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.
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/module.h>
16 #include <linux/errno.h>
17 #include <linux/slab.h>
18 #include <linux/skbuff.h>
19 #include <linux/rculist.h>
20 #include <linux/netdevice.h>
23 #include <linux/udp.h>
24 #include <linux/igmp.h>
25 #include <linux/etherdevice.h>
26 #include <linux/if_ether.h>
27 #include <linux/if_vlan.h>
28 #include <linux/hash.h>
29 #include <linux/ethtool.h>
31 #include <net/ndisc.h>
33 #include <net/ip_tunnels.h>
36 #include <net/rtnetlink.h>
37 #include <net/route.h>
38 #include <net/dsfield.h>
39 #include <net/inet_ecn.h>
40 #include <net/net_namespace.h>
41 #include <net/netns/generic.h>
42 #include <net/vxlan.h>
43 #if IS_ENABLED(CONFIG_IPV6)
45 #include <net/addrconf.h>
46 #include <net/ip6_tunnel.h>
47 #include <net/ip6_checksum.h>
50 #define VXLAN_VERSION "0.1"
52 #define PORT_HASH_BITS 8
53 #define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
54 #define VNI_HASH_BITS 10
55 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
56 #define FDB_HASH_BITS 8
57 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
58 #define FDB_AGE_DEFAULT 300 /* 5 min */
59 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
61 #define VXLAN_N_VID (1u << 24)
62 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
63 /* IP header + UDP + VXLAN + Ethernet header */
64 #define VXLAN_HEADROOM (20 + 8 + 8 + 14)
65 /* IPv6 header + UDP + VXLAN + Ethernet header */
66 #define VXLAN6_HEADROOM (40 + 8 + 8 + 14)
67 #define VXLAN_HLEN (sizeof(struct udphdr) + sizeof(struct vxlanhdr))
69 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
71 /* VXLAN protocol header */
77 /* UDP port for VXLAN traffic.
78 * The IANA assigned port is 4789, but the Linux default is 8472
79 * for compatibility with early adopters.
81 static unsigned short vxlan_port __read_mostly = 8472;
82 module_param_named(udp_port, vxlan_port, ushort, 0444);
83 MODULE_PARM_DESC(udp_port, "Destination UDP port");
85 static bool log_ecn_error = true;
86 module_param(log_ecn_error, bool, 0644);
87 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
89 static int vxlan_net_id;
91 static const u8 all_zeros_mac[ETH_ALEN];
93 /* per-network namespace private data for this module */
95 struct list_head vxlan_list;
96 struct hlist_head sock_list[PORT_HASH_SIZE];
101 struct sockaddr_in sin;
102 struct sockaddr_in6 sin6;
107 union vxlan_addr remote_ip;
111 struct list_head list;
115 /* Forwarding table entry */
117 struct hlist_node hlist; /* linked list of entries */
119 unsigned long updated; /* jiffies */
121 struct list_head remotes;
122 u16 state; /* see ndm_state */
123 u8 flags; /* see ndm_flags */
124 u8 eth_addr[ETH_ALEN];
127 /* Pseudo network device */
129 struct hlist_node hlist; /* vni hash table */
130 struct list_head next; /* vxlan's per namespace list */
131 struct vxlan_sock *vn_sock; /* listening socket */
132 struct net_device *dev;
133 struct vxlan_rdst default_dst; /* default destination */
134 union vxlan_addr saddr; /* source address */
136 __u16 port_min; /* source port range */
138 __u8 tos; /* TOS override */
140 u32 flags; /* VXLAN_F_* below */
142 struct work_struct sock_work;
143 struct work_struct igmp_join;
144 struct work_struct igmp_leave;
146 unsigned long age_interval;
147 struct timer_list age_timer;
148 spinlock_t hash_lock;
149 unsigned int addrcnt;
150 unsigned int addrmax;
152 struct hlist_head fdb_head[FDB_HASH_SIZE];
155 #define VXLAN_F_LEARN 0x01
156 #define VXLAN_F_PROXY 0x02
157 #define VXLAN_F_RSC 0x04
158 #define VXLAN_F_L2MISS 0x08
159 #define VXLAN_F_L3MISS 0x10
160 #define VXLAN_F_IPV6 0x20 /* internal flag */
162 /* salt for hash table */
163 static u32 vxlan_salt __read_mostly;
164 static struct workqueue_struct *vxlan_wq;
166 static void vxlan_sock_work(struct work_struct *work);
168 #if IS_ENABLED(CONFIG_IPV6)
170 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
172 if (a->sa.sa_family != b->sa.sa_family)
174 if (a->sa.sa_family == AF_INET6)
175 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
177 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
180 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
182 if (ipa->sa.sa_family == AF_INET6)
183 return ipv6_addr_any(&ipa->sin6.sin6_addr);
185 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
188 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
190 if (ipa->sa.sa_family == AF_INET6)
191 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
193 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
196 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
198 if (nla_len(nla) >= sizeof(struct in6_addr)) {
199 nla_memcpy(&ip->sin6.sin6_addr, nla, sizeof(struct in6_addr));
200 ip->sa.sa_family = AF_INET6;
202 } else if (nla_len(nla) >= sizeof(__be32)) {
203 ip->sin.sin_addr.s_addr = nla_get_be32(nla);
204 ip->sa.sa_family = AF_INET;
207 return -EAFNOSUPPORT;
211 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
212 const union vxlan_addr *ip)
214 if (ip->sa.sa_family == AF_INET6)
215 return nla_put(skb, attr, sizeof(struct in6_addr), &ip->sin6.sin6_addr);
217 return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
220 #else /* !CONFIG_IPV6 */
223 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
225 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
228 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
230 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
233 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
235 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
238 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
240 if (nla_len(nla) >= sizeof(struct in6_addr)) {
241 return -EAFNOSUPPORT;
242 } else if (nla_len(nla) >= sizeof(__be32)) {
243 ip->sin.sin_addr.s_addr = nla_get_be32(nla);
244 ip->sa.sa_family = AF_INET;
247 return -EAFNOSUPPORT;
251 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
252 const union vxlan_addr *ip)
254 return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
258 /* Virtual Network hash table head */
259 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
261 return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
264 /* Socket hash table head */
265 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
267 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
269 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
272 /* First remote destination for a forwarding entry.
273 * Guaranteed to be non-NULL because remotes are never deleted.
275 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
277 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
280 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
282 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
285 /* Find VXLAN socket based on network namespace and UDP port */
286 static struct vxlan_sock *vxlan_find_sock(struct net *net, __be16 port)
288 struct vxlan_sock *vs;
290 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
291 if (inet_sk(vs->sock->sk)->inet_sport == port)
297 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
299 struct vxlan_dev *vxlan;
301 hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
302 if (vxlan->default_dst.remote_vni == id)
309 /* Look up VNI in a per net namespace table */
310 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id, __be16 port)
312 struct vxlan_sock *vs;
314 vs = vxlan_find_sock(net, port);
318 return vxlan_vs_find_vni(vs, id);
321 /* Fill in neighbour message in skbuff. */
322 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
323 const struct vxlan_fdb *fdb,
324 u32 portid, u32 seq, int type, unsigned int flags,
325 const struct vxlan_rdst *rdst)
327 unsigned long now = jiffies;
328 struct nda_cacheinfo ci;
329 struct nlmsghdr *nlh;
331 bool send_ip, send_eth;
333 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
337 ndm = nlmsg_data(nlh);
338 memset(ndm, 0, sizeof(*ndm));
340 send_eth = send_ip = true;
342 if (type == RTM_GETNEIGH) {
343 ndm->ndm_family = AF_INET;
344 send_ip = !vxlan_addr_any(&rdst->remote_ip);
345 send_eth = !is_zero_ether_addr(fdb->eth_addr);
347 ndm->ndm_family = AF_BRIDGE;
348 ndm->ndm_state = fdb->state;
349 ndm->ndm_ifindex = vxlan->dev->ifindex;
350 ndm->ndm_flags = fdb->flags;
351 ndm->ndm_type = NDA_DST;
353 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
354 goto nla_put_failure;
356 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
357 goto nla_put_failure;
359 if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
360 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
361 goto nla_put_failure;
362 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
363 nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
364 goto nla_put_failure;
365 if (rdst->remote_ifindex &&
366 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
367 goto nla_put_failure;
369 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
370 ci.ndm_confirmed = 0;
371 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
374 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
375 goto nla_put_failure;
377 return nlmsg_end(skb, nlh);
380 nlmsg_cancel(skb, nlh);
384 static inline size_t vxlan_nlmsg_size(void)
386 return NLMSG_ALIGN(sizeof(struct ndmsg))
387 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
388 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
389 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
390 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
391 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
392 + nla_total_size(sizeof(struct nda_cacheinfo));
395 static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
396 struct vxlan_fdb *fdb, int type)
398 struct net *net = dev_net(vxlan->dev);
402 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
406 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0,
407 first_remote_rtnl(fdb));
409 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
410 WARN_ON(err == -EMSGSIZE);
415 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
419 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
422 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
424 struct vxlan_dev *vxlan = netdev_priv(dev);
425 struct vxlan_fdb f = {
428 struct vxlan_rdst remote = {
429 .remote_ip = *ipa, /* goes to NDA_DST */
430 .remote_vni = VXLAN_N_VID,
433 INIT_LIST_HEAD(&f.remotes);
434 list_add_rcu(&remote.list, &f.remotes);
436 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
439 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
441 struct vxlan_fdb f = {
445 INIT_LIST_HEAD(&f.remotes);
446 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
448 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
451 /* Hash Ethernet address */
452 static u32 eth_hash(const unsigned char *addr)
454 u64 value = get_unaligned((u64 *)addr);
456 /* only want 6 bytes */
462 return hash_64(value, FDB_HASH_BITS);
465 /* Hash chain to use given mac address */
466 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
469 return &vxlan->fdb_head[eth_hash(mac)];
472 /* Look up Ethernet address in forwarding table */
473 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
477 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
480 hlist_for_each_entry_rcu(f, head, hlist) {
481 if (ether_addr_equal(mac, f->eth_addr))
488 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
493 f = __vxlan_find_mac(vxlan, mac);
500 /* caller should hold vxlan->hash_lock */
501 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
502 union vxlan_addr *ip, __be16 port,
503 __u32 vni, __u32 ifindex)
505 struct vxlan_rdst *rd;
507 list_for_each_entry(rd, &f->remotes, list) {
508 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
509 rd->remote_port == port &&
510 rd->remote_vni == vni &&
511 rd->remote_ifindex == ifindex)
518 /* Replace destination of unicast mac */
519 static int vxlan_fdb_replace(struct vxlan_fdb *f,
520 union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
522 struct vxlan_rdst *rd;
524 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
528 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
532 rd->remote_port = port;
533 rd->remote_vni = vni;
534 rd->remote_ifindex = ifindex;
538 /* Add/update destinations for multicast */
539 static int vxlan_fdb_append(struct vxlan_fdb *f,
540 union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
542 struct vxlan_rdst *rd;
544 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
548 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
552 rd->remote_port = port;
553 rd->remote_vni = vni;
554 rd->remote_ifindex = ifindex;
556 list_add_tail_rcu(&rd->list, &f->remotes);
561 /* Notify netdevs that UDP port started listening */
562 static void vxlan_notify_add_rx_port(struct sock *sk)
564 struct net_device *dev;
565 struct net *net = sock_net(sk);
566 sa_family_t sa_family = sk->sk_family;
567 u16 port = htons(inet_sk(sk)->inet_sport);
570 for_each_netdev_rcu(net, dev) {
571 if (dev->netdev_ops->ndo_add_vxlan_port)
572 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
578 /* Notify netdevs that UDP port is no more listening */
579 static void vxlan_notify_del_rx_port(struct sock *sk)
581 struct net_device *dev;
582 struct net *net = sock_net(sk);
583 sa_family_t sa_family = sk->sk_family;
584 u16 port = htons(inet_sk(sk)->inet_sport);
587 for_each_netdev_rcu(net, dev) {
588 if (dev->netdev_ops->ndo_del_vxlan_port)
589 dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
595 /* Add new entry to forwarding table -- assumes lock held */
596 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
597 const u8 *mac, union vxlan_addr *ip,
598 __u16 state, __u16 flags,
599 __be16 port, __u32 vni, __u32 ifindex,
605 f = __vxlan_find_mac(vxlan, mac);
607 if (flags & NLM_F_EXCL) {
608 netdev_dbg(vxlan->dev,
609 "lost race to create %pM\n", mac);
612 if (f->state != state) {
614 f->updated = jiffies;
617 if (f->flags != ndm_flags) {
618 f->flags = ndm_flags;
619 f->updated = jiffies;
622 if ((flags & NLM_F_REPLACE)) {
623 /* Only change unicasts */
624 if (!(is_multicast_ether_addr(f->eth_addr) ||
625 is_zero_ether_addr(f->eth_addr))) {
626 int rc = vxlan_fdb_replace(f, ip, port, vni,
635 if ((flags & NLM_F_APPEND) &&
636 (is_multicast_ether_addr(f->eth_addr) ||
637 is_zero_ether_addr(f->eth_addr))) {
638 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex);
645 if (!(flags & NLM_F_CREATE))
648 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
651 /* Disallow replace to add a multicast entry */
652 if ((flags & NLM_F_REPLACE) &&
653 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
656 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
657 f = kmalloc(sizeof(*f), GFP_ATOMIC);
663 f->flags = ndm_flags;
664 f->updated = f->used = jiffies;
665 INIT_LIST_HEAD(&f->remotes);
666 memcpy(f->eth_addr, mac, ETH_ALEN);
668 vxlan_fdb_append(f, ip, port, vni, ifindex);
671 hlist_add_head_rcu(&f->hlist,
672 vxlan_fdb_head(vxlan, mac));
676 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
681 static void vxlan_fdb_free(struct rcu_head *head)
683 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
684 struct vxlan_rdst *rd, *nd;
686 list_for_each_entry_safe(rd, nd, &f->remotes, list)
691 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
693 netdev_dbg(vxlan->dev,
694 "delete %pM\n", f->eth_addr);
697 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
699 hlist_del_rcu(&f->hlist);
700 call_rcu(&f->rcu, vxlan_fdb_free);
703 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
704 union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
706 struct net *net = dev_net(vxlan->dev);
710 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
714 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
715 if (remote->sa.sa_family == AF_INET) {
716 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
717 ip->sa.sa_family = AF_INET;
718 #if IS_ENABLED(CONFIG_IPV6)
720 ip->sin6.sin6_addr = in6addr_any;
721 ip->sa.sa_family = AF_INET6;
727 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
729 *port = nla_get_be16(tb[NDA_PORT]);
731 *port = vxlan->dst_port;
735 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
737 *vni = nla_get_u32(tb[NDA_VNI]);
739 *vni = vxlan->default_dst.remote_vni;
742 if (tb[NDA_IFINDEX]) {
743 struct net_device *tdev;
745 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
747 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
748 tdev = dev_get_by_index(net, *ifindex);
750 return -EADDRNOTAVAIL;
759 /* Add static entry (via netlink) */
760 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
761 struct net_device *dev,
762 const unsigned char *addr, u16 flags)
764 struct vxlan_dev *vxlan = netdev_priv(dev);
765 /* struct net *net = dev_net(vxlan->dev); */
771 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
772 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
777 if (tb[NDA_DST] == NULL)
780 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
784 spin_lock_bh(&vxlan->hash_lock);
785 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
786 port, vni, ifindex, ndm->ndm_flags);
787 spin_unlock_bh(&vxlan->hash_lock);
792 /* Delete entry (via netlink) */
793 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
794 struct net_device *dev,
795 const unsigned char *addr)
797 struct vxlan_dev *vxlan = netdev_priv(dev);
799 struct vxlan_rdst *rd = NULL;
805 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
811 spin_lock_bh(&vxlan->hash_lock);
812 f = vxlan_find_mac(vxlan, addr);
816 if (!vxlan_addr_any(&ip)) {
817 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
824 /* remove a destination if it's not the only one on the list,
825 * otherwise destroy the fdb entry
827 if (rd && !list_is_singular(&f->remotes)) {
828 list_del_rcu(&rd->list);
833 vxlan_fdb_destroy(vxlan, f);
836 spin_unlock_bh(&vxlan->hash_lock);
841 /* Dump forwarding table */
842 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
843 struct net_device *dev, int idx)
845 struct vxlan_dev *vxlan = netdev_priv(dev);
848 for (h = 0; h < FDB_HASH_SIZE; ++h) {
852 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
853 struct vxlan_rdst *rd;
855 if (idx < cb->args[0])
858 list_for_each_entry_rcu(rd, &f->remotes, list) {
859 err = vxlan_fdb_info(skb, vxlan, f,
860 NETLINK_CB(cb->skb).portid,
875 /* Watch incoming packets to learn mapping between Ethernet address
876 * and Tunnel endpoint.
877 * Return true if packet is bogus and should be droppped.
879 static bool vxlan_snoop(struct net_device *dev,
880 union vxlan_addr *src_ip, const u8 *src_mac)
882 struct vxlan_dev *vxlan = netdev_priv(dev);
885 f = vxlan_find_mac(vxlan, src_mac);
887 struct vxlan_rdst *rdst = first_remote_rcu(f);
889 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
892 /* Don't migrate static entries, drop packets */
893 if (f->state & NUD_NOARP)
898 "%pM migrated from %pIS to %pIS\n",
899 src_mac, &rdst->remote_ip, &src_ip);
901 rdst->remote_ip = *src_ip;
902 f->updated = jiffies;
903 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
905 /* learned new entry */
906 spin_lock(&vxlan->hash_lock);
908 /* close off race between vxlan_flush and incoming packets */
909 if (netif_running(dev))
910 vxlan_fdb_create(vxlan, src_mac, src_ip,
912 NLM_F_EXCL|NLM_F_CREATE,
914 vxlan->default_dst.remote_vni,
916 spin_unlock(&vxlan->hash_lock);
922 /* See if multicast group is already in use by other ID */
923 static bool vxlan_group_used(struct vxlan_net *vn, union vxlan_addr *remote_ip)
925 struct vxlan_dev *vxlan;
927 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
928 if (!netif_running(vxlan->dev))
931 if (vxlan_addr_equal(&vxlan->default_dst.remote_ip,
939 static void vxlan_sock_hold(struct vxlan_sock *vs)
941 atomic_inc(&vs->refcnt);
944 void vxlan_sock_release(struct vxlan_sock *vs)
946 struct sock *sk = vs->sock->sk;
947 struct net *net = sock_net(sk);
948 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
950 if (!atomic_dec_and_test(&vs->refcnt))
953 spin_lock(&vn->sock_lock);
954 hlist_del_rcu(&vs->hlist);
956 vs->sock->sk->sk_user_data = NULL;
957 vxlan_notify_del_rx_port(sk);
958 spin_unlock(&vn->sock_lock);
960 queue_work(vxlan_wq, &vs->del_work);
962 EXPORT_SYMBOL_GPL(vxlan_sock_release);
964 /* Callback to update multicast group membership when first VNI on
965 * multicast asddress is brought up
966 * Done as workqueue because ip_mc_join_group acquires RTNL.
968 static void vxlan_igmp_join(struct work_struct *work)
970 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join);
971 struct vxlan_sock *vs = vxlan->vn_sock;
972 struct sock *sk = vs->sock->sk;
973 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
974 int ifindex = vxlan->default_dst.remote_ifindex;
977 if (ip->sa.sa_family == AF_INET) {
978 struct ip_mreqn mreq = {
979 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
980 .imr_ifindex = ifindex,
983 ip_mc_join_group(sk, &mreq);
984 #if IS_ENABLED(CONFIG_IPV6)
986 ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
987 &ip->sin6.sin6_addr);
992 vxlan_sock_release(vs);
996 /* Inverse of vxlan_igmp_join when last VNI is brought down */
997 static void vxlan_igmp_leave(struct work_struct *work)
999 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave);
1000 struct vxlan_sock *vs = vxlan->vn_sock;
1001 struct sock *sk = vs->sock->sk;
1002 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1003 int ifindex = vxlan->default_dst.remote_ifindex;
1006 if (ip->sa.sa_family == AF_INET) {
1007 struct ip_mreqn mreq = {
1008 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1009 .imr_ifindex = ifindex,
1012 ip_mc_leave_group(sk, &mreq);
1013 #if IS_ENABLED(CONFIG_IPV6)
1015 ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1016 &ip->sin6.sin6_addr);
1022 vxlan_sock_release(vs);
1023 dev_put(vxlan->dev);
1026 /* Callback from net/ipv4/udp.c to receive packets */
1027 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1029 struct vxlan_sock *vs;
1030 struct vxlanhdr *vxh;
1033 /* Need Vxlan and inner Ethernet header to be present */
1034 if (!pskb_may_pull(skb, VXLAN_HLEN))
1037 /* Return packets with reserved bits set */
1038 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1039 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
1040 (vxh->vx_vni & htonl(0xff))) {
1041 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1042 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1046 if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1049 port = inet_sk(sk)->inet_sport;
1051 smp_read_barrier_depends();
1052 vs = (struct vxlan_sock *)sk->sk_user_data;
1056 vs->rcv(vs, skb, vxh->vx_vni);
1060 /* Consume bad packet */
1065 /* Return non vxlan pkt */
1069 static void vxlan_rcv(struct vxlan_sock *vs,
1070 struct sk_buff *skb, __be32 vx_vni)
1072 struct iphdr *oip = NULL;
1073 struct ipv6hdr *oip6 = NULL;
1074 struct vxlan_dev *vxlan;
1075 struct pcpu_tstats *stats;
1076 union vxlan_addr saddr;
1079 union vxlan_addr *remote_ip;
1081 vni = ntohl(vx_vni) >> 8;
1082 /* Is this VNI defined? */
1083 vxlan = vxlan_vs_find_vni(vs, vni);
1087 remote_ip = &vxlan->default_dst.remote_ip;
1088 skb_reset_mac_header(skb);
1089 skb->protocol = eth_type_trans(skb, vxlan->dev);
1091 /* Ignore packet loops (and multicast echo) */
1092 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1095 /* Re-examine inner Ethernet packet */
1096 if (remote_ip->sa.sa_family == AF_INET) {
1098 saddr.sin.sin_addr.s_addr = oip->saddr;
1099 saddr.sa.sa_family = AF_INET;
1100 #if IS_ENABLED(CONFIG_IPV6)
1102 oip6 = ipv6_hdr(skb);
1103 saddr.sin6.sin6_addr = oip6->saddr;
1104 saddr.sa.sa_family = AF_INET6;
1108 if ((vxlan->flags & VXLAN_F_LEARN) &&
1109 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1112 skb_reset_network_header(skb);
1114 /* If the NIC driver gave us an encapsulated packet with
1115 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
1116 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
1117 * for us. Otherwise force the upper layers to verify it.
1119 if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
1120 !(vxlan->dev->features & NETIF_F_RXCSUM))
1121 skb->ip_summed = CHECKSUM_NONE;
1123 skb->encapsulation = 0;
1126 err = IP6_ECN_decapsulate(oip6, skb);
1128 err = IP_ECN_decapsulate(oip, skb);
1130 if (unlikely(err)) {
1131 if (log_ecn_error) {
1133 net_info_ratelimited("non-ECT from %pI6\n",
1136 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1137 &oip->saddr, oip->tos);
1140 ++vxlan->dev->stats.rx_frame_errors;
1141 ++vxlan->dev->stats.rx_errors;
1146 stats = this_cpu_ptr(vxlan->dev->tstats);
1147 u64_stats_update_begin(&stats->syncp);
1148 stats->rx_packets++;
1149 stats->rx_bytes += skb->len;
1150 u64_stats_update_end(&stats->syncp);
1156 /* Consume bad packet */
1160 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1162 struct vxlan_dev *vxlan = netdev_priv(dev);
1163 struct arphdr *parp;
1166 struct neighbour *n;
1168 if (dev->flags & IFF_NOARP)
1171 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1172 dev->stats.tx_dropped++;
1175 parp = arp_hdr(skb);
1177 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1178 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1179 parp->ar_pro != htons(ETH_P_IP) ||
1180 parp->ar_op != htons(ARPOP_REQUEST) ||
1181 parp->ar_hln != dev->addr_len ||
1184 arpptr = (u8 *)parp + sizeof(struct arphdr);
1186 arpptr += dev->addr_len; /* sha */
1187 memcpy(&sip, arpptr, sizeof(sip));
1188 arpptr += sizeof(sip);
1189 arpptr += dev->addr_len; /* tha */
1190 memcpy(&tip, arpptr, sizeof(tip));
1192 if (ipv4_is_loopback(tip) ||
1193 ipv4_is_multicast(tip))
1196 n = neigh_lookup(&arp_tbl, &tip, dev);
1199 struct vxlan_fdb *f;
1200 struct sk_buff *reply;
1202 if (!(n->nud_state & NUD_CONNECTED)) {
1207 f = vxlan_find_mac(vxlan, n->ha);
1208 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1209 /* bridge-local neighbor */
1214 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1219 skb_reset_mac_header(reply);
1220 __skb_pull(reply, skb_network_offset(reply));
1221 reply->ip_summed = CHECKSUM_UNNECESSARY;
1222 reply->pkt_type = PACKET_HOST;
1224 if (netif_rx_ni(reply) == NET_RX_DROP)
1225 dev->stats.rx_dropped++;
1226 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1227 union vxlan_addr ipa = {
1228 .sin.sin_addr.s_addr = tip,
1229 .sa.sa_family = AF_INET,
1232 vxlan_ip_miss(dev, &ipa);
1236 return NETDEV_TX_OK;
1239 #if IS_ENABLED(CONFIG_IPV6)
1240 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1242 struct vxlan_dev *vxlan = netdev_priv(dev);
1243 struct neighbour *n;
1244 union vxlan_addr ipa;
1245 const struct ipv6hdr *iphdr;
1246 const struct in6_addr *saddr, *daddr;
1248 struct inet6_dev *in6_dev = NULL;
1250 in6_dev = __in6_dev_get(dev);
1254 if (!pskb_may_pull(skb, skb->len))
1257 iphdr = ipv6_hdr(skb);
1258 saddr = &iphdr->saddr;
1259 daddr = &iphdr->daddr;
1261 if (ipv6_addr_loopback(daddr) ||
1262 ipv6_addr_is_multicast(daddr))
1265 msg = (struct nd_msg *)skb_transport_header(skb);
1266 if (msg->icmph.icmp6_code != 0 ||
1267 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1270 n = neigh_lookup(ipv6_stub->nd_tbl, daddr, dev);
1273 struct vxlan_fdb *f;
1275 if (!(n->nud_state & NUD_CONNECTED)) {
1280 f = vxlan_find_mac(vxlan, n->ha);
1281 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1282 /* bridge-local neighbor */
1287 ipv6_stub->ndisc_send_na(dev, n, saddr, &msg->target,
1288 !!in6_dev->cnf.forwarding,
1289 true, false, false);
1291 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1292 ipa.sin6.sin6_addr = *daddr;
1293 ipa.sa.sa_family = AF_INET6;
1294 vxlan_ip_miss(dev, &ipa);
1299 return NETDEV_TX_OK;
1303 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1305 struct vxlan_dev *vxlan = netdev_priv(dev);
1306 struct neighbour *n;
1308 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1312 switch (ntohs(eth_hdr(skb)->h_proto)) {
1317 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1320 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1321 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1322 union vxlan_addr ipa = {
1323 .sin.sin_addr.s_addr = pip->daddr,
1324 .sa.sa_family = AF_INET,
1327 vxlan_ip_miss(dev, &ipa);
1333 #if IS_ENABLED(CONFIG_IPV6)
1336 struct ipv6hdr *pip6;
1338 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1340 pip6 = ipv6_hdr(skb);
1341 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1342 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1343 union vxlan_addr ipa = {
1344 .sin6.sin6_addr = pip6->daddr,
1345 .sa.sa_family = AF_INET6,
1348 vxlan_ip_miss(dev, &ipa);
1362 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1364 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1366 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1375 static void vxlan_sock_put(struct sk_buff *skb)
1380 /* On transmit, associate with the tunnel socket */
1381 static void vxlan_set_owner(struct sock *sk, struct sk_buff *skb)
1386 skb->destructor = vxlan_sock_put;
1389 /* Compute source port for outgoing packet
1390 * first choice to use L4 flow hash since it will spread
1391 * better and maybe available from hardware
1392 * secondary choice is to use jhash on the Ethernet header
1394 __be16 vxlan_src_port(__u16 port_min, __u16 port_max, struct sk_buff *skb)
1396 unsigned int range = (port_max - port_min) + 1;
1399 hash = skb_get_rxhash(skb);
1401 hash = jhash(skb->data, 2 * ETH_ALEN,
1402 (__force u32) skb->protocol);
1404 return htons((((u64) hash * range) >> 32) + port_min);
1406 EXPORT_SYMBOL_GPL(vxlan_src_port);
1408 static int handle_offloads(struct sk_buff *skb)
1410 if (skb_is_gso(skb)) {
1411 int err = skb_unclone(skb, GFP_ATOMIC);
1415 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
1416 } else if (skb->ip_summed != CHECKSUM_PARTIAL)
1417 skb->ip_summed = CHECKSUM_NONE;
1422 #if IS_ENABLED(CONFIG_IPV6)
1423 static int vxlan6_xmit_skb(struct vxlan_sock *vs,
1424 struct dst_entry *dst, struct sk_buff *skb,
1425 struct net_device *dev, struct in6_addr *saddr,
1426 struct in6_addr *daddr, __u8 prio, __u8 ttl,
1427 __be16 src_port, __be16 dst_port, __be32 vni)
1429 struct ipv6hdr *ip6h;
1430 struct vxlanhdr *vxh;
1435 if (!skb->encapsulation) {
1436 skb_reset_inner_headers(skb);
1437 skb->encapsulation = 1;
1440 skb_scrub_packet(skb, false);
1442 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1443 + VXLAN_HLEN + sizeof(struct ipv6hdr)
1444 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1446 /* Need space for new headers (invalidates iph ptr) */
1447 err = skb_cow_head(skb, min_headroom);
1451 if (vlan_tx_tag_present(skb)) {
1452 if (WARN_ON(!__vlan_put_tag(skb,
1454 vlan_tx_tag_get(skb))))
1460 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1461 vxh->vx_flags = htonl(VXLAN_FLAGS);
1464 __skb_push(skb, sizeof(*uh));
1465 skb_reset_transport_header(skb);
1468 uh->dest = dst_port;
1469 uh->source = src_port;
1471 uh->len = htons(skb->len);
1474 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1475 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1477 skb_dst_set(skb, dst);
1479 if (!skb_is_gso(skb) && !(dst->dev->features & NETIF_F_IPV6_CSUM)) {
1480 __wsum csum = skb_checksum(skb, 0, skb->len, 0);
1481 skb->ip_summed = CHECKSUM_UNNECESSARY;
1482 uh->check = csum_ipv6_magic(saddr, daddr, skb->len,
1485 uh->check = CSUM_MANGLED_0;
1487 skb->ip_summed = CHECKSUM_PARTIAL;
1488 skb->csum_start = skb_transport_header(skb) - skb->head;
1489 skb->csum_offset = offsetof(struct udphdr, check);
1490 uh->check = ~csum_ipv6_magic(saddr, daddr,
1491 skb->len, IPPROTO_UDP, 0);
1494 __skb_push(skb, sizeof(*ip6h));
1495 skb_reset_network_header(skb);
1496 ip6h = ipv6_hdr(skb);
1498 ip6h->priority = prio;
1499 ip6h->flow_lbl[0] = 0;
1500 ip6h->flow_lbl[1] = 0;
1501 ip6h->flow_lbl[2] = 0;
1502 ip6h->payload_len = htons(skb->len);
1503 ip6h->nexthdr = IPPROTO_UDP;
1504 ip6h->hop_limit = ttl;
1505 ip6h->daddr = *daddr;
1506 ip6h->saddr = *saddr;
1508 vxlan_set_owner(vs->sock->sk, skb);
1510 err = handle_offloads(skb);
1514 ip6tunnel_xmit(skb, dev);
1519 int vxlan_xmit_skb(struct vxlan_sock *vs,
1520 struct rtable *rt, struct sk_buff *skb,
1521 __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1522 __be16 src_port, __be16 dst_port, __be32 vni)
1524 struct vxlanhdr *vxh;
1529 if (!skb->encapsulation) {
1530 skb_reset_inner_headers(skb);
1531 skb->encapsulation = 1;
1534 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1535 + VXLAN_HLEN + sizeof(struct iphdr)
1536 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1538 /* Need space for new headers (invalidates iph ptr) */
1539 err = skb_cow_head(skb, min_headroom);
1543 if (vlan_tx_tag_present(skb)) {
1544 if (WARN_ON(!__vlan_put_tag(skb,
1546 vlan_tx_tag_get(skb))))
1552 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1553 vxh->vx_flags = htonl(VXLAN_FLAGS);
1556 __skb_push(skb, sizeof(*uh));
1557 skb_reset_transport_header(skb);
1560 uh->dest = dst_port;
1561 uh->source = src_port;
1563 uh->len = htons(skb->len);
1566 vxlan_set_owner(vs->sock->sk, skb);
1568 err = handle_offloads(skb);
1572 return iptunnel_xmit(rt, skb, src, dst, IPPROTO_UDP, tos, ttl, df,
1575 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1577 /* Bypass encapsulation if the destination is local */
1578 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1579 struct vxlan_dev *dst_vxlan)
1581 struct pcpu_tstats *tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1582 struct pcpu_tstats *rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1583 union vxlan_addr loopback;
1584 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1586 skb->pkt_type = PACKET_HOST;
1587 skb->encapsulation = 0;
1588 skb->dev = dst_vxlan->dev;
1589 __skb_pull(skb, skb_network_offset(skb));
1591 if (remote_ip->sa.sa_family == AF_INET) {
1592 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1593 loopback.sa.sa_family = AF_INET;
1594 #if IS_ENABLED(CONFIG_IPV6)
1596 loopback.sin6.sin6_addr = in6addr_loopback;
1597 loopback.sa.sa_family = AF_INET6;
1601 if (dst_vxlan->flags & VXLAN_F_LEARN)
1602 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1604 u64_stats_update_begin(&tx_stats->syncp);
1605 tx_stats->tx_packets++;
1606 tx_stats->tx_bytes += skb->len;
1607 u64_stats_update_end(&tx_stats->syncp);
1609 if (netif_rx(skb) == NET_RX_SUCCESS) {
1610 u64_stats_update_begin(&rx_stats->syncp);
1611 rx_stats->rx_packets++;
1612 rx_stats->rx_bytes += skb->len;
1613 u64_stats_update_end(&rx_stats->syncp);
1615 skb->dev->stats.rx_dropped++;
1619 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1620 struct vxlan_rdst *rdst, bool did_rsc)
1622 struct vxlan_dev *vxlan = netdev_priv(dev);
1623 struct rtable *rt = NULL;
1624 const struct iphdr *old_iph;
1626 union vxlan_addr *dst;
1627 __be16 src_port = 0, dst_port;
1633 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1634 vni = rdst->remote_vni;
1635 dst = &rdst->remote_ip;
1637 if (vxlan_addr_any(dst)) {
1639 /* short-circuited back to local bridge */
1640 vxlan_encap_bypass(skb, vxlan, vxlan);
1646 old_iph = ip_hdr(skb);
1649 if (!ttl && vxlan_addr_multicast(dst))
1654 tos = ip_tunnel_get_dsfield(old_iph, skb);
1656 src_port = vxlan_src_port(vxlan->port_min, vxlan->port_max, skb);
1658 if (dst->sa.sa_family == AF_INET) {
1659 memset(&fl4, 0, sizeof(fl4));
1660 fl4.flowi4_oif = rdst->remote_ifindex;
1661 fl4.flowi4_tos = RT_TOS(tos);
1662 fl4.daddr = dst->sin.sin_addr.s_addr;
1663 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1665 rt = ip_route_output_key(dev_net(dev), &fl4);
1667 netdev_dbg(dev, "no route to %pI4\n",
1668 &dst->sin.sin_addr.s_addr);
1669 dev->stats.tx_carrier_errors++;
1673 if (rt->dst.dev == dev) {
1674 netdev_dbg(dev, "circular route to %pI4\n",
1675 &dst->sin.sin_addr.s_addr);
1676 dev->stats.collisions++;
1680 /* Bypass encapsulation if the destination is local */
1681 if (rt->rt_flags & RTCF_LOCAL &&
1682 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1683 struct vxlan_dev *dst_vxlan;
1686 dst_vxlan = vxlan_find_vni(dev_net(dev), vni, dst_port);
1689 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1693 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1694 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1696 err = vxlan_xmit_skb(vxlan->vn_sock, rt, skb,
1697 fl4.saddr, dst->sin.sin_addr.s_addr,
1698 tos, ttl, df, src_port, dst_port,
1703 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1704 #if IS_ENABLED(CONFIG_IPV6)
1706 struct sock *sk = vxlan->vn_sock->sock->sk;
1707 struct dst_entry *ndst;
1711 memset(&fl6, 0, sizeof(fl6));
1712 fl6.flowi6_oif = rdst->remote_ifindex;
1713 fl6.daddr = dst->sin6.sin6_addr;
1714 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
1715 fl6.flowi6_proto = IPPROTO_UDP;
1717 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
1718 netdev_dbg(dev, "no route to %pI6\n",
1719 &dst->sin6.sin6_addr);
1720 dev->stats.tx_carrier_errors++;
1724 if (ndst->dev == dev) {
1725 netdev_dbg(dev, "circular route to %pI6\n",
1726 &dst->sin6.sin6_addr);
1728 dev->stats.collisions++;
1732 /* Bypass encapsulation if the destination is local */
1733 flags = ((struct rt6_info *)ndst)->rt6i_flags;
1734 if (flags & RTF_LOCAL &&
1735 !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1736 struct vxlan_dev *dst_vxlan;
1739 dst_vxlan = vxlan_find_vni(dev_net(dev), vni, dst_port);
1742 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1746 ttl = ttl ? : ip6_dst_hoplimit(ndst);
1748 err = vxlan6_xmit_skb(vxlan->vn_sock, ndst, skb,
1749 dev, &fl6.saddr, &fl6.daddr, 0, ttl,
1750 src_port, dst_port, htonl(vni << 8));
1757 dev->stats.tx_dropped++;
1763 dev->stats.tx_errors++;
1768 /* Transmit local packets over Vxlan
1770 * Outer IP header inherits ECN and DF from inner header.
1771 * Outer UDP destination is the VXLAN assigned port.
1772 * source port is based on hash of flow
1774 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1776 struct vxlan_dev *vxlan = netdev_priv(dev);
1778 bool did_rsc = false;
1779 struct vxlan_rdst *rdst;
1780 struct vxlan_fdb *f;
1782 skb_reset_mac_header(skb);
1785 if ((vxlan->flags & VXLAN_F_PROXY)) {
1786 if (ntohs(eth->h_proto) == ETH_P_ARP)
1787 return arp_reduce(dev, skb);
1788 #if IS_ENABLED(CONFIG_IPV6)
1789 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
1790 skb->len >= sizeof(struct ipv6hdr) + sizeof(struct nd_msg) &&
1791 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
1794 msg = (struct nd_msg *)skb_transport_header(skb);
1795 if (msg->icmph.icmp6_code == 0 &&
1796 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
1797 return neigh_reduce(dev, skb);
1802 f = vxlan_find_mac(vxlan, eth->h_dest);
1805 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1806 (ntohs(eth->h_proto) == ETH_P_IP ||
1807 ntohs(eth->h_proto) == ETH_P_IPV6)) {
1808 did_rsc = route_shortcircuit(dev, skb);
1810 f = vxlan_find_mac(vxlan, eth->h_dest);
1814 f = vxlan_find_mac(vxlan, all_zeros_mac);
1816 if ((vxlan->flags & VXLAN_F_L2MISS) &&
1817 !is_multicast_ether_addr(eth->h_dest))
1818 vxlan_fdb_miss(vxlan, eth->h_dest);
1820 dev->stats.tx_dropped++;
1822 return NETDEV_TX_OK;
1826 list_for_each_entry_rcu(rdst, &f->remotes, list) {
1827 struct sk_buff *skb1;
1829 skb1 = skb_clone(skb, GFP_ATOMIC);
1831 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1835 return NETDEV_TX_OK;
1838 /* Walk the forwarding table and purge stale entries */
1839 static void vxlan_cleanup(unsigned long arg)
1841 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1842 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1845 if (!netif_running(vxlan->dev))
1848 spin_lock_bh(&vxlan->hash_lock);
1849 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1850 struct hlist_node *p, *n;
1851 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1853 = container_of(p, struct vxlan_fdb, hlist);
1854 unsigned long timeout;
1856 if (f->state & NUD_PERMANENT)
1859 timeout = f->used + vxlan->age_interval * HZ;
1860 if (time_before_eq(timeout, jiffies)) {
1861 netdev_dbg(vxlan->dev,
1862 "garbage collect %pM\n",
1864 f->state = NUD_STALE;
1865 vxlan_fdb_destroy(vxlan, f);
1866 } else if (time_before(timeout, next_timer))
1867 next_timer = timeout;
1870 spin_unlock_bh(&vxlan->hash_lock);
1872 mod_timer(&vxlan->age_timer, next_timer);
1875 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
1877 __u32 vni = vxlan->default_dst.remote_vni;
1879 vxlan->vn_sock = vs;
1880 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
1883 /* Setup stats when device is created */
1884 static int vxlan_init(struct net_device *dev)
1886 struct vxlan_dev *vxlan = netdev_priv(dev);
1887 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
1888 struct vxlan_sock *vs;
1890 dev->tstats = alloc_percpu(struct pcpu_tstats);
1894 spin_lock(&vn->sock_lock);
1895 vs = vxlan_find_sock(dev_net(dev), vxlan->dst_port);
1897 /* If we have a socket with same port already, reuse it */
1898 atomic_inc(&vs->refcnt);
1899 vxlan_vs_add_dev(vs, vxlan);
1901 /* otherwise make new socket outside of RTNL */
1903 queue_work(vxlan_wq, &vxlan->sock_work);
1905 spin_unlock(&vn->sock_lock);
1910 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
1912 struct vxlan_fdb *f;
1914 spin_lock_bh(&vxlan->hash_lock);
1915 f = __vxlan_find_mac(vxlan, all_zeros_mac);
1917 vxlan_fdb_destroy(vxlan, f);
1918 spin_unlock_bh(&vxlan->hash_lock);
1921 static void vxlan_uninit(struct net_device *dev)
1923 struct vxlan_dev *vxlan = netdev_priv(dev);
1924 struct vxlan_sock *vs = vxlan->vn_sock;
1926 vxlan_fdb_delete_default(vxlan);
1929 vxlan_sock_release(vs);
1930 free_percpu(dev->tstats);
1933 /* Start ageing timer and join group when device is brought up */
1934 static int vxlan_open(struct net_device *dev)
1936 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
1937 struct vxlan_dev *vxlan = netdev_priv(dev);
1938 struct vxlan_sock *vs = vxlan->vn_sock;
1940 /* socket hasn't been created */
1944 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
1945 vxlan_group_used(vn, &vxlan->default_dst.remote_ip)) {
1946 vxlan_sock_hold(vs);
1948 queue_work(vxlan_wq, &vxlan->igmp_join);
1951 if (vxlan->age_interval)
1952 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
1957 /* Purge the forwarding table */
1958 static void vxlan_flush(struct vxlan_dev *vxlan)
1962 spin_lock_bh(&vxlan->hash_lock);
1963 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1964 struct hlist_node *p, *n;
1965 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1967 = container_of(p, struct vxlan_fdb, hlist);
1968 /* the all_zeros_mac entry is deleted at vxlan_uninit */
1969 if (!is_zero_ether_addr(f->eth_addr))
1970 vxlan_fdb_destroy(vxlan, f);
1973 spin_unlock_bh(&vxlan->hash_lock);
1976 /* Cleanup timer and forwarding table on shutdown */
1977 static int vxlan_stop(struct net_device *dev)
1979 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
1980 struct vxlan_dev *vxlan = netdev_priv(dev);
1981 struct vxlan_sock *vs = vxlan->vn_sock;
1983 if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
1984 ! vxlan_group_used(vn, &vxlan->default_dst.remote_ip)) {
1985 vxlan_sock_hold(vs);
1987 queue_work(vxlan_wq, &vxlan->igmp_leave);
1990 del_timer_sync(&vxlan->age_timer);
1997 /* Stub, nothing needs to be done. */
1998 static void vxlan_set_multicast_list(struct net_device *dev)
2002 static const struct net_device_ops vxlan_netdev_ops = {
2003 .ndo_init = vxlan_init,
2004 .ndo_uninit = vxlan_uninit,
2005 .ndo_open = vxlan_open,
2006 .ndo_stop = vxlan_stop,
2007 .ndo_start_xmit = vxlan_xmit,
2008 .ndo_get_stats64 = ip_tunnel_get_stats64,
2009 .ndo_set_rx_mode = vxlan_set_multicast_list,
2010 .ndo_change_mtu = eth_change_mtu,
2011 .ndo_validate_addr = eth_validate_addr,
2012 .ndo_set_mac_address = eth_mac_addr,
2013 .ndo_fdb_add = vxlan_fdb_add,
2014 .ndo_fdb_del = vxlan_fdb_delete,
2015 .ndo_fdb_dump = vxlan_fdb_dump,
2018 /* Info for udev, that this is a virtual tunnel endpoint */
2019 static struct device_type vxlan_type = {
2023 /* Calls the ndo_add_vxlan_port of the caller in order to
2024 * supply the listening VXLAN udp ports.
2026 void vxlan_get_rx_port(struct net_device *dev)
2028 struct vxlan_sock *vs;
2029 struct net *net = dev_net(dev);
2030 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2031 sa_family_t sa_family;
2035 if (!dev || !dev->netdev_ops || !dev->netdev_ops->ndo_add_vxlan_port)
2038 spin_lock(&vn->sock_lock);
2039 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2040 hlist_for_each_entry_rcu(vs, vs_head(net, i), hlist) {
2041 port = htons(inet_sk(vs->sock->sk)->inet_sport);
2042 sa_family = vs->sock->sk->sk_family;
2043 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2047 spin_unlock(&vn->sock_lock);
2049 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2051 /* Initialize the device structure. */
2052 static void vxlan_setup(struct net_device *dev)
2054 struct vxlan_dev *vxlan = netdev_priv(dev);
2058 eth_hw_addr_random(dev);
2060 if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2061 dev->hard_header_len = ETH_HLEN + VXLAN6_HEADROOM;
2063 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
2065 dev->netdev_ops = &vxlan_netdev_ops;
2066 dev->destructor = free_netdev;
2067 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2069 dev->tx_queue_len = 0;
2070 dev->features |= NETIF_F_LLTX;
2071 dev->features |= NETIF_F_NETNS_LOCAL;
2072 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2073 dev->features |= NETIF_F_RXCSUM;
2074 dev->features |= NETIF_F_GSO_SOFTWARE;
2076 dev->vlan_features = dev->features;
2077 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2078 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2079 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2080 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2081 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
2082 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2084 INIT_LIST_HEAD(&vxlan->next);
2085 spin_lock_init(&vxlan->hash_lock);
2086 INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join);
2087 INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave);
2088 INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
2090 init_timer_deferrable(&vxlan->age_timer);
2091 vxlan->age_timer.function = vxlan_cleanup;
2092 vxlan->age_timer.data = (unsigned long) vxlan;
2094 inet_get_local_port_range(&low, &high);
2095 vxlan->port_min = low;
2096 vxlan->port_max = high;
2097 vxlan->dst_port = htons(vxlan_port);
2101 for (h = 0; h < FDB_HASH_SIZE; ++h)
2102 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2105 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2106 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2107 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2108 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2109 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2110 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2111 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2112 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2113 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2114 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2115 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2116 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2117 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2118 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2119 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2120 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2121 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2122 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2125 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2127 if (tb[IFLA_ADDRESS]) {
2128 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2129 pr_debug("invalid link address (not ethernet)\n");
2133 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2134 pr_debug("invalid all zero ethernet address\n");
2135 return -EADDRNOTAVAIL;
2142 if (data[IFLA_VXLAN_ID]) {
2143 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2144 if (id >= VXLAN_VID_MASK)
2148 if (data[IFLA_VXLAN_PORT_RANGE]) {
2149 const struct ifla_vxlan_port_range *p
2150 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2152 if (ntohs(p->high) < ntohs(p->low)) {
2153 pr_debug("port range %u .. %u not valid\n",
2154 ntohs(p->low), ntohs(p->high));
2162 static void vxlan_get_drvinfo(struct net_device *netdev,
2163 struct ethtool_drvinfo *drvinfo)
2165 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2166 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2169 static const struct ethtool_ops vxlan_ethtool_ops = {
2170 .get_drvinfo = vxlan_get_drvinfo,
2171 .get_link = ethtool_op_get_link,
2174 static void vxlan_del_work(struct work_struct *work)
2176 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2178 sk_release_kernel(vs->sock->sk);
2182 #if IS_ENABLED(CONFIG_IPV6)
2183 /* Create UDP socket for encapsulation receive. AF_INET6 socket
2184 * could be used for both IPv4 and IPv6 communications, but
2185 * users may set bindv6only=1.
2187 static int create_v6_sock(struct net *net, __be16 port, struct socket **psock)
2190 struct socket *sock;
2191 struct sockaddr_in6 vxlan_addr = {
2192 .sin6_family = AF_INET6,
2197 rc = sock_create_kern(AF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
2199 pr_debug("UDPv6 socket create failed\n");
2203 /* Put in proper namespace */
2205 sk_change_net(sk, net);
2207 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
2208 (char *)&val, sizeof(val));
2209 rc = kernel_bind(sock, (struct sockaddr *)&vxlan_addr,
2210 sizeof(struct sockaddr_in6));
2212 pr_debug("bind for UDPv6 socket %pI6:%u (%d)\n",
2213 &vxlan_addr.sin6_addr, ntohs(vxlan_addr.sin6_port), rc);
2214 sk_release_kernel(sk);
2217 /* At this point, IPv6 module should have been loaded in
2218 * sock_create_kern().
2223 /* Disable multicast loopback */
2224 inet_sk(sk)->mc_loop = 0;
2230 static int create_v6_sock(struct net *net, __be16 port, struct socket **psock)
2232 return -EPFNOSUPPORT;
2236 static int create_v4_sock(struct net *net, __be16 port, struct socket **psock)
2239 struct socket *sock;
2240 struct sockaddr_in vxlan_addr = {
2241 .sin_family = AF_INET,
2242 .sin_addr.s_addr = htonl(INADDR_ANY),
2247 /* Create UDP socket for encapsulation receive. */
2248 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
2250 pr_debug("UDP socket create failed\n");
2254 /* Put in proper namespace */
2256 sk_change_net(sk, net);
2258 rc = kernel_bind(sock, (struct sockaddr *) &vxlan_addr,
2259 sizeof(vxlan_addr));
2261 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
2262 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
2263 sk_release_kernel(sk);
2268 /* Disable multicast loopback */
2269 inet_sk(sk)->mc_loop = 0;
2273 /* Create new listen socket if needed */
2274 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2275 vxlan_rcv_t *rcv, void *data, bool ipv6)
2277 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2278 struct vxlan_sock *vs;
2279 struct socket *sock;
2284 vs = kmalloc(sizeof(*vs), GFP_KERNEL);
2286 return ERR_PTR(-ENOMEM);
2288 for (h = 0; h < VNI_HASH_SIZE; ++h)
2289 INIT_HLIST_HEAD(&vs->vni_list[h]);
2291 INIT_WORK(&vs->del_work, vxlan_del_work);
2294 rc = create_v6_sock(net, port, &sock);
2296 rc = create_v4_sock(net, port, &sock);
2304 atomic_set(&vs->refcnt, 1);
2308 vs->sock->sk->sk_user_data = vs;
2310 spin_lock(&vn->sock_lock);
2311 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2312 vxlan_notify_add_rx_port(sk);
2313 spin_unlock(&vn->sock_lock);
2315 /* Mark socket as an encapsulation socket. */
2316 udp_sk(sk)->encap_type = 1;
2317 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
2318 #if IS_ENABLED(CONFIG_IPV6)
2320 ipv6_stub->udpv6_encap_enable();
2328 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2329 vxlan_rcv_t *rcv, void *data,
2330 bool no_share, bool ipv6)
2332 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2333 struct vxlan_sock *vs;
2335 vs = vxlan_socket_create(net, port, rcv, data, ipv6);
2339 if (no_share) /* Return error if sharing is not allowed. */
2342 spin_lock(&vn->sock_lock);
2343 vs = vxlan_find_sock(net, port);
2346 atomic_inc(&vs->refcnt);
2348 vs = ERR_PTR(-EBUSY);
2350 spin_unlock(&vn->sock_lock);
2353 vs = ERR_PTR(-EINVAL);
2357 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2359 /* Scheduled at device creation to bind to a socket */
2360 static void vxlan_sock_work(struct work_struct *work)
2362 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2363 struct net *net = dev_net(vxlan->dev);
2364 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2365 __be16 port = vxlan->dst_port;
2366 struct vxlan_sock *nvs;
2368 nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags & VXLAN_F_IPV6);
2369 spin_lock(&vn->sock_lock);
2371 vxlan_vs_add_dev(nvs, vxlan);
2372 spin_unlock(&vn->sock_lock);
2374 dev_put(vxlan->dev);
2377 static int vxlan_newlink(struct net *net, struct net_device *dev,
2378 struct nlattr *tb[], struct nlattr *data[])
2380 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2381 struct vxlan_dev *vxlan = netdev_priv(dev);
2382 struct vxlan_rdst *dst = &vxlan->default_dst;
2385 bool use_ipv6 = false;
2387 if (!data[IFLA_VXLAN_ID])
2390 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2391 dst->remote_vni = vni;
2393 if (data[IFLA_VXLAN_GROUP]) {
2394 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2395 dst->remote_ip.sa.sa_family = AF_INET;
2396 } else if (data[IFLA_VXLAN_GROUP6]) {
2397 if (!IS_ENABLED(CONFIG_IPV6))
2398 return -EPFNOSUPPORT;
2400 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2401 sizeof(struct in6_addr));
2402 dst->remote_ip.sa.sa_family = AF_INET6;
2406 if (data[IFLA_VXLAN_LOCAL]) {
2407 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2408 vxlan->saddr.sa.sa_family = AF_INET;
2409 } else if (data[IFLA_VXLAN_LOCAL6]) {
2410 if (!IS_ENABLED(CONFIG_IPV6))
2411 return -EPFNOSUPPORT;
2413 /* TODO: respect scope id */
2414 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2415 sizeof(struct in6_addr));
2416 vxlan->saddr.sa.sa_family = AF_INET6;
2420 if (data[IFLA_VXLAN_LINK] &&
2421 (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2422 struct net_device *lowerdev
2423 = __dev_get_by_index(net, dst->remote_ifindex);
2426 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2430 #if IS_ENABLED(CONFIG_IPV6)
2432 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2433 if (idev && idev->cnf.disable_ipv6) {
2434 pr_info("IPv6 is disabled via sysctl\n");
2437 vxlan->flags |= VXLAN_F_IPV6;
2442 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2444 /* update header length based on lower device */
2445 dev->hard_header_len = lowerdev->hard_header_len +
2446 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2449 if (data[IFLA_VXLAN_TOS])
2450 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2452 if (data[IFLA_VXLAN_TTL])
2453 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2455 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2456 vxlan->flags |= VXLAN_F_LEARN;
2458 if (data[IFLA_VXLAN_AGEING])
2459 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2461 vxlan->age_interval = FDB_AGE_DEFAULT;
2463 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2464 vxlan->flags |= VXLAN_F_PROXY;
2466 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2467 vxlan->flags |= VXLAN_F_RSC;
2469 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2470 vxlan->flags |= VXLAN_F_L2MISS;
2472 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2473 vxlan->flags |= VXLAN_F_L3MISS;
2475 if (data[IFLA_VXLAN_LIMIT])
2476 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2478 if (data[IFLA_VXLAN_PORT_RANGE]) {
2479 const struct ifla_vxlan_port_range *p
2480 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2481 vxlan->port_min = ntohs(p->low);
2482 vxlan->port_max = ntohs(p->high);
2485 if (data[IFLA_VXLAN_PORT])
2486 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2488 if (vxlan_find_vni(net, vni, vxlan->dst_port)) {
2489 pr_info("duplicate VNI %u\n", vni);
2493 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
2495 /* create an fdb entry for default destination */
2496 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2497 &vxlan->default_dst.remote_ip,
2498 NUD_REACHABLE|NUD_PERMANENT,
2499 NLM_F_EXCL|NLM_F_CREATE,
2500 vxlan->dst_port, vxlan->default_dst.remote_vni,
2501 vxlan->default_dst.remote_ifindex, NTF_SELF);
2505 err = register_netdevice(dev);
2507 vxlan_fdb_delete_default(vxlan);
2511 list_add(&vxlan->next, &vn->vxlan_list);
2516 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2518 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2519 struct vxlan_dev *vxlan = netdev_priv(dev);
2521 spin_lock(&vn->sock_lock);
2522 if (!hlist_unhashed(&vxlan->hlist))
2523 hlist_del_rcu(&vxlan->hlist);
2524 spin_unlock(&vn->sock_lock);
2526 list_del(&vxlan->next);
2527 unregister_netdevice_queue(dev, head);
2530 static size_t vxlan_get_size(const struct net_device *dev)
2533 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
2534 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2535 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2536 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2537 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
2538 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
2539 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
2540 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
2541 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
2542 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
2543 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
2544 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2545 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2546 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2547 nla_total_size(sizeof(__be16))+ /* IFLA_VXLAN_PORT */
2551 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2553 const struct vxlan_dev *vxlan = netdev_priv(dev);
2554 const struct vxlan_rdst *dst = &vxlan->default_dst;
2555 struct ifla_vxlan_port_range ports = {
2556 .low = htons(vxlan->port_min),
2557 .high = htons(vxlan->port_max),
2560 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2561 goto nla_put_failure;
2563 if (!vxlan_addr_any(&dst->remote_ip)) {
2564 if (dst->remote_ip.sa.sa_family == AF_INET) {
2565 if (nla_put_be32(skb, IFLA_VXLAN_GROUP,
2566 dst->remote_ip.sin.sin_addr.s_addr))
2567 goto nla_put_failure;
2568 #if IS_ENABLED(CONFIG_IPV6)
2570 if (nla_put(skb, IFLA_VXLAN_GROUP6, sizeof(struct in6_addr),
2571 &dst->remote_ip.sin6.sin6_addr))
2572 goto nla_put_failure;
2577 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2578 goto nla_put_failure;
2580 if (!vxlan_addr_any(&vxlan->saddr)) {
2581 if (vxlan->saddr.sa.sa_family == AF_INET) {
2582 if (nla_put_be32(skb, IFLA_VXLAN_LOCAL,
2583 vxlan->saddr.sin.sin_addr.s_addr))
2584 goto nla_put_failure;
2585 #if IS_ENABLED(CONFIG_IPV6)
2587 if (nla_put(skb, IFLA_VXLAN_LOCAL6, sizeof(struct in6_addr),
2588 &vxlan->saddr.sin6.sin6_addr))
2589 goto nla_put_failure;
2594 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2595 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2596 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2597 !!(vxlan->flags & VXLAN_F_LEARN)) ||
2598 nla_put_u8(skb, IFLA_VXLAN_PROXY,
2599 !!(vxlan->flags & VXLAN_F_PROXY)) ||
2600 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2601 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2602 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2603 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2604 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2605 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2606 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2607 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port))
2608 goto nla_put_failure;
2610 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2611 goto nla_put_failure;
2619 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2621 .maxtype = IFLA_VXLAN_MAX,
2622 .policy = vxlan_policy,
2623 .priv_size = sizeof(struct vxlan_dev),
2624 .setup = vxlan_setup,
2625 .validate = vxlan_validate,
2626 .newlink = vxlan_newlink,
2627 .dellink = vxlan_dellink,
2628 .get_size = vxlan_get_size,
2629 .fill_info = vxlan_fill_info,
2632 static __net_init int vxlan_init_net(struct net *net)
2634 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2637 INIT_LIST_HEAD(&vn->vxlan_list);
2638 spin_lock_init(&vn->sock_lock);
2640 for (h = 0; h < PORT_HASH_SIZE; ++h)
2641 INIT_HLIST_HEAD(&vn->sock_list[h]);
2646 static __net_exit void vxlan_exit_net(struct net *net)
2648 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2649 struct vxlan_dev *vxlan;
2653 list_for_each_entry(vxlan, &vn->vxlan_list, next)
2654 unregister_netdevice_queue(vxlan->dev, &list);
2655 unregister_netdevice_many(&list);
2659 static struct pernet_operations vxlan_net_ops = {
2660 .init = vxlan_init_net,
2661 .exit = vxlan_exit_net,
2662 .id = &vxlan_net_id,
2663 .size = sizeof(struct vxlan_net),
2666 static int __init vxlan_init_module(void)
2670 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
2674 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
2676 rc = register_pernet_device(&vxlan_net_ops);
2680 rc = rtnl_link_register(&vxlan_link_ops);
2687 unregister_pernet_device(&vxlan_net_ops);
2689 destroy_workqueue(vxlan_wq);
2692 late_initcall(vxlan_init_module);
2694 static void __exit vxlan_cleanup_module(void)
2696 rtnl_link_unregister(&vxlan_link_ops);
2697 destroy_workqueue(vxlan_wq);
2698 unregister_pernet_device(&vxlan_net_ops);
2701 module_exit(vxlan_cleanup_module);
2703 MODULE_LICENSE("GPL");
2704 MODULE_VERSION(VXLAN_VERSION);
2705 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
2706 MODULE_ALIAS_RTNL_LINK("vxlan");