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 #define VXLAN_HLEN (sizeof(struct udphdr) + sizeof(struct vxlanhdr))
65 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
67 /* VXLAN protocol header */
73 /* UDP port for VXLAN traffic.
74 * The IANA assigned port is 4789, but the Linux default is 8472
75 * for compatibility with early adopters.
77 static unsigned short vxlan_port __read_mostly = 8472;
78 module_param_named(udp_port, vxlan_port, ushort, 0444);
79 MODULE_PARM_DESC(udp_port, "Destination UDP port");
81 static bool log_ecn_error = true;
82 module_param(log_ecn_error, bool, 0644);
83 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
85 static int vxlan_net_id;
87 static const u8 all_zeros_mac[ETH_ALEN];
89 /* per-network namespace private data for this module */
91 struct list_head vxlan_list;
92 struct hlist_head sock_list[PORT_HASH_SIZE];
97 struct sockaddr_in sin;
98 struct sockaddr_in6 sin6;
103 union vxlan_addr remote_ip;
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 union vxlan_addr 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_join;
140 struct work_struct igmp_leave;
142 unsigned long age_interval;
143 struct timer_list age_timer;
144 spinlock_t hash_lock;
145 unsigned int addrcnt;
146 unsigned int addrmax;
148 struct hlist_head fdb_head[FDB_HASH_SIZE];
151 #define VXLAN_F_LEARN 0x01
152 #define VXLAN_F_PROXY 0x02
153 #define VXLAN_F_RSC 0x04
154 #define VXLAN_F_L2MISS 0x08
155 #define VXLAN_F_L3MISS 0x10
156 #define VXLAN_F_IPV6 0x20 /* internal flag */
158 /* salt for hash table */
159 static u32 vxlan_salt __read_mostly;
160 static struct workqueue_struct *vxlan_wq;
162 static void vxlan_sock_work(struct work_struct *work);
164 #if IS_ENABLED(CONFIG_IPV6)
166 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
168 if (a->sa.sa_family != b->sa.sa_family)
170 if (a->sa.sa_family == AF_INET6)
171 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
173 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
176 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
178 if (ipa->sa.sa_family == AF_INET6)
179 return ipv6_addr_any(&ipa->sin6.sin6_addr);
181 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
184 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
186 if (ipa->sa.sa_family == AF_INET6)
187 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
189 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
192 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
194 if (nla_len(nla) >= sizeof(struct in6_addr)) {
195 nla_memcpy(&ip->sin6.sin6_addr, nla, sizeof(struct in6_addr));
196 ip->sa.sa_family = AF_INET6;
198 } else if (nla_len(nla) >= sizeof(__be32)) {
199 ip->sin.sin_addr.s_addr = nla_get_be32(nla);
200 ip->sa.sa_family = AF_INET;
203 return -EAFNOSUPPORT;
207 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
208 const union vxlan_addr *ip)
210 if (ip->sa.sa_family == AF_INET6)
211 return nla_put(skb, attr, sizeof(struct in6_addr), &ip->sin6.sin6_addr);
213 return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
216 #else /* !CONFIG_IPV6 */
219 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
221 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
224 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
226 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
229 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
231 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
234 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
236 if (nla_len(nla) >= sizeof(struct in6_addr)) {
237 return -EAFNOSUPPORT;
238 } else if (nla_len(nla) >= sizeof(__be32)) {
239 ip->sin.sin_addr.s_addr = nla_get_be32(nla);
240 ip->sa.sa_family = AF_INET;
243 return -EAFNOSUPPORT;
247 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
248 const union vxlan_addr *ip)
250 return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
254 /* Virtual Network hash table head */
255 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
257 return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
260 /* Socket hash table head */
261 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
263 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
265 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
268 /* First remote destination for a forwarding entry.
269 * Guaranteed to be non-NULL because remotes are never deleted.
271 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
273 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
276 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
278 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
281 /* Find VXLAN socket based on network namespace and UDP port */
282 static struct vxlan_sock *vxlan_find_sock(struct net *net, __be16 port)
284 struct vxlan_sock *vs;
286 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
287 if (inet_sk(vs->sock->sk)->inet_sport == port)
293 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
295 struct vxlan_dev *vxlan;
297 hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
298 if (vxlan->default_dst.remote_vni == id)
305 /* Look up VNI in a per net namespace table */
306 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id, __be16 port)
308 struct vxlan_sock *vs;
310 vs = vxlan_find_sock(net, port);
314 return vxlan_vs_find_vni(vs, id);
317 /* Fill in neighbour message in skbuff. */
318 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
319 const struct vxlan_fdb *fdb,
320 u32 portid, u32 seq, int type, unsigned int flags,
321 const struct vxlan_rdst *rdst)
323 unsigned long now = jiffies;
324 struct nda_cacheinfo ci;
325 struct nlmsghdr *nlh;
327 bool send_ip, send_eth;
329 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
333 ndm = nlmsg_data(nlh);
334 memset(ndm, 0, sizeof(*ndm));
336 send_eth = send_ip = true;
338 if (type == RTM_GETNEIGH) {
339 ndm->ndm_family = AF_INET;
340 send_ip = !vxlan_addr_any(&rdst->remote_ip);
341 send_eth = !is_zero_ether_addr(fdb->eth_addr);
343 ndm->ndm_family = AF_BRIDGE;
344 ndm->ndm_state = fdb->state;
345 ndm->ndm_ifindex = vxlan->dev->ifindex;
346 ndm->ndm_flags = fdb->flags;
347 ndm->ndm_type = NDA_DST;
349 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
350 goto nla_put_failure;
352 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
353 goto nla_put_failure;
355 if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
356 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
357 goto nla_put_failure;
358 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
359 nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
360 goto nla_put_failure;
361 if (rdst->remote_ifindex &&
362 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
363 goto nla_put_failure;
365 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
366 ci.ndm_confirmed = 0;
367 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
370 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
371 goto nla_put_failure;
373 return nlmsg_end(skb, nlh);
376 nlmsg_cancel(skb, nlh);
380 static inline size_t vxlan_nlmsg_size(void)
382 return NLMSG_ALIGN(sizeof(struct ndmsg))
383 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
384 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
385 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
386 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
387 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
388 + nla_total_size(sizeof(struct nda_cacheinfo));
391 static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
392 struct vxlan_fdb *fdb, int type)
394 struct net *net = dev_net(vxlan->dev);
398 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
402 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0,
403 first_remote_rtnl(fdb));
405 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
406 WARN_ON(err == -EMSGSIZE);
411 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
415 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
418 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
420 struct vxlan_dev *vxlan = netdev_priv(dev);
421 struct vxlan_fdb f = {
424 struct vxlan_rdst remote = {
425 .remote_ip = *ipa, /* goes to NDA_DST */
426 .remote_vni = VXLAN_N_VID,
429 INIT_LIST_HEAD(&f.remotes);
430 list_add_rcu(&remote.list, &f.remotes);
432 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
435 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
437 struct vxlan_fdb f = {
441 INIT_LIST_HEAD(&f.remotes);
442 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
444 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
447 /* Hash Ethernet address */
448 static u32 eth_hash(const unsigned char *addr)
450 u64 value = get_unaligned((u64 *)addr);
452 /* only want 6 bytes */
458 return hash_64(value, FDB_HASH_BITS);
461 /* Hash chain to use given mac address */
462 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
465 return &vxlan->fdb_head[eth_hash(mac)];
468 /* Look up Ethernet address in forwarding table */
469 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
473 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
476 hlist_for_each_entry_rcu(f, head, hlist) {
477 if (ether_addr_equal(mac, f->eth_addr))
484 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
489 f = __vxlan_find_mac(vxlan, mac);
496 /* caller should hold vxlan->hash_lock */
497 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
498 union vxlan_addr *ip, __be16 port,
499 __u32 vni, __u32 ifindex)
501 struct vxlan_rdst *rd;
503 list_for_each_entry(rd, &f->remotes, list) {
504 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
505 rd->remote_port == port &&
506 rd->remote_vni == vni &&
507 rd->remote_ifindex == ifindex)
514 /* Replace destination of unicast mac */
515 static int vxlan_fdb_replace(struct vxlan_fdb *f,
516 union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
518 struct vxlan_rdst *rd;
520 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
524 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
528 rd->remote_port = port;
529 rd->remote_vni = vni;
530 rd->remote_ifindex = ifindex;
534 /* Add/update destinations for multicast */
535 static int vxlan_fdb_append(struct vxlan_fdb *f,
536 union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
538 struct vxlan_rdst *rd;
540 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
544 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
548 rd->remote_port = port;
549 rd->remote_vni = vni;
550 rd->remote_ifindex = ifindex;
552 list_add_tail_rcu(&rd->list, &f->remotes);
557 /* Notify netdevs that UDP port started listening */
558 static void vxlan_notify_add_rx_port(struct sock *sk)
560 struct net_device *dev;
561 struct net *net = sock_net(sk);
562 sa_family_t sa_family = sk->sk_family;
563 __be16 port = inet_sk(sk)->inet_sport;
566 for_each_netdev_rcu(net, dev) {
567 if (dev->netdev_ops->ndo_add_vxlan_port)
568 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
574 /* Notify netdevs that UDP port is no more listening */
575 static void vxlan_notify_del_rx_port(struct sock *sk)
577 struct net_device *dev;
578 struct net *net = sock_net(sk);
579 sa_family_t sa_family = sk->sk_family;
580 __be16 port = inet_sk(sk)->inet_sport;
583 for_each_netdev_rcu(net, dev) {
584 if (dev->netdev_ops->ndo_del_vxlan_port)
585 dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
591 /* Add new entry to forwarding table -- assumes lock held */
592 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
593 const u8 *mac, union vxlan_addr *ip,
594 __u16 state, __u16 flags,
595 __be16 port, __u32 vni, __u32 ifindex,
601 f = __vxlan_find_mac(vxlan, mac);
603 if (flags & NLM_F_EXCL) {
604 netdev_dbg(vxlan->dev,
605 "lost race to create %pM\n", mac);
608 if (f->state != state) {
610 f->updated = jiffies;
613 if (f->flags != ndm_flags) {
614 f->flags = ndm_flags;
615 f->updated = jiffies;
618 if ((flags & NLM_F_REPLACE)) {
619 /* Only change unicasts */
620 if (!(is_multicast_ether_addr(f->eth_addr) ||
621 is_zero_ether_addr(f->eth_addr))) {
622 int rc = vxlan_fdb_replace(f, ip, port, vni,
631 if ((flags & NLM_F_APPEND) &&
632 (is_multicast_ether_addr(f->eth_addr) ||
633 is_zero_ether_addr(f->eth_addr))) {
634 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex);
641 if (!(flags & NLM_F_CREATE))
644 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
647 /* Disallow replace to add a multicast entry */
648 if ((flags & NLM_F_REPLACE) &&
649 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
652 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
653 f = kmalloc(sizeof(*f), GFP_ATOMIC);
659 f->flags = ndm_flags;
660 f->updated = f->used = jiffies;
661 INIT_LIST_HEAD(&f->remotes);
662 memcpy(f->eth_addr, mac, ETH_ALEN);
664 vxlan_fdb_append(f, ip, port, vni, ifindex);
667 hlist_add_head_rcu(&f->hlist,
668 vxlan_fdb_head(vxlan, mac));
672 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
677 static void vxlan_fdb_free(struct rcu_head *head)
679 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
680 struct vxlan_rdst *rd, *nd;
682 list_for_each_entry_safe(rd, nd, &f->remotes, list)
687 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
689 netdev_dbg(vxlan->dev,
690 "delete %pM\n", f->eth_addr);
693 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
695 hlist_del_rcu(&f->hlist);
696 call_rcu(&f->rcu, vxlan_fdb_free);
699 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
700 union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
702 struct net *net = dev_net(vxlan->dev);
706 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
710 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
711 if (remote->sa.sa_family == AF_INET) {
712 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
713 ip->sa.sa_family = AF_INET;
714 #if IS_ENABLED(CONFIG_IPV6)
716 ip->sin6.sin6_addr = in6addr_any;
717 ip->sa.sa_family = AF_INET6;
723 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
725 *port = nla_get_be16(tb[NDA_PORT]);
727 *port = vxlan->dst_port;
731 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
733 *vni = nla_get_u32(tb[NDA_VNI]);
735 *vni = vxlan->default_dst.remote_vni;
738 if (tb[NDA_IFINDEX]) {
739 struct net_device *tdev;
741 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
743 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
744 tdev = dev_get_by_index(net, *ifindex);
746 return -EADDRNOTAVAIL;
755 /* Add static entry (via netlink) */
756 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
757 struct net_device *dev,
758 const unsigned char *addr, u16 flags)
760 struct vxlan_dev *vxlan = netdev_priv(dev);
761 /* struct net *net = dev_net(vxlan->dev); */
767 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
768 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
773 if (tb[NDA_DST] == NULL)
776 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
780 spin_lock_bh(&vxlan->hash_lock);
781 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
782 port, vni, ifindex, ndm->ndm_flags);
783 spin_unlock_bh(&vxlan->hash_lock);
788 /* Delete entry (via netlink) */
789 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
790 struct net_device *dev,
791 const unsigned char *addr)
793 struct vxlan_dev *vxlan = netdev_priv(dev);
795 struct vxlan_rdst *rd = NULL;
801 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
807 spin_lock_bh(&vxlan->hash_lock);
808 f = vxlan_find_mac(vxlan, addr);
812 if (!vxlan_addr_any(&ip)) {
813 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
820 /* remove a destination if it's not the only one on the list,
821 * otherwise destroy the fdb entry
823 if (rd && !list_is_singular(&f->remotes)) {
824 list_del_rcu(&rd->list);
829 vxlan_fdb_destroy(vxlan, f);
832 spin_unlock_bh(&vxlan->hash_lock);
837 /* Dump forwarding table */
838 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
839 struct net_device *dev, int idx)
841 struct vxlan_dev *vxlan = netdev_priv(dev);
844 for (h = 0; h < FDB_HASH_SIZE; ++h) {
848 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
849 struct vxlan_rdst *rd;
851 if (idx < cb->args[0])
854 list_for_each_entry_rcu(rd, &f->remotes, list) {
855 err = vxlan_fdb_info(skb, vxlan, f,
856 NETLINK_CB(cb->skb).portid,
871 /* Watch incoming packets to learn mapping between Ethernet address
872 * and Tunnel endpoint.
873 * Return true if packet is bogus and should be droppped.
875 static bool vxlan_snoop(struct net_device *dev,
876 union vxlan_addr *src_ip, const u8 *src_mac)
878 struct vxlan_dev *vxlan = netdev_priv(dev);
881 f = vxlan_find_mac(vxlan, src_mac);
883 struct vxlan_rdst *rdst = first_remote_rcu(f);
885 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
888 /* Don't migrate static entries, drop packets */
889 if (f->state & NUD_NOARP)
894 "%pM migrated from %pIS to %pIS\n",
895 src_mac, &rdst->remote_ip, &src_ip);
897 rdst->remote_ip = *src_ip;
898 f->updated = jiffies;
899 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
901 /* learned new entry */
902 spin_lock(&vxlan->hash_lock);
904 /* close off race between vxlan_flush and incoming packets */
905 if (netif_running(dev))
906 vxlan_fdb_create(vxlan, src_mac, src_ip,
908 NLM_F_EXCL|NLM_F_CREATE,
910 vxlan->default_dst.remote_vni,
912 spin_unlock(&vxlan->hash_lock);
918 /* See if multicast group is already in use by other ID */
919 static bool vxlan_group_used(struct vxlan_net *vn, union vxlan_addr *remote_ip)
921 struct vxlan_dev *vxlan;
923 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
924 if (!netif_running(vxlan->dev))
927 if (vxlan_addr_equal(&vxlan->default_dst.remote_ip,
935 static void vxlan_sock_hold(struct vxlan_sock *vs)
937 atomic_inc(&vs->refcnt);
940 void vxlan_sock_release(struct vxlan_sock *vs)
942 struct sock *sk = vs->sock->sk;
943 struct net *net = sock_net(sk);
944 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
946 if (!atomic_dec_and_test(&vs->refcnt))
949 spin_lock(&vn->sock_lock);
950 hlist_del_rcu(&vs->hlist);
951 rcu_assign_sk_user_data(vs->sock->sk, NULL);
952 vxlan_notify_del_rx_port(sk);
953 spin_unlock(&vn->sock_lock);
955 queue_work(vxlan_wq, &vs->del_work);
957 EXPORT_SYMBOL_GPL(vxlan_sock_release);
959 /* Callback to update multicast group membership when first VNI on
960 * multicast asddress is brought up
961 * Done as workqueue because ip_mc_join_group acquires RTNL.
963 static void vxlan_igmp_join(struct work_struct *work)
965 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join);
966 struct vxlan_sock *vs = vxlan->vn_sock;
967 struct sock *sk = vs->sock->sk;
968 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
969 int ifindex = vxlan->default_dst.remote_ifindex;
972 if (ip->sa.sa_family == AF_INET) {
973 struct ip_mreqn mreq = {
974 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
975 .imr_ifindex = ifindex,
978 ip_mc_join_group(sk, &mreq);
979 #if IS_ENABLED(CONFIG_IPV6)
981 ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
982 &ip->sin6.sin6_addr);
987 vxlan_sock_release(vs);
991 /* Inverse of vxlan_igmp_join when last VNI is brought down */
992 static void vxlan_igmp_leave(struct work_struct *work)
994 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave);
995 struct vxlan_sock *vs = vxlan->vn_sock;
996 struct sock *sk = vs->sock->sk;
997 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
998 int ifindex = vxlan->default_dst.remote_ifindex;
1001 if (ip->sa.sa_family == AF_INET) {
1002 struct ip_mreqn mreq = {
1003 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1004 .imr_ifindex = ifindex,
1007 ip_mc_leave_group(sk, &mreq);
1008 #if IS_ENABLED(CONFIG_IPV6)
1010 ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1011 &ip->sin6.sin6_addr);
1017 vxlan_sock_release(vs);
1018 dev_put(vxlan->dev);
1021 /* Callback from net/ipv4/udp.c to receive packets */
1022 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1024 struct vxlan_sock *vs;
1025 struct vxlanhdr *vxh;
1028 /* Need Vxlan and inner Ethernet header to be present */
1029 if (!pskb_may_pull(skb, VXLAN_HLEN))
1032 /* Return packets with reserved bits set */
1033 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1034 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
1035 (vxh->vx_vni & htonl(0xff))) {
1036 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1037 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1041 if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1044 port = inet_sk(sk)->inet_sport;
1046 vs = rcu_dereference_sk_user_data(sk);
1050 vs->rcv(vs, skb, vxh->vx_vni);
1054 /* Consume bad packet */
1059 /* Return non vxlan pkt */
1063 static void vxlan_rcv(struct vxlan_sock *vs,
1064 struct sk_buff *skb, __be32 vx_vni)
1066 struct iphdr *oip = NULL;
1067 struct ipv6hdr *oip6 = NULL;
1068 struct vxlan_dev *vxlan;
1069 struct pcpu_tstats *stats;
1070 union vxlan_addr saddr;
1073 union vxlan_addr *remote_ip;
1075 vni = ntohl(vx_vni) >> 8;
1076 /* Is this VNI defined? */
1077 vxlan = vxlan_vs_find_vni(vs, vni);
1081 remote_ip = &vxlan->default_dst.remote_ip;
1082 skb_reset_mac_header(skb);
1083 skb->protocol = eth_type_trans(skb, vxlan->dev);
1085 /* Ignore packet loops (and multicast echo) */
1086 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1089 /* Re-examine inner Ethernet packet */
1090 if (remote_ip->sa.sa_family == AF_INET) {
1092 saddr.sin.sin_addr.s_addr = oip->saddr;
1093 saddr.sa.sa_family = AF_INET;
1094 #if IS_ENABLED(CONFIG_IPV6)
1096 oip6 = ipv6_hdr(skb);
1097 saddr.sin6.sin6_addr = oip6->saddr;
1098 saddr.sa.sa_family = AF_INET6;
1102 if ((vxlan->flags & VXLAN_F_LEARN) &&
1103 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1106 skb_reset_network_header(skb);
1108 /* If the NIC driver gave us an encapsulated packet with
1109 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
1110 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
1111 * for us. Otherwise force the upper layers to verify it.
1113 if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
1114 !(vxlan->dev->features & NETIF_F_RXCSUM))
1115 skb->ip_summed = CHECKSUM_NONE;
1117 skb->encapsulation = 0;
1120 err = IP6_ECN_decapsulate(oip6, skb);
1122 err = IP_ECN_decapsulate(oip, skb);
1124 if (unlikely(err)) {
1125 if (log_ecn_error) {
1127 net_info_ratelimited("non-ECT from %pI6\n",
1130 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1131 &oip->saddr, oip->tos);
1134 ++vxlan->dev->stats.rx_frame_errors;
1135 ++vxlan->dev->stats.rx_errors;
1140 stats = this_cpu_ptr(vxlan->dev->tstats);
1141 u64_stats_update_begin(&stats->syncp);
1142 stats->rx_packets++;
1143 stats->rx_bytes += skb->len;
1144 u64_stats_update_end(&stats->syncp);
1150 /* Consume bad packet */
1154 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1156 struct vxlan_dev *vxlan = netdev_priv(dev);
1157 struct arphdr *parp;
1160 struct neighbour *n;
1162 if (dev->flags & IFF_NOARP)
1165 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1166 dev->stats.tx_dropped++;
1169 parp = arp_hdr(skb);
1171 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1172 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1173 parp->ar_pro != htons(ETH_P_IP) ||
1174 parp->ar_op != htons(ARPOP_REQUEST) ||
1175 parp->ar_hln != dev->addr_len ||
1178 arpptr = (u8 *)parp + sizeof(struct arphdr);
1180 arpptr += dev->addr_len; /* sha */
1181 memcpy(&sip, arpptr, sizeof(sip));
1182 arpptr += sizeof(sip);
1183 arpptr += dev->addr_len; /* tha */
1184 memcpy(&tip, arpptr, sizeof(tip));
1186 if (ipv4_is_loopback(tip) ||
1187 ipv4_is_multicast(tip))
1190 n = neigh_lookup(&arp_tbl, &tip, dev);
1193 struct vxlan_fdb *f;
1194 struct sk_buff *reply;
1196 if (!(n->nud_state & NUD_CONNECTED)) {
1201 f = vxlan_find_mac(vxlan, n->ha);
1202 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1203 /* bridge-local neighbor */
1208 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1213 skb_reset_mac_header(reply);
1214 __skb_pull(reply, skb_network_offset(reply));
1215 reply->ip_summed = CHECKSUM_UNNECESSARY;
1216 reply->pkt_type = PACKET_HOST;
1218 if (netif_rx_ni(reply) == NET_RX_DROP)
1219 dev->stats.rx_dropped++;
1220 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1221 union vxlan_addr ipa = {
1222 .sin.sin_addr.s_addr = tip,
1223 .sa.sa_family = AF_INET,
1226 vxlan_ip_miss(dev, &ipa);
1230 return NETDEV_TX_OK;
1233 #if IS_ENABLED(CONFIG_IPV6)
1234 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1236 struct vxlan_dev *vxlan = netdev_priv(dev);
1237 struct neighbour *n;
1238 union vxlan_addr ipa;
1239 const struct ipv6hdr *iphdr;
1240 const struct in6_addr *saddr, *daddr;
1242 struct inet6_dev *in6_dev = NULL;
1244 in6_dev = __in6_dev_get(dev);
1248 if (!pskb_may_pull(skb, skb->len))
1251 iphdr = ipv6_hdr(skb);
1252 saddr = &iphdr->saddr;
1253 daddr = &iphdr->daddr;
1255 if (ipv6_addr_loopback(daddr) ||
1256 ipv6_addr_is_multicast(daddr))
1259 msg = (struct nd_msg *)skb_transport_header(skb);
1260 if (msg->icmph.icmp6_code != 0 ||
1261 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1264 n = neigh_lookup(ipv6_stub->nd_tbl, daddr, dev);
1267 struct vxlan_fdb *f;
1269 if (!(n->nud_state & NUD_CONNECTED)) {
1274 f = vxlan_find_mac(vxlan, n->ha);
1275 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1276 /* bridge-local neighbor */
1281 ipv6_stub->ndisc_send_na(dev, n, saddr, &msg->target,
1282 !!in6_dev->cnf.forwarding,
1283 true, false, false);
1285 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1286 ipa.sin6.sin6_addr = *daddr;
1287 ipa.sa.sa_family = AF_INET6;
1288 vxlan_ip_miss(dev, &ipa);
1293 return NETDEV_TX_OK;
1297 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1299 struct vxlan_dev *vxlan = netdev_priv(dev);
1300 struct neighbour *n;
1302 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1306 switch (ntohs(eth_hdr(skb)->h_proto)) {
1311 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1314 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1315 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1316 union vxlan_addr ipa = {
1317 .sin.sin_addr.s_addr = pip->daddr,
1318 .sa.sa_family = AF_INET,
1321 vxlan_ip_miss(dev, &ipa);
1327 #if IS_ENABLED(CONFIG_IPV6)
1330 struct ipv6hdr *pip6;
1332 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1334 pip6 = ipv6_hdr(skb);
1335 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1336 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1337 union vxlan_addr ipa = {
1338 .sin6.sin6_addr = pip6->daddr,
1339 .sa.sa_family = AF_INET6,
1342 vxlan_ip_miss(dev, &ipa);
1356 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1358 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1360 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1369 static void vxlan_sock_put(struct sk_buff *skb)
1374 /* On transmit, associate with the tunnel socket */
1375 static void vxlan_set_owner(struct sock *sk, struct sk_buff *skb)
1380 skb->destructor = vxlan_sock_put;
1383 /* Compute source port for outgoing packet
1384 * first choice to use L4 flow hash since it will spread
1385 * better and maybe available from hardware
1386 * secondary choice is to use jhash on the Ethernet header
1388 __be16 vxlan_src_port(__u16 port_min, __u16 port_max, struct sk_buff *skb)
1390 unsigned int range = (port_max - port_min) + 1;
1393 hash = skb_get_rxhash(skb);
1395 hash = jhash(skb->data, 2 * ETH_ALEN,
1396 (__force u32) skb->protocol);
1398 return htons((((u64) hash * range) >> 32) + port_min);
1400 EXPORT_SYMBOL_GPL(vxlan_src_port);
1402 static int handle_offloads(struct sk_buff *skb)
1404 if (skb_is_gso(skb)) {
1405 int err = skb_unclone(skb, GFP_ATOMIC);
1409 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
1410 } else if (skb->ip_summed != CHECKSUM_PARTIAL)
1411 skb->ip_summed = CHECKSUM_NONE;
1416 #if IS_ENABLED(CONFIG_IPV6)
1417 static int vxlan6_xmit_skb(struct vxlan_sock *vs,
1418 struct dst_entry *dst, struct sk_buff *skb,
1419 struct net_device *dev, struct in6_addr *saddr,
1420 struct in6_addr *daddr, __u8 prio, __u8 ttl,
1421 __be16 src_port, __be16 dst_port, __be32 vni)
1423 struct ipv6hdr *ip6h;
1424 struct vxlanhdr *vxh;
1429 if (!skb->encapsulation) {
1430 skb_reset_inner_headers(skb);
1431 skb->encapsulation = 1;
1434 skb_scrub_packet(skb, false);
1436 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1437 + VXLAN_HLEN + sizeof(struct ipv6hdr)
1438 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1440 /* Need space for new headers (invalidates iph ptr) */
1441 err = skb_cow_head(skb, min_headroom);
1445 if (vlan_tx_tag_present(skb)) {
1446 if (WARN_ON(!__vlan_put_tag(skb,
1448 vlan_tx_tag_get(skb))))
1454 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1455 vxh->vx_flags = htonl(VXLAN_FLAGS);
1458 __skb_push(skb, sizeof(*uh));
1459 skb_reset_transport_header(skb);
1462 uh->dest = dst_port;
1463 uh->source = src_port;
1465 uh->len = htons(skb->len);
1468 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1469 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1471 skb_dst_set(skb, dst);
1473 if (!skb_is_gso(skb) && !(dst->dev->features & NETIF_F_IPV6_CSUM)) {
1474 __wsum csum = skb_checksum(skb, 0, skb->len, 0);
1475 skb->ip_summed = CHECKSUM_UNNECESSARY;
1476 uh->check = csum_ipv6_magic(saddr, daddr, skb->len,
1479 uh->check = CSUM_MANGLED_0;
1481 skb->ip_summed = CHECKSUM_PARTIAL;
1482 skb->csum_start = skb_transport_header(skb) - skb->head;
1483 skb->csum_offset = offsetof(struct udphdr, check);
1484 uh->check = ~csum_ipv6_magic(saddr, daddr,
1485 skb->len, IPPROTO_UDP, 0);
1488 __skb_push(skb, sizeof(*ip6h));
1489 skb_reset_network_header(skb);
1490 ip6h = ipv6_hdr(skb);
1492 ip6h->priority = prio;
1493 ip6h->flow_lbl[0] = 0;
1494 ip6h->flow_lbl[1] = 0;
1495 ip6h->flow_lbl[2] = 0;
1496 ip6h->payload_len = htons(skb->len);
1497 ip6h->nexthdr = IPPROTO_UDP;
1498 ip6h->hop_limit = ttl;
1499 ip6h->daddr = *daddr;
1500 ip6h->saddr = *saddr;
1502 vxlan_set_owner(vs->sock->sk, skb);
1504 err = handle_offloads(skb);
1508 ip6tunnel_xmit(skb, dev);
1513 int vxlan_xmit_skb(struct vxlan_sock *vs,
1514 struct rtable *rt, struct sk_buff *skb,
1515 __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1516 __be16 src_port, __be16 dst_port, __be32 vni)
1518 struct vxlanhdr *vxh;
1523 if (!skb->encapsulation) {
1524 skb_reset_inner_headers(skb);
1525 skb->encapsulation = 1;
1528 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1529 + VXLAN_HLEN + sizeof(struct iphdr)
1530 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1532 /* Need space for new headers (invalidates iph ptr) */
1533 err = skb_cow_head(skb, min_headroom);
1537 if (vlan_tx_tag_present(skb)) {
1538 if (WARN_ON(!__vlan_put_tag(skb,
1540 vlan_tx_tag_get(skb))))
1546 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1547 vxh->vx_flags = htonl(VXLAN_FLAGS);
1550 __skb_push(skb, sizeof(*uh));
1551 skb_reset_transport_header(skb);
1554 uh->dest = dst_port;
1555 uh->source = src_port;
1557 uh->len = htons(skb->len);
1560 vxlan_set_owner(vs->sock->sk, skb);
1562 err = handle_offloads(skb);
1566 return iptunnel_xmit(rt, skb, src, dst, IPPROTO_UDP, tos, ttl, df,
1569 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1571 /* Bypass encapsulation if the destination is local */
1572 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1573 struct vxlan_dev *dst_vxlan)
1575 struct pcpu_tstats *tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1576 struct pcpu_tstats *rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1577 union vxlan_addr loopback;
1578 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1580 skb->pkt_type = PACKET_HOST;
1581 skb->encapsulation = 0;
1582 skb->dev = dst_vxlan->dev;
1583 __skb_pull(skb, skb_network_offset(skb));
1585 if (remote_ip->sa.sa_family == AF_INET) {
1586 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1587 loopback.sa.sa_family = AF_INET;
1588 #if IS_ENABLED(CONFIG_IPV6)
1590 loopback.sin6.sin6_addr = in6addr_loopback;
1591 loopback.sa.sa_family = AF_INET6;
1595 if (dst_vxlan->flags & VXLAN_F_LEARN)
1596 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1598 u64_stats_update_begin(&tx_stats->syncp);
1599 tx_stats->tx_packets++;
1600 tx_stats->tx_bytes += skb->len;
1601 u64_stats_update_end(&tx_stats->syncp);
1603 if (netif_rx(skb) == NET_RX_SUCCESS) {
1604 u64_stats_update_begin(&rx_stats->syncp);
1605 rx_stats->rx_packets++;
1606 rx_stats->rx_bytes += skb->len;
1607 u64_stats_update_end(&rx_stats->syncp);
1609 skb->dev->stats.rx_dropped++;
1613 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1614 struct vxlan_rdst *rdst, bool did_rsc)
1616 struct vxlan_dev *vxlan = netdev_priv(dev);
1617 struct rtable *rt = NULL;
1618 const struct iphdr *old_iph;
1620 union vxlan_addr *dst;
1621 __be16 src_port = 0, dst_port;
1627 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1628 vni = rdst->remote_vni;
1629 dst = &rdst->remote_ip;
1631 if (vxlan_addr_any(dst)) {
1633 /* short-circuited back to local bridge */
1634 vxlan_encap_bypass(skb, vxlan, vxlan);
1640 old_iph = ip_hdr(skb);
1643 if (!ttl && vxlan_addr_multicast(dst))
1648 tos = ip_tunnel_get_dsfield(old_iph, skb);
1650 src_port = vxlan_src_port(vxlan->port_min, vxlan->port_max, skb);
1652 if (dst->sa.sa_family == AF_INET) {
1653 memset(&fl4, 0, sizeof(fl4));
1654 fl4.flowi4_oif = rdst->remote_ifindex;
1655 fl4.flowi4_tos = RT_TOS(tos);
1656 fl4.daddr = dst->sin.sin_addr.s_addr;
1657 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1659 rt = ip_route_output_key(dev_net(dev), &fl4);
1661 netdev_dbg(dev, "no route to %pI4\n",
1662 &dst->sin.sin_addr.s_addr);
1663 dev->stats.tx_carrier_errors++;
1667 if (rt->dst.dev == dev) {
1668 netdev_dbg(dev, "circular route to %pI4\n",
1669 &dst->sin.sin_addr.s_addr);
1670 dev->stats.collisions++;
1674 /* Bypass encapsulation if the destination is local */
1675 if (rt->rt_flags & RTCF_LOCAL &&
1676 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1677 struct vxlan_dev *dst_vxlan;
1680 dst_vxlan = vxlan_find_vni(dev_net(dev), vni, dst_port);
1683 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1687 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1688 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1690 err = vxlan_xmit_skb(vxlan->vn_sock, rt, skb,
1691 fl4.saddr, dst->sin.sin_addr.s_addr,
1692 tos, ttl, df, src_port, dst_port,
1697 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1698 #if IS_ENABLED(CONFIG_IPV6)
1700 struct sock *sk = vxlan->vn_sock->sock->sk;
1701 struct dst_entry *ndst;
1705 memset(&fl6, 0, sizeof(fl6));
1706 fl6.flowi6_oif = rdst->remote_ifindex;
1707 fl6.daddr = dst->sin6.sin6_addr;
1708 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
1709 fl6.flowi6_proto = IPPROTO_UDP;
1711 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
1712 netdev_dbg(dev, "no route to %pI6\n",
1713 &dst->sin6.sin6_addr);
1714 dev->stats.tx_carrier_errors++;
1718 if (ndst->dev == dev) {
1719 netdev_dbg(dev, "circular route to %pI6\n",
1720 &dst->sin6.sin6_addr);
1722 dev->stats.collisions++;
1726 /* Bypass encapsulation if the destination is local */
1727 flags = ((struct rt6_info *)ndst)->rt6i_flags;
1728 if (flags & RTF_LOCAL &&
1729 !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1730 struct vxlan_dev *dst_vxlan;
1733 dst_vxlan = vxlan_find_vni(dev_net(dev), vni, dst_port);
1736 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1740 ttl = ttl ? : ip6_dst_hoplimit(ndst);
1742 err = vxlan6_xmit_skb(vxlan->vn_sock, ndst, skb,
1743 dev, &fl6.saddr, &fl6.daddr, 0, ttl,
1744 src_port, dst_port, htonl(vni << 8));
1751 dev->stats.tx_dropped++;
1757 dev->stats.tx_errors++;
1762 /* Transmit local packets over Vxlan
1764 * Outer IP header inherits ECN and DF from inner header.
1765 * Outer UDP destination is the VXLAN assigned port.
1766 * source port is based on hash of flow
1768 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1770 struct vxlan_dev *vxlan = netdev_priv(dev);
1772 bool did_rsc = false;
1773 struct vxlan_rdst *rdst;
1774 struct vxlan_fdb *f;
1776 skb_reset_mac_header(skb);
1779 if ((vxlan->flags & VXLAN_F_PROXY)) {
1780 if (ntohs(eth->h_proto) == ETH_P_ARP)
1781 return arp_reduce(dev, skb);
1782 #if IS_ENABLED(CONFIG_IPV6)
1783 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
1784 skb->len >= sizeof(struct ipv6hdr) + sizeof(struct nd_msg) &&
1785 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
1788 msg = (struct nd_msg *)skb_transport_header(skb);
1789 if (msg->icmph.icmp6_code == 0 &&
1790 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
1791 return neigh_reduce(dev, skb);
1796 f = vxlan_find_mac(vxlan, eth->h_dest);
1799 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1800 (ntohs(eth->h_proto) == ETH_P_IP ||
1801 ntohs(eth->h_proto) == ETH_P_IPV6)) {
1802 did_rsc = route_shortcircuit(dev, skb);
1804 f = vxlan_find_mac(vxlan, eth->h_dest);
1808 f = vxlan_find_mac(vxlan, all_zeros_mac);
1810 if ((vxlan->flags & VXLAN_F_L2MISS) &&
1811 !is_multicast_ether_addr(eth->h_dest))
1812 vxlan_fdb_miss(vxlan, eth->h_dest);
1814 dev->stats.tx_dropped++;
1816 return NETDEV_TX_OK;
1820 list_for_each_entry_rcu(rdst, &f->remotes, list) {
1821 struct sk_buff *skb1;
1823 skb1 = skb_clone(skb, GFP_ATOMIC);
1825 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1829 return NETDEV_TX_OK;
1832 /* Walk the forwarding table and purge stale entries */
1833 static void vxlan_cleanup(unsigned long arg)
1835 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1836 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1839 if (!netif_running(vxlan->dev))
1842 spin_lock_bh(&vxlan->hash_lock);
1843 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1844 struct hlist_node *p, *n;
1845 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1847 = container_of(p, struct vxlan_fdb, hlist);
1848 unsigned long timeout;
1850 if (f->state & NUD_PERMANENT)
1853 timeout = f->used + vxlan->age_interval * HZ;
1854 if (time_before_eq(timeout, jiffies)) {
1855 netdev_dbg(vxlan->dev,
1856 "garbage collect %pM\n",
1858 f->state = NUD_STALE;
1859 vxlan_fdb_destroy(vxlan, f);
1860 } else if (time_before(timeout, next_timer))
1861 next_timer = timeout;
1864 spin_unlock_bh(&vxlan->hash_lock);
1866 mod_timer(&vxlan->age_timer, next_timer);
1869 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
1871 __u32 vni = vxlan->default_dst.remote_vni;
1873 vxlan->vn_sock = vs;
1874 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
1877 /* Setup stats when device is created */
1878 static int vxlan_init(struct net_device *dev)
1880 struct vxlan_dev *vxlan = netdev_priv(dev);
1881 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
1882 struct vxlan_sock *vs;
1885 dev->tstats = alloc_percpu(struct pcpu_tstats);
1889 for_each_possible_cpu(i) {
1890 struct pcpu_tstats *vxlan_stats;
1891 vxlan_stats = per_cpu_ptr(dev->tstats, i);
1892 u64_stats_init(&vxlan_stats->syncp);
1896 spin_lock(&vn->sock_lock);
1897 vs = vxlan_find_sock(dev_net(dev), vxlan->dst_port);
1899 /* If we have a socket with same port already, reuse it */
1900 atomic_inc(&vs->refcnt);
1901 vxlan_vs_add_dev(vs, vxlan);
1903 /* otherwise make new socket outside of RTNL */
1905 queue_work(vxlan_wq, &vxlan->sock_work);
1907 spin_unlock(&vn->sock_lock);
1912 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
1914 struct vxlan_fdb *f;
1916 spin_lock_bh(&vxlan->hash_lock);
1917 f = __vxlan_find_mac(vxlan, all_zeros_mac);
1919 vxlan_fdb_destroy(vxlan, f);
1920 spin_unlock_bh(&vxlan->hash_lock);
1923 static void vxlan_uninit(struct net_device *dev)
1925 struct vxlan_dev *vxlan = netdev_priv(dev);
1926 struct vxlan_sock *vs = vxlan->vn_sock;
1928 vxlan_fdb_delete_default(vxlan);
1931 vxlan_sock_release(vs);
1932 free_percpu(dev->tstats);
1935 /* Start ageing timer and join group when device is brought up */
1936 static int vxlan_open(struct net_device *dev)
1938 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
1939 struct vxlan_dev *vxlan = netdev_priv(dev);
1940 struct vxlan_sock *vs = vxlan->vn_sock;
1942 /* socket hasn't been created */
1946 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
1947 vxlan_group_used(vn, &vxlan->default_dst.remote_ip)) {
1948 vxlan_sock_hold(vs);
1950 queue_work(vxlan_wq, &vxlan->igmp_join);
1953 if (vxlan->age_interval)
1954 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
1959 /* Purge the forwarding table */
1960 static void vxlan_flush(struct vxlan_dev *vxlan)
1964 spin_lock_bh(&vxlan->hash_lock);
1965 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1966 struct hlist_node *p, *n;
1967 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1969 = container_of(p, struct vxlan_fdb, hlist);
1970 /* the all_zeros_mac entry is deleted at vxlan_uninit */
1971 if (!is_zero_ether_addr(f->eth_addr))
1972 vxlan_fdb_destroy(vxlan, f);
1975 spin_unlock_bh(&vxlan->hash_lock);
1978 /* Cleanup timer and forwarding table on shutdown */
1979 static int vxlan_stop(struct net_device *dev)
1981 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
1982 struct vxlan_dev *vxlan = netdev_priv(dev);
1983 struct vxlan_sock *vs = vxlan->vn_sock;
1985 if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
1986 ! vxlan_group_used(vn, &vxlan->default_dst.remote_ip)) {
1987 vxlan_sock_hold(vs);
1989 queue_work(vxlan_wq, &vxlan->igmp_leave);
1992 del_timer_sync(&vxlan->age_timer);
1999 /* Stub, nothing needs to be done. */
2000 static void vxlan_set_multicast_list(struct net_device *dev)
2004 static const struct net_device_ops vxlan_netdev_ops = {
2005 .ndo_init = vxlan_init,
2006 .ndo_uninit = vxlan_uninit,
2007 .ndo_open = vxlan_open,
2008 .ndo_stop = vxlan_stop,
2009 .ndo_start_xmit = vxlan_xmit,
2010 .ndo_get_stats64 = ip_tunnel_get_stats64,
2011 .ndo_set_rx_mode = vxlan_set_multicast_list,
2012 .ndo_change_mtu = eth_change_mtu,
2013 .ndo_validate_addr = eth_validate_addr,
2014 .ndo_set_mac_address = eth_mac_addr,
2015 .ndo_fdb_add = vxlan_fdb_add,
2016 .ndo_fdb_del = vxlan_fdb_delete,
2017 .ndo_fdb_dump = vxlan_fdb_dump,
2020 /* Info for udev, that this is a virtual tunnel endpoint */
2021 static struct device_type vxlan_type = {
2025 /* Calls the ndo_add_vxlan_port of the caller in order to
2026 * supply the listening VXLAN udp ports. Callers are expected
2027 * to implement the ndo_add_vxlan_port.
2029 void vxlan_get_rx_port(struct net_device *dev)
2031 struct vxlan_sock *vs;
2032 struct net *net = dev_net(dev);
2033 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2034 sa_family_t sa_family;
2038 spin_lock(&vn->sock_lock);
2039 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2040 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2041 port = 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(dev_net(dev), &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 struct socket *create_v6_sock(struct net *net, __be16 port)
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().
2222 /* Disable multicast loopback */
2223 inet_sk(sk)->mc_loop = 0;
2229 static struct socket *create_v6_sock(struct net *net, __be16 port)
2231 return ERR_PTR(-EPFNOSUPPORT);
2235 static struct socket *create_v4_sock(struct net *net, __be16 port)
2238 struct socket *sock;
2239 struct sockaddr_in vxlan_addr = {
2240 .sin_family = AF_INET,
2241 .sin_addr.s_addr = htonl(INADDR_ANY),
2246 /* Create UDP socket for encapsulation receive. */
2247 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
2249 pr_debug("UDP socket create failed\n");
2253 /* Put in proper namespace */
2255 sk_change_net(sk, net);
2257 rc = kernel_bind(sock, (struct sockaddr *) &vxlan_addr,
2258 sizeof(vxlan_addr));
2260 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
2261 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
2262 sk_release_kernel(sk);
2266 /* Disable multicast loopback */
2267 inet_sk(sk)->mc_loop = 0;
2271 /* Create new listen socket if needed */
2272 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2273 vxlan_rcv_t *rcv, void *data, bool ipv6)
2275 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2276 struct vxlan_sock *vs;
2277 struct socket *sock;
2281 vs = kmalloc(sizeof(*vs), GFP_KERNEL);
2283 return ERR_PTR(-ENOMEM);
2285 for (h = 0; h < VNI_HASH_SIZE; ++h)
2286 INIT_HLIST_HEAD(&vs->vni_list[h]);
2288 INIT_WORK(&vs->del_work, vxlan_del_work);
2291 sock = create_v6_sock(net, port);
2293 sock = create_v4_sock(net, port);
2296 return ERR_CAST(sock);
2301 atomic_set(&vs->refcnt, 1);
2304 rcu_assign_sk_user_data(vs->sock->sk, vs);
2306 spin_lock(&vn->sock_lock);
2307 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2308 vxlan_notify_add_rx_port(sk);
2309 spin_unlock(&vn->sock_lock);
2311 /* Mark socket as an encapsulation socket. */
2312 udp_sk(sk)->encap_type = 1;
2313 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
2314 #if IS_ENABLED(CONFIG_IPV6)
2316 ipv6_stub->udpv6_encap_enable();
2324 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2325 vxlan_rcv_t *rcv, void *data,
2326 bool no_share, bool ipv6)
2328 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2329 struct vxlan_sock *vs;
2331 vs = vxlan_socket_create(net, port, rcv, data, ipv6);
2335 if (no_share) /* Return error if sharing is not allowed. */
2338 spin_lock(&vn->sock_lock);
2339 vs = vxlan_find_sock(net, port);
2342 atomic_inc(&vs->refcnt);
2344 vs = ERR_PTR(-EBUSY);
2346 spin_unlock(&vn->sock_lock);
2349 vs = ERR_PTR(-EINVAL);
2353 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2355 /* Scheduled at device creation to bind to a socket */
2356 static void vxlan_sock_work(struct work_struct *work)
2358 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2359 struct net *net = dev_net(vxlan->dev);
2360 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2361 __be16 port = vxlan->dst_port;
2362 struct vxlan_sock *nvs;
2364 nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags & VXLAN_F_IPV6);
2365 spin_lock(&vn->sock_lock);
2367 vxlan_vs_add_dev(nvs, vxlan);
2368 spin_unlock(&vn->sock_lock);
2370 dev_put(vxlan->dev);
2373 static int vxlan_newlink(struct net *net, struct net_device *dev,
2374 struct nlattr *tb[], struct nlattr *data[])
2376 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2377 struct vxlan_dev *vxlan = netdev_priv(dev);
2378 struct vxlan_rdst *dst = &vxlan->default_dst;
2381 bool use_ipv6 = false;
2383 if (!data[IFLA_VXLAN_ID])
2386 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2387 dst->remote_vni = vni;
2389 if (data[IFLA_VXLAN_GROUP]) {
2390 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2391 dst->remote_ip.sa.sa_family = AF_INET;
2392 } else if (data[IFLA_VXLAN_GROUP6]) {
2393 if (!IS_ENABLED(CONFIG_IPV6))
2394 return -EPFNOSUPPORT;
2396 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2397 sizeof(struct in6_addr));
2398 dst->remote_ip.sa.sa_family = AF_INET6;
2402 if (data[IFLA_VXLAN_LOCAL]) {
2403 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2404 vxlan->saddr.sa.sa_family = AF_INET;
2405 } else if (data[IFLA_VXLAN_LOCAL6]) {
2406 if (!IS_ENABLED(CONFIG_IPV6))
2407 return -EPFNOSUPPORT;
2409 /* TODO: respect scope id */
2410 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2411 sizeof(struct in6_addr));
2412 vxlan->saddr.sa.sa_family = AF_INET6;
2416 if (data[IFLA_VXLAN_LINK] &&
2417 (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2418 struct net_device *lowerdev
2419 = __dev_get_by_index(net, dst->remote_ifindex);
2422 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2426 #if IS_ENABLED(CONFIG_IPV6)
2428 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2429 if (idev && idev->cnf.disable_ipv6) {
2430 pr_info("IPv6 is disabled via sysctl\n");
2433 vxlan->flags |= VXLAN_F_IPV6;
2438 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2440 /* update header length based on lower device */
2441 dev->hard_header_len = lowerdev->hard_header_len +
2442 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2445 if (data[IFLA_VXLAN_TOS])
2446 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2448 if (data[IFLA_VXLAN_TTL])
2449 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2451 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2452 vxlan->flags |= VXLAN_F_LEARN;
2454 if (data[IFLA_VXLAN_AGEING])
2455 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2457 vxlan->age_interval = FDB_AGE_DEFAULT;
2459 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2460 vxlan->flags |= VXLAN_F_PROXY;
2462 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2463 vxlan->flags |= VXLAN_F_RSC;
2465 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2466 vxlan->flags |= VXLAN_F_L2MISS;
2468 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2469 vxlan->flags |= VXLAN_F_L3MISS;
2471 if (data[IFLA_VXLAN_LIMIT])
2472 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2474 if (data[IFLA_VXLAN_PORT_RANGE]) {
2475 const struct ifla_vxlan_port_range *p
2476 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2477 vxlan->port_min = ntohs(p->low);
2478 vxlan->port_max = ntohs(p->high);
2481 if (data[IFLA_VXLAN_PORT])
2482 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2484 if (vxlan_find_vni(net, vni, vxlan->dst_port)) {
2485 pr_info("duplicate VNI %u\n", vni);
2489 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
2491 /* create an fdb entry for a valid default destination */
2492 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2493 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2494 &vxlan->default_dst.remote_ip,
2495 NUD_REACHABLE|NUD_PERMANENT,
2496 NLM_F_EXCL|NLM_F_CREATE,
2498 vxlan->default_dst.remote_vni,
2499 vxlan->default_dst.remote_ifindex,
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");