dev *vxlan = netdev_priv(dev);
664 for (h = 0; h < FDB_HASH_SIZE; ++h) {
668 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
669 struct vxlan_rdst *rd;
671 if (idx < cb->args[0])
674 list_for_each_entry_rcu(rd, &f->remotes, list) {
675 err = vxlan_fdb_info(skb, vxlan, f,
676 NETLINK_CB(cb->skb).portid,
691 /* Watch incoming packets to learn mapping between Ethernet address
692 * and Tunnel endpoint.
693 * Return true if packet is bogus and should be droppped.
695 static bool vxlan_snoop(struct net_device *dev,
696 __be32 src_ip, const u8 *src_mac)
698 struct vxlan_dev *vxlan = netdev_priv(dev);
701 f = vxlan_find_mac(vxlan, src_mac);
703 struct vxlan_rdst *rdst = first_remote(f);
705 if (likely(rdst->remote_ip == src_ip))
708 /* Don't migrate static entries, drop packets */
709 if (f->state & NUD_NOARP)
714 "%pM migrated from %pI4 to %pI4\n",
715 src_mac, &rdst->remote_ip, &src_ip);
717 rdst->remote_ip = src_ip;
718 f->updated = jiffies;
719 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
721 /* learned new entry */
722 spin_lock(&vxlan->hash_lock);
724 /* close off race between vxlan_flush and incoming packets */
725 if (netif_running(dev))
726 vxlan_fdb_create(vxlan, src_mac, src_ip,
728 NLM_F_EXCL|NLM_F_CREATE,
730 vxlan->default_dst.remote_vni,
732 spin_unlock(&vxlan->hash_lock);
739 /* See if multicast group is already in use by other ID */
740 static bool vxlan_group_used(struct vxlan_net *vn, __be32 remote_ip)
742 struct vxlan_dev *vxlan;
744 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
745 if (!netif_running(vxlan->dev))
748 if (vxlan->default_dst.remote_ip == remote_ip)
755 static void vxlan_sock_hold(struct vxlan_sock *vs)
757 atomic_inc(&vs->refcnt);
760 static void vxlan_sock_release(struct vxlan_net *vn, struct vxlan_sock *vs)
762 if (!atomic_dec_and_test(&vs->refcnt))
765 spin_lock(&vn->sock_lock);
766 hlist_del_rcu(&vs->hlist);
767 spin_unlock(&vn->sock_lock);
769 queue_work(vxlan_wq, &vs->del_work);
772 /* Callback to update multicast group membership.
773 * Scheduled when vxlan goes up/down.
775 static void vxlan_igmp_work(struct work_struct *work)
777 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_work);
778 struct vxlan_net *vn = net_generic(dev_net(vxlan->dev), vxlan_net_id);
779 struct vxlan_sock *vs = vxlan->vn_sock;
780 struct sock *sk = vs->sock->sk;
781 struct ip_mreqn mreq = {
782 .imr_multiaddr.s_addr = vxlan->default_dst.remote_ip,
783 .imr_ifindex = vxlan->default_dst.remote_ifindex,
787 if (vxlan_group_used(vn, vxlan->default_dst.remote_ip))
788 ip_mc_join_group(sk, &mreq);
790 ip_mc_leave_group(sk, &mreq);
793 vxlan_sock_release(vn, vs);
797 /* Callback from net/ipv4/udp.c to receive packets */
798 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
801 struct vxlanhdr *vxh;
802 struct vxlan_dev *vxlan;
803 struct pcpu_tstats *stats;
808 /* pop off outer UDP header */
809 __skb_pull(skb, sizeof(struct udphdr));
811 /* Need Vxlan and inner Ethernet header to be present */
812 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
815 /* Drop packets with reserved bits set */
816 vxh = (struct vxlanhdr *) skb->data;
817 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
818 (vxh->vx_vni & htonl(0xff))) {
819 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
820 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
824 __skb_pull(skb, sizeof(struct vxlanhdr));
826 /* Is this VNI defined? */
827 vni = ntohl(vxh->vx_vni) >> 8;
828 port = inet_sk(sk)->inet_sport;
829 vxlan = vxlan_find_vni(sock_net(sk), vni, port);
831 netdev_dbg(skb->dev, "unknown vni %d port %u\n",
836 if (!pskb_may_pull(skb, ETH_HLEN)) {
837 vxlan->dev->stats.rx_length_errors++;
838 vxlan->dev->stats.rx_errors++;
842 skb_reset_mac_header(skb);
844 /* Re-examine inner Ethernet packet */
846 skb->protocol = eth_type_trans(skb, vxlan->dev);
848 /* Ignore packet loops (and multicast echo) */
849 if (compare_ether_addr(eth_hdr(skb)->h_source,
850 vxlan->dev->dev_addr) == 0)
853 if ((vxlan->flags & VXLAN_F_LEARN) &&
854 vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source))
857 __skb_tunnel_rx(skb, vxlan->dev);
858 skb_reset_network_header(skb);
860 /* If the NIC driver gave us an encapsulated packet with
861 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
862 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
863 * for us. Otherwise force the upper layers to verify it.
865 if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
866 !(vxlan->dev->features & NETIF_F_RXCSUM))
867 skb->ip_summed = CHECKSUM_NONE;
869 skb->encapsulation = 0;
871 err = IP_ECN_decapsulate(oip, skb);
874 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
875 &oip->saddr, oip->tos);
877 ++vxlan->dev->stats.rx_frame_errors;
878 ++vxlan->dev->stats.rx_errors;
883 stats = this_cpu_ptr(vxlan->dev->tstats);
884 u64_stats_update_begin(&stats->syncp);
886 stats->rx_bytes += skb->len;
887 u64_stats_update_end(&stats->syncp);
893 /* Put UDP header back */
894 __skb_push(skb, sizeof(struct udphdr));
898 /* Consume bad packet */
903 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
905 struct vxlan_dev *vxlan = netdev_priv(dev);
911 if (dev->flags & IFF_NOARP)
914 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
915 dev->stats.tx_dropped++;
920 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
921 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
922 parp->ar_pro != htons(ETH_P_IP) ||
923 parp->ar_op != htons(ARPOP_REQUEST) ||
924 parp->ar_hln != dev->addr_len ||
927 arpptr = (u8 *)parp + sizeof(struct arphdr);
929 arpptr += dev->addr_len; /* sha */
930 memcpy(&sip, arpptr, sizeof(sip));
931 arpptr += sizeof(sip);
932 arpptr += dev->addr_len; /* tha */
933 memcpy(&tip, arpptr, sizeof(tip));
935 if (ipv4_is_loopback(tip) ||
936 ipv4_is_multicast(tip))
939 n = neigh_lookup(&arp_tbl, &tip, dev);
943 struct sk_buff *reply;
945 if (!(n->nud_state & NUD_CONNECTED)) {
950 f = vxlan_find_mac(vxlan, n->ha);
951 if (f && first_remote(f)->remote_ip == htonl(INADDR_ANY)) {
952 /* bridge-local neighbor */
957 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
962 skb_reset_mac_header(reply);
963 __skb_pull(reply, skb_network_offset(reply));
964 reply->ip_summed = CHECKSUM_UNNECESSARY;
965 reply->pkt_type = PACKET_HOST;
967 if (netif_rx_ni(reply) == NET_RX_DROP)
968 dev->stats.rx_dropped++;
969 } else if (vxlan->flags & VXLAN_F_L3MISS)
970 vxlan_ip_miss(dev, tip);
976 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
978 struct vxlan_dev *vxlan = netdev_priv(dev);
982 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
986 switch (ntohs(eth_hdr(skb)->h_proto)) {
988 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
991 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1000 diff = compare_ether_addr(eth_hdr(skb)->h_dest, n->ha) != 0;
1002 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1004 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1008 } else if (vxlan->flags & VXLAN_F_L3MISS)
1009 vxlan_ip_miss(dev, pip->daddr);
1013 static void vxlan_sock_put(struct sk_buff *skb)
1018 /* On transmit, associate with the tunnel socket */
1019 static void vxlan_set_owner(struct net_device *dev, struct sk_buff *skb)
1021 struct vxlan_dev *vxlan = netdev_priv(dev);
1022 struct sock *sk = vxlan->vn_sock->sock->sk;
1027 skb->destructor = vxlan_sock_put;
1030 /* Compute source port for outgoing packet
1031 * first choice to use L4 flow hash since it will spread
1032 * better and maybe available from hardware
1033 * secondary choice is to use jhash on the Ethernet header
1035 static __be16 vxlan_src_port(const struct vxlan_dev *vxlan, struct sk_buff *skb)
1037 unsigned int range = (vxlan->port_max - vxlan->port_min) + 1;
1040 hash = skb_get_rxhash(skb);
1042 hash = jhash(skb->data, 2 * ETH_ALEN,
1043 (__force u32) skb->protocol);
1045 return htons((((u64) hash * range) >> 32) + vxlan->port_min);
1048 static int handle_offloads(struct sk_buff *skb)
1050 if (skb_is_gso(skb)) {
1051 int err = skb_unclone(skb, GFP_ATOMIC);
1055 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
1056 } else if (skb->ip_summed != CHECKSUM_PARTIAL)
1057 skb->ip_summed = CHECKSUM_NONE;
1062 /* Bypass encapsulation if the destination is local */
1063 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1064 struct vxlan_dev *dst_vxlan)
1066 struct pcpu_tstats *tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1067 struct pcpu_tstats *rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1069 skb->pkt_type = PACKET_HOST;
1070 skb->encapsulation = 0;
1071 skb->dev = dst_vxlan->dev;
1072 __skb_pull(skb, skb_network_offset(skb));
1074 if (dst_vxlan->flags & VXLAN_F_LEARN)
1075 vxlan_snoop(skb->dev, htonl(INADDR_LOOPBACK),
1076 eth_hdr(skb)->h_source);
1078 u64_stats_update_begin(&tx_stats->syncp);
1079 tx_stats->tx_packets++;
1080 tx_stats->tx_bytes += skb->len;
1081 u64_stats_update_end(&tx_stats->syncp);
1083 if (netif_rx(skb) == NET_RX_SUCCESS) {
1084 u64_stats_update_begin(&rx_stats->syncp);
1085 rx_stats->rx_packets++;
1086 rx_stats->rx_bytes += skb->len;
1087 u64_stats_update_end(&rx_stats->syncp);
1089 skb->dev->stats.rx_dropped++;
1093 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1094 struct vxlan_rdst *rdst, bool did_rsc)
1096 struct vxlan_dev *vxlan = netdev_priv(dev);
1098 const struct iphdr *old_iph;
1099 struct vxlanhdr *vxh;
1103 __be16 src_port, dst_port;
1109 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1110 vni = rdst->remote_vni;
1111 dst = rdst->remote_ip;
1115 /* short-circuited back to local bridge */
1116 vxlan_encap_bypass(skb, vxlan, vxlan);
1122 if (!skb->encapsulation) {
1123 skb_reset_inner_headers(skb);
1124 skb->encapsulation = 1;
1127 /* Need space for new headers (invalidates iph ptr) */
1128 if (skb_cow_head(skb, VXLAN_HEADROOM))
1131 old_iph = ip_hdr(skb);
1134 if (!ttl && IN_MULTICAST(ntohl(dst)))
1139 tos = ip_tunnel_get_dsfield(old_iph, skb);
1141 src_port = vxlan_src_port(vxlan, skb);
1143 memset(&fl4, 0, sizeof(fl4));
1144 fl4.flowi4_oif = rdst->remote_ifindex;
1145 fl4.flowi4_tos = RT_TOS(tos);
1147 fl4.saddr = vxlan->saddr;
1149 rt = ip_route_output_key(dev_net(dev), &fl4);
1151 netdev_dbg(dev, "no route to %pI4\n", &dst);
1152 dev->stats.tx_carrier_errors++;
1156 if (rt->dst.dev == dev) {
1157 netdev_dbg(dev, "circular route to %pI4\n", &dst);
1159 dev->stats.collisions++;
1163 /* Bypass encapsulation if the destination is local */
1164 if (rt->rt_flags & RTCF_LOCAL &&
1165 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1166 struct vxlan_dev *dst_vxlan;
1169 dst_vxlan = vxlan_find_vni(dev_net(dev), vni, dst_port);
1172 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1175 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1176 vxh->vx_flags = htonl(VXLAN_FLAGS);
1177 vxh->vx_vni = htonl(vni << 8);
1179 __skb_push(skb, sizeof(*uh));
1180 skb_reset_transport_header(skb);
1183 uh->dest = dst_port;
1184 uh->source = src_port;
1186 uh->len = htons(skb->len);
1189 vxlan_set_owner(dev, skb);
1191 if (handle_offloads(skb))
1194 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1195 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1197 err = iptunnel_xmit(dev_net(dev), rt, skb, fl4.saddr, dst,
1198 IPPROTO_UDP, tos, ttl, df);
1199 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1204 dev->stats.tx_dropped++;
1208 dev->stats.tx_errors++;
1213 /* Transmit local packets over Vxlan
1215 * Outer IP header inherits ECN and DF from inner header.
1216 * Outer UDP destination is the VXLAN assigned port.
1217 * source port is based on hash of flow
1219 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1221 struct vxlan_dev *vxlan = netdev_priv(dev);
1223 bool did_rsc = false;
1224 struct vxlan_rdst *rdst;
1225 struct vxlan_fdb *f;
1227 skb_reset_mac_header(skb);
1230 if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP)
1231 return arp_reduce(dev, skb);
1233 f = vxlan_find_mac(vxlan, eth->h_dest);
1236 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1237 ntohs(eth->h_proto) == ETH_P_IP) {
1238 did_rsc = route_shortcircuit(dev, skb);
1240 f = vxlan_find_mac(vxlan, eth->h_dest);
1244 f = vxlan_find_mac(vxlan, all_zeros_mac);
1246 if ((vxlan->flags & VXLAN_F_L2MISS) &&
1247 !is_multicast_ether_addr(eth->h_dest))
1248 vxlan_fdb_miss(vxlan, eth->h_dest);
1250 dev->stats.tx_dropped++;
1252 return NETDEV_TX_OK;
1256 list_for_each_entry_rcu(rdst, &f->remotes, list) {
1257 struct sk_buff *skb1;
1259 skb1 = skb_clone(skb, GFP_ATOMIC);
1261 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1265 return NETDEV_TX_OK;
1268 /* Walk the forwarding table and purge stale entries */
1269 static void vxlan_cleanup(unsigned long arg)
1271 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1272 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1275 if (!netif_running(vxlan->dev))
1278 spin_lock_bh(&vxlan->hash_lock);
1279 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1280 struct hlist_node *p, *n;
1281 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1283 = container_of(p, struct vxlan_fdb, hlist);
1284 unsigned long timeout;
1286 if (f->state & NUD_PERMANENT)
1289 timeout = f->used + vxlan->age_interval * HZ;
1290 if (time_before_eq(timeout, jiffies)) {
1291 netdev_dbg(vxlan->dev,
1292 "garbage collect %pM\n",
1294 f->state = NUD_STALE;
1295 vxlan_fdb_destroy(vxlan, f);
1296 } else if (time_before(timeout, next_timer))
1297 next_timer = timeout;
1300 spin_unlock_bh(&vxlan->hash_lock);
1302 mod_timer(&vxlan->age_timer, next_timer);
1305 /* Setup stats when device is created */
1306 static int vxlan_init(struct net_device *dev)
1308 struct vxlan_dev *vxlan = netdev_priv(dev);
1309 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
1310 struct vxlan_sock *vs;
1311 __u32 vni = vxlan->default_dst.remote_vni;
1313 dev->tstats = alloc_percpu(struct pcpu_tstats);
1317 spin_lock(&vn->sock_lock);
1318 vs = vxlan_find_port(dev_net(dev), vxlan->dst_port);
1320 /* If we have a socket with same port already, reuse it */
1321 atomic_inc(&vs->refcnt);
1322 vxlan->vn_sock = vs;
1323 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
1325 /* otherwise make new socket outside of RTNL */
1327 queue_work(vxlan_wq, &vxlan->sock_work);
1329 spin_unlock(&vn->sock_lock);
1334 static void vxlan_fdb_delete_defualt(struct vxlan_dev *vxlan)
1336 struct vxlan_fdb *f;
1338 spin_lock_bh(&vxlan->hash_lock);
1339 f = __vxlan_find_mac(vxlan, all_zeros_mac);
1341 vxlan_fdb_destroy(vxlan, f);
1342 spin_unlock_bh(&vxlan->hash_lock);
1345 static void vxlan_uninit(struct net_device *dev)
1347 struct vxlan_dev *vxlan = netdev_priv(dev);
1348 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
1349 struct vxlan_sock *vs = vxlan->vn_sock;
1351 vxlan_fdb_delete_defualt(vxlan);
1354 vxlan_sock_release(vn, vs);
1355 free_percpu(dev->tstats);
1358 /* Start ageing timer and join group when device is brought up */
1359 static int vxlan_open(struct net_device *dev)
1361 struct vxlan_dev *vxlan = netdev_priv(dev);
1362 struct vxlan_sock *vs = vxlan->vn_sock;
1364 /* socket hasn't been created */
1368 if (IN_MULTICAST(ntohl(vxlan->default_dst.remote_ip))) {
1369 vxlan_sock_hold(vs);
1371 queue_work(vxlan_wq, &vxlan->igmp_work);
1374 if (vxlan->age_interval)
1375 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
1380 /* Purge the forwarding table */
1381 static void vxlan_flush(struct vxlan_dev *vxlan)
1385 spin_lock_bh(&vxlan->hash_lock);
1386 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1387 struct hlist_node *p, *n;
1388 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1390 = container_of(p, struct vxlan_fdb, hlist);
1391 /* the all_zeros_mac entry is deleted at vxlan_uninit */
1392 if (!is_zero_ether_addr(f->eth_addr))
1393 vxlan_fdb_destroy(vxlan, f);
1396 spin_unlock_bh(&vxlan->hash_lock);
1399 /* Cleanup timer and forwarding table on shutdown */
1400 static int vxlan_stop(struct net_device *dev)
1402 struct vxlan_dev *vxlan = netdev_priv(dev);
1403 struct vxlan_sock *vs = vxlan->vn_sock;
1405 if (vs && IN_MULTICAST(ntohl(vxlan->default_dst.remote_ip))) {
1406 vxlan_sock_hold(vs);
1408 queue_work(vxlan_wq, &vxlan->igmp_work);
1411 del_timer_sync(&vxlan->age_timer);
1418 /* Stub, nothing needs to be done. */
1419 static void vxlan_set_multicast_list(struct net_device *dev)
1423 static const struct net_device_ops vxlan_netdev_ops = {
1424 .ndo_init = vxlan_init,
1425 .ndo_uninit = vxlan_uninit,
1426 .ndo_open = vxlan_open,
1427 .ndo_stop = vxlan_stop,
1428 .ndo_start_xmit = vxlan_xmit,
1429 .ndo_get_stats64 = ip_tunnel_get_stats64,
1430 .ndo_set_rx_mode = vxlan_set_multicast_list,
1431 .ndo_change_mtu = eth_change_mtu,
1432 .ndo_validate_addr = eth_validate_addr,
1433 .ndo_set_mac_address = eth_mac_addr,
1434 .ndo_fdb_add = vxlan_fdb_add,
1435 .ndo_fdb_del = vxlan_fdb_delete,
1436 .ndo_fdb_dump = vxlan_fdb_dump,
1439 /* Info for udev, that this is a virtual tunnel endpoint */
1440 static struct device_type vxlan_type = {
1444 /* Initialize the device structure. */
1445 static void vxlan_setup(struct net_device *dev)
1447 struct vxlan_dev *vxlan = netdev_priv(dev);
1451 eth_hw_addr_random(dev);
1453 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
1455 dev->netdev_ops = &vxlan_netdev_ops;
1456 dev->destructor = free_netdev;
1457 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
1459 dev->tx_queue_len = 0;
1460 dev->features |= NETIF_F_LLTX;
1461 dev->features |= NETIF_F_NETNS_LOCAL;
1462 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1463 dev->features |= NETIF_F_RXCSUM;
1464 dev->features |= NETIF_F_GSO_SOFTWARE;
1466 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1467 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1468 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1469 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1471 INIT_LIST_HEAD(&vxlan->next);
1472 spin_lock_init(&vxlan->hash_lock);
1473 INIT_WORK(&vxlan->igmp_work, vxlan_igmp_work);
1474 INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
1476 init_timer_deferrable(&vxlan->age_timer);
1477 vxlan->age_timer.function = vxlan_cleanup;
1478 vxlan->age_timer.data = (unsigned long) vxlan;
1480 inet_get_local_port_range(&low, &high);
1481 vxlan->port_min = low;
1482 vxlan->port_max = high;
1483 vxlan->dst_port = htons(vxlan_port);
1487 for (h = 0; h < FDB_HASH_SIZE; ++h)
1488 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
1491 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
1492 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
1493 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1494 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
1495 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1496 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
1497 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
1498 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
1499 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
1500 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
1501 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
1502 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
1503 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
1504 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
1505 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
1506 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
1509 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
1511 if (tb[IFLA_ADDRESS]) {
1512 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1513 pr_debug("invalid link address (not ethernet)\n");
1517 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1518 pr_debug("invalid all zero ethernet address\n");
1519 return -EADDRNOTAVAIL;
1526 if (data[IFLA_VXLAN_ID]) {
1527 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
1528 if (id >= VXLAN_VID_MASK)
1532 if (data[IFLA_VXLAN_PORT_RANGE]) {
1533 const struct ifla_vxlan_port_range *p
1534 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1536 if (ntohs(p->high) < ntohs(p->low)) {
1537 pr_debug("port range %u .. %u not valid\n",
1538 ntohs(p->low), ntohs(p->high));
1546 static void vxlan_get_drvinfo(struct net_device *netdev,
1547 struct ethtool_drvinfo *drvinfo)
1549 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
1550 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
1553 static const struct ethtool_ops vxlan_ethtool_ops = {
1554 .get_drvinfo = vxlan_get_drvinfo,
1555 .get_link = ethtool_op_get_link,
1558 static void vxlan_del_work(struct work_struct *work)
1560 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
1562 sk_release_kernel(vs->sock->sk);
1566 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port)
1568 struct vxlan_sock *vs;
1570 struct sockaddr_in vxlan_addr = {
1571 .sin_family = AF_INET,
1572 .sin_addr.s_addr = htonl(INADDR_ANY),
1578 vs = kmalloc(sizeof(*vs), GFP_KERNEL);
1580 return ERR_PTR(-ENOMEM);
1582 for (h = 0; h < VNI_HASH_SIZE; ++h)
1583 INIT_HLIST_HEAD(&vs->vni_list[h]);
1585 INIT_WORK(&vs->del_work, vxlan_del_work);
1587 /* Create UDP socket for encapsulation receive. */
1588 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vs->sock);
1590 pr_debug("UDP socket create failed\n");
1595 /* Put in proper namespace */
1597 sk_change_net(sk, net);
1599 rc = kernel_bind(vs->sock, (struct sockaddr *) &vxlan_addr,
1600 sizeof(vxlan_addr));
1602 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1603 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
1604 sk_release_kernel(sk);
1609 /* Disable multicast loopback */
1610 inet_sk(sk)->mc_loop = 0;
1612 /* Mark socket as an encapsulation socket. */
1613 udp_sk(sk)->encap_type = 1;
1614 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1616 atomic_set(&vs->refcnt, 1);
1621 /* Scheduled at device creation to bind to a socket */
1622 static void vxlan_sock_work(struct work_struct *work)
1624 struct vxlan_dev *vxlan
1625 = container_of(work, struct vxlan_dev, sock_work);
1626 struct net_device *dev = vxlan->dev;
1627 struct net *net = dev_net(dev);
1628 __u32 vni = vxlan->default_dst.remote_vni;
1629 __be16 port = vxlan->dst_port;
1630 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1631 struct vxlan_sock *nvs, *ovs;
1633 nvs = vxlan_socket_create(net, port);
1635 netdev_err(vxlan->dev, "Can not create UDP socket, %ld\n",
1640 spin_lock(&vn->sock_lock);
1641 /* Look again to see if can reuse socket */
1642 ovs = vxlan_find_port(net, port);
1644 atomic_inc(&ovs->refcnt);
1645 vxlan->vn_sock = ovs;
1646 hlist_add_head_rcu(&vxlan->hlist, vni_head(ovs, vni));
1647 spin_unlock(&vn->sock_lock);
1649 sk_release_kernel(nvs->sock->sk);
1652 vxlan->vn_sock = nvs;
1653 hlist_add_head_rcu(&nvs->hlist, vs_head(net, port));
1654 hlist_add_head_rcu(&vxlan->hlist, vni_head(nvs, vni));
1655 spin_unlock(&vn->sock_lock);
1661 static int vxlan_newlink(struct net *net, struct net_device *dev,
1662 struct nlattr *tb[], struct nlattr *data[])
1664 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1665 struct vxlan_dev *vxlan = netdev_priv(dev);
1666 struct vxlan_rdst *dst = &vxlan->default_dst;
1670 if (!data[IFLA_VXLAN_ID])
1673 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1674 dst->remote_vni = vni;
1676 if (data[IFLA_VXLAN_GROUP])
1677 dst->remote_ip = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1679 if (data[IFLA_VXLAN_LOCAL])
1680 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1682 if (data[IFLA_VXLAN_LINK] &&
1683 (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
1684 struct net_device *lowerdev
1685 = __dev_get_by_index(net, dst->remote_ifindex);
1688 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
1693 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
1695 /* update header length based on lower device */
1696 dev->hard_header_len = lowerdev->hard_header_len +
1700 if (data[IFLA_VXLAN_TOS])
1701 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1703 if (data[IFLA_VXLAN_TTL])
1704 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
1706 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
1707 vxlan->flags |= VXLAN_F_LEARN;
1709 if (data[IFLA_VXLAN_AGEING])
1710 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1712 vxlan->age_interval = FDB_AGE_DEFAULT;
1714 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
1715 vxlan->flags |= VXLAN_F_PROXY;
1717 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
1718 vxlan->flags |= VXLAN_F_RSC;
1720 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
1721 vxlan->flags |= VXLAN_F_L2MISS;
1723 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
1724 vxlan->flags |= VXLAN_F_L3MISS;
1726 if (data[IFLA_VXLAN_LIMIT])
1727 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1729 if (data[IFLA_VXLAN_PORT_RANGE]) {
1730 const struct ifla_vxlan_port_range *p
1731 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1732 vxlan->port_min = ntohs(p->low);
1733 vxlan->port_max = ntohs(p->high);
1736 if (data[IFLA_VXLAN_PORT])
1737 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
1739 if (vxlan_find_vni(net, vni, vxlan->dst_port)) {
1740 pr_info("duplicate VNI %u\n", vni);
1744 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
1746 /* create an fdb entry for default destination */
1747 err = vxlan_fdb_create(vxlan, all_zeros_mac,
1748 vxlan->default_dst.remote_ip,
1749 NUD_REACHABLE|NUD_PERMANENT,
1750 NLM_F_EXCL|NLM_F_CREATE,
1751 vxlan->dst_port, vxlan->default_dst.remote_vni,
1752 vxlan->default_dst.remote_ifindex, NTF_SELF);
1756 err = register_netdevice(dev);
1758 vxlan_fdb_delete_defualt(vxlan);
1762 list_add(&vxlan->next, &vn->vxlan_list);
1767 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1769 struct vxlan_dev *vxlan = netdev_priv(dev);
1771 hlist_del_rcu(&vxlan->hlist);
1772 list_del(&vxlan->next);
1773 unregister_netdevice_queue(dev, head);
1776 static size_t vxlan_get_size(const struct net_device *dev)
1779 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1780 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1781 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1782 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1783 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1784 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1785 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
1786 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
1787 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
1788 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
1789 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
1790 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1791 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
1792 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
1793 nla_total_size(sizeof(__be16))+ /* IFLA_VXLAN_PORT */
1797 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1799 const struct vxlan_dev *vxlan = netdev_priv(dev);
1800 const struct vxlan_rdst *dst = &vxlan->default_dst;
1801 struct ifla_vxlan_port_range ports = {
1802 .low = htons(vxlan->port_min),
1803 .high = htons(vxlan->port_max),
1806 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
1807 goto nla_put_failure;
1809 if (dst->remote_ip && nla_put_be32(skb, IFLA_VXLAN_GROUP, dst->remote_ip))
1810 goto nla_put_failure;
1812 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
1813 goto nla_put_failure;
1815 if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
1816 goto nla_put_failure;
1818 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1819 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
1820 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
1821 !!(vxlan->flags & VXLAN_F_LEARN)) ||
1822 nla_put_u8(skb, IFLA_VXLAN_PROXY,
1823 !!(vxlan->flags & VXLAN_F_PROXY)) ||
1824 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
1825 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
1826 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
1827 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
1828 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
1829 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1830 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
1831 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port))
1832 goto nla_put_failure;
1834 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
1835 goto nla_put_failure;
1843 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1845 .maxtype = IFLA_VXLAN_MAX,
1846 .policy = vxlan_policy,
1847 .priv_size = sizeof(struct vxlan_dev),
1848 .setup = vxlan_setup,
1849 .validate = vxlan_validate,
1850 .newlink = vxlan_newlink,
1851 .dellink = vxlan_dellink,
1852 .get_size = vxlan_get_size,
1853 .fill_info = vxlan_fill_info,
1856 static __net_init int vxlan_init_net(struct net *net)
1858 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1861 INIT_LIST_HEAD(&vn->vxlan_list);
1862 spin_lock_init(&vn->sock_lock);
1864 for (h = 0; h < PORT_HASH_SIZE; ++h)
1865 INIT_HLIST_HEAD(&vn->sock_list[h]);
1870 static __net_exit void vxlan_exit_net(struct net *net)
1872 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1873 struct vxlan_dev *vxlan;
1876 list_for_each_entry(vxlan, &vn->vxlan_list, next)
1877 dev_close(vxlan->dev);
1881 static struct pernet_operations vxlan_net_ops = {
1882 .init = vxlan_init_net,
1883 .exit = vxlan_exit_net,
1884 .id = &vxlan_net_id,
1885 .size = sizeof(struct vxlan_net),
1888 static int __init vxlan_init_module(void)
1892 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
1896 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1898 rc = register_pernet_device(&vxlan_net_ops);
1902 rc = rtnl_link_register(&vxlan_link_ops);
1909 unregister_pernet_device(&vxlan_net_ops);
1911 destroy_workqueue(vxlan_wq);
1914 late_initcall(vxlan_init_module);
1916 static void __exit vxlan_cleanup_module(void)
1918 unregister_pernet_device(&vxlan_net_ops);
1919 rtnl_link_unregister(&vxlan_link_ops);
1920 destroy_workqueue(vxlan_wq);
1923 module_exit(vxlan_cleanup_module);
1925 MODULE_LICENSE("GPL");
1926 MODULE_VERSION(VXLAN_VERSION);
1927 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
1928 MODULE_ALIAS_RTNL_LINK("vxlan");