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/udp_tunnel.h>
37 #include <net/rtnetlink.h>
38 #include <net/route.h>
39 #include <net/dsfield.h>
40 #include <net/inet_ecn.h>
41 #include <net/net_namespace.h>
42 #include <net/netns/generic.h>
43 #include <net/vxlan.h>
44 #include <net/protocol.h>
46 #if IS_ENABLED(CONFIG_IPV6)
48 #include <net/addrconf.h>
49 #include <net/ip6_tunnel.h>
50 #include <net/ip6_checksum.h>
52 #include <net/dst_metadata.h>
54 #define VXLAN_VERSION "0.1"
56 #define PORT_HASH_BITS 8
57 #define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
58 #define FDB_AGE_DEFAULT 300 /* 5 min */
59 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
61 /* UDP port for VXLAN traffic.
62 * The IANA assigned port is 4789, but the Linux default is 8472
63 * for compatibility with early adopters.
65 static unsigned short vxlan_port __read_mostly = 8472;
66 module_param_named(udp_port, vxlan_port, ushort, 0444);
67 MODULE_PARM_DESC(udp_port, "Destination UDP port");
69 static bool log_ecn_error = true;
70 module_param(log_ecn_error, bool, 0644);
71 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
73 static int vxlan_net_id;
74 static struct rtnl_link_ops vxlan_link_ops;
76 static const u8 all_zeros_mac[ETH_ALEN + 2];
78 static int vxlan_sock_add(struct vxlan_dev *vxlan);
80 /* per-network namespace private data for this module */
82 struct list_head vxlan_list;
83 struct hlist_head sock_list[PORT_HASH_SIZE];
87 /* Forwarding table entry */
89 struct hlist_node hlist; /* linked list of entries */
91 unsigned long updated; /* jiffies */
93 struct list_head remotes;
94 u8 eth_addr[ETH_ALEN];
95 u16 state; /* see ndm_state */
96 u8 flags; /* see ndm_flags */
99 /* salt for hash table */
100 static u32 vxlan_salt __read_mostly;
101 static struct workqueue_struct *vxlan_wq;
103 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
105 return vs->flags & VXLAN_F_COLLECT_METADATA ||
106 ip_tunnel_collect_metadata();
109 #if IS_ENABLED(CONFIG_IPV6)
111 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
113 if (a->sa.sa_family != b->sa.sa_family)
115 if (a->sa.sa_family == AF_INET6)
116 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
118 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
121 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
123 if (ipa->sa.sa_family == AF_INET6)
124 return ipv6_addr_any(&ipa->sin6.sin6_addr);
126 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
129 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
131 if (ipa->sa.sa_family == AF_INET6)
132 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
134 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
137 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
139 if (nla_len(nla) >= sizeof(struct in6_addr)) {
140 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
141 ip->sa.sa_family = AF_INET6;
143 } else if (nla_len(nla) >= sizeof(__be32)) {
144 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
145 ip->sa.sa_family = AF_INET;
148 return -EAFNOSUPPORT;
152 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
153 const union vxlan_addr *ip)
155 if (ip->sa.sa_family == AF_INET6)
156 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
158 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
161 #else /* !CONFIG_IPV6 */
164 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
166 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
169 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
171 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
174 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
176 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
179 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
181 if (nla_len(nla) >= sizeof(struct in6_addr)) {
182 return -EAFNOSUPPORT;
183 } else if (nla_len(nla) >= sizeof(__be32)) {
184 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
185 ip->sa.sa_family = AF_INET;
188 return -EAFNOSUPPORT;
192 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
193 const union vxlan_addr *ip)
195 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
199 /* Virtual Network hash table head */
200 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
202 return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
205 /* Socket hash table head */
206 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
208 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
210 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
213 /* First remote destination for a forwarding entry.
214 * Guaranteed to be non-NULL because remotes are never deleted.
216 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
218 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
221 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
223 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
226 /* Find VXLAN socket based on network namespace, address family and UDP port
227 * and enabled unshareable flags.
229 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
230 __be16 port, u32 flags)
232 struct vxlan_sock *vs;
234 flags &= VXLAN_F_RCV_FLAGS;
236 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
237 if (inet_sk(vs->sock->sk)->inet_sport == port &&
238 vxlan_get_sk_family(vs) == family &&
245 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
247 struct vxlan_dev *vxlan;
249 /* For flow based devices, map all packets to VNI 0 */
250 if (vs->flags & VXLAN_F_COLLECT_METADATA)
253 hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
254 if (vxlan->default_dst.remote_vni == vni)
261 /* Look up VNI in a per net namespace table */
262 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
263 sa_family_t family, __be16 port,
266 struct vxlan_sock *vs;
268 vs = vxlan_find_sock(net, family, port, flags);
272 return vxlan_vs_find_vni(vs, vni);
275 /* Fill in neighbour message in skbuff. */
276 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
277 const struct vxlan_fdb *fdb,
278 u32 portid, u32 seq, int type, unsigned int flags,
279 const struct vxlan_rdst *rdst)
281 unsigned long now = jiffies;
282 struct nda_cacheinfo ci;
283 struct nlmsghdr *nlh;
285 bool send_ip, send_eth;
287 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
291 ndm = nlmsg_data(nlh);
292 memset(ndm, 0, sizeof(*ndm));
294 send_eth = send_ip = true;
296 if (type == RTM_GETNEIGH) {
297 ndm->ndm_family = AF_INET;
298 send_ip = !vxlan_addr_any(&rdst->remote_ip);
299 send_eth = !is_zero_ether_addr(fdb->eth_addr);
301 ndm->ndm_family = AF_BRIDGE;
302 ndm->ndm_state = fdb->state;
303 ndm->ndm_ifindex = vxlan->dev->ifindex;
304 ndm->ndm_flags = fdb->flags;
305 ndm->ndm_type = RTN_UNICAST;
307 if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
308 nla_put_s32(skb, NDA_LINK_NETNSID,
309 peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
310 goto nla_put_failure;
312 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
313 goto nla_put_failure;
315 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
316 goto nla_put_failure;
318 if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
319 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
320 goto nla_put_failure;
321 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
322 nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
323 goto nla_put_failure;
324 if (rdst->remote_ifindex &&
325 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
326 goto nla_put_failure;
328 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
329 ci.ndm_confirmed = 0;
330 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
333 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
334 goto nla_put_failure;
340 nlmsg_cancel(skb, nlh);
344 static inline size_t vxlan_nlmsg_size(void)
346 return NLMSG_ALIGN(sizeof(struct ndmsg))
347 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
348 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
349 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
350 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
351 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
352 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
353 + nla_total_size(sizeof(struct nda_cacheinfo));
356 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
357 struct vxlan_rdst *rd, int type)
359 struct net *net = dev_net(vxlan->dev);
363 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
367 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
369 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
370 WARN_ON(err == -EMSGSIZE);
375 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
379 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
382 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
384 struct vxlan_dev *vxlan = netdev_priv(dev);
385 struct vxlan_fdb f = {
388 struct vxlan_rdst remote = {
389 .remote_ip = *ipa, /* goes to NDA_DST */
390 .remote_vni = cpu_to_be32(VXLAN_N_VID),
393 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
396 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
398 struct vxlan_fdb f = {
401 struct vxlan_rdst remote = { };
403 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
405 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
408 /* Hash Ethernet address */
409 static u32 eth_hash(const unsigned char *addr)
411 u64 value = get_unaligned((u64 *)addr);
413 /* only want 6 bytes */
419 return hash_64(value, FDB_HASH_BITS);
422 /* Hash chain to use given mac address */
423 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
426 return &vxlan->fdb_head[eth_hash(mac)];
429 /* Look up Ethernet address in forwarding table */
430 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
433 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
436 hlist_for_each_entry_rcu(f, head, hlist) {
437 if (ether_addr_equal(mac, f->eth_addr))
444 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
449 f = __vxlan_find_mac(vxlan, mac);
456 /* caller should hold vxlan->hash_lock */
457 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
458 union vxlan_addr *ip, __be16 port,
459 __be32 vni, __u32 ifindex)
461 struct vxlan_rdst *rd;
463 list_for_each_entry(rd, &f->remotes, list) {
464 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
465 rd->remote_port == port &&
466 rd->remote_vni == vni &&
467 rd->remote_ifindex == ifindex)
474 /* Replace destination of unicast mac */
475 static int vxlan_fdb_replace(struct vxlan_fdb *f,
476 union vxlan_addr *ip, __be16 port, __be32 vni,
479 struct vxlan_rdst *rd;
481 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
485 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
489 dst_cache_reset(&rd->dst_cache);
491 rd->remote_port = port;
492 rd->remote_vni = vni;
493 rd->remote_ifindex = ifindex;
497 /* Add/update destinations for multicast */
498 static int vxlan_fdb_append(struct vxlan_fdb *f,
499 union vxlan_addr *ip, __be16 port, __be32 vni,
500 __u32 ifindex, struct vxlan_rdst **rdp)
502 struct vxlan_rdst *rd;
504 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
508 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
512 if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
518 rd->remote_port = port;
519 rd->remote_vni = vni;
520 rd->remote_ifindex = ifindex;
522 list_add_tail_rcu(&rd->list, &f->remotes);
528 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
530 struct vxlanhdr *vh, size_t hdrlen,
532 struct gro_remcsum *grc,
535 size_t start, offset;
537 if (skb->remcsum_offload)
540 if (!NAPI_GRO_CB(skb)->csum_valid)
543 start = vxlan_rco_start(vni_field);
544 offset = start + vxlan_rco_offset(vni_field);
546 vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
547 start, offset, grc, nopartial);
549 skb->remcsum_offload = 1;
554 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
555 struct sk_buff **head,
558 struct sk_buff *p, **pp = NULL;
559 struct vxlanhdr *vh, *vh2;
560 unsigned int hlen, off_vx;
562 struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
564 struct gro_remcsum grc;
566 skb_gro_remcsum_init(&grc);
568 off_vx = skb_gro_offset(skb);
569 hlen = off_vx + sizeof(*vh);
570 vh = skb_gro_header_fast(skb, off_vx);
571 if (skb_gro_header_hard(skb, hlen)) {
572 vh = skb_gro_header_slow(skb, hlen, off_vx);
577 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
579 flags = vh->vx_flags;
581 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
582 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
585 VXLAN_F_REMCSUM_NOPARTIAL));
591 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
593 for (p = *head; p; p = p->next) {
594 if (!NAPI_GRO_CB(p)->same_flow)
597 vh2 = (struct vxlanhdr *)(p->data + off_vx);
598 if (vh->vx_flags != vh2->vx_flags ||
599 vh->vx_vni != vh2->vx_vni) {
600 NAPI_GRO_CB(p)->same_flow = 0;
605 pp = eth_gro_receive(head, skb);
609 skb_gro_remcsum_cleanup(skb, &grc);
610 NAPI_GRO_CB(skb)->flush |= flush;
615 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
617 udp_tunnel_gro_complete(skb, nhoff);
619 return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
622 /* Notify netdevs that UDP port started listening */
623 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
625 struct net_device *dev;
626 struct sock *sk = vs->sock->sk;
627 struct net *net = sock_net(sk);
628 sa_family_t sa_family = vxlan_get_sk_family(vs);
629 __be16 port = inet_sk(sk)->inet_sport;
632 for_each_netdev_rcu(net, dev) {
633 if (dev->netdev_ops->ndo_add_vxlan_port)
634 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
640 /* Notify netdevs that UDP port is no more listening */
641 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
643 struct net_device *dev;
644 struct sock *sk = vs->sock->sk;
645 struct net *net = sock_net(sk);
646 sa_family_t sa_family = vxlan_get_sk_family(vs);
647 __be16 port = inet_sk(sk)->inet_sport;
650 for_each_netdev_rcu(net, dev) {
651 if (dev->netdev_ops->ndo_del_vxlan_port)
652 dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
658 /* Add new entry to forwarding table -- assumes lock held */
659 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
660 const u8 *mac, union vxlan_addr *ip,
661 __u16 state, __u16 flags,
662 __be16 port, __be32 vni, __u32 ifindex,
665 struct vxlan_rdst *rd = NULL;
669 f = __vxlan_find_mac(vxlan, mac);
671 if (flags & NLM_F_EXCL) {
672 netdev_dbg(vxlan->dev,
673 "lost race to create %pM\n", mac);
676 if (f->state != state) {
678 f->updated = jiffies;
681 if (f->flags != ndm_flags) {
682 f->flags = ndm_flags;
683 f->updated = jiffies;
686 if ((flags & NLM_F_REPLACE)) {
687 /* Only change unicasts */
688 if (!(is_multicast_ether_addr(f->eth_addr) ||
689 is_zero_ether_addr(f->eth_addr))) {
690 notify |= vxlan_fdb_replace(f, ip, port, vni,
695 if ((flags & NLM_F_APPEND) &&
696 (is_multicast_ether_addr(f->eth_addr) ||
697 is_zero_ether_addr(f->eth_addr))) {
698 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
706 if (!(flags & NLM_F_CREATE))
709 if (vxlan->cfg.addrmax &&
710 vxlan->addrcnt >= vxlan->cfg.addrmax)
713 /* Disallow replace to add a multicast entry */
714 if ((flags & NLM_F_REPLACE) &&
715 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
718 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
719 f = kmalloc(sizeof(*f), GFP_ATOMIC);
725 f->flags = ndm_flags;
726 f->updated = f->used = jiffies;
727 INIT_LIST_HEAD(&f->remotes);
728 memcpy(f->eth_addr, mac, ETH_ALEN);
730 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
733 hlist_add_head_rcu(&f->hlist,
734 vxlan_fdb_head(vxlan, mac));
739 rd = first_remote_rtnl(f);
740 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
746 static void vxlan_fdb_free(struct rcu_head *head)
748 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
749 struct vxlan_rdst *rd, *nd;
751 list_for_each_entry_safe(rd, nd, &f->remotes, list) {
752 dst_cache_destroy(&rd->dst_cache);
758 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
760 netdev_dbg(vxlan->dev,
761 "delete %pM\n", f->eth_addr);
764 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
766 hlist_del_rcu(&f->hlist);
767 call_rcu(&f->rcu, vxlan_fdb_free);
770 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
771 union vxlan_addr *ip, __be16 *port, __be32 *vni,
774 struct net *net = dev_net(vxlan->dev);
778 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
782 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
783 if (remote->sa.sa_family == AF_INET) {
784 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
785 ip->sa.sa_family = AF_INET;
786 #if IS_ENABLED(CONFIG_IPV6)
788 ip->sin6.sin6_addr = in6addr_any;
789 ip->sa.sa_family = AF_INET6;
795 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
797 *port = nla_get_be16(tb[NDA_PORT]);
799 *port = vxlan->cfg.dst_port;
803 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
805 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
807 *vni = vxlan->default_dst.remote_vni;
810 if (tb[NDA_IFINDEX]) {
811 struct net_device *tdev;
813 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
815 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
816 tdev = __dev_get_by_index(net, *ifindex);
818 return -EADDRNOTAVAIL;
826 /* Add static entry (via netlink) */
827 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
828 struct net_device *dev,
829 const unsigned char *addr, u16 vid, u16 flags)
831 struct vxlan_dev *vxlan = netdev_priv(dev);
832 /* struct net *net = dev_net(vxlan->dev); */
839 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
840 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
845 if (tb[NDA_DST] == NULL)
848 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
852 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
853 return -EAFNOSUPPORT;
855 spin_lock_bh(&vxlan->hash_lock);
856 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
857 port, vni, ifindex, ndm->ndm_flags);
858 spin_unlock_bh(&vxlan->hash_lock);
863 /* Delete entry (via netlink) */
864 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
865 struct net_device *dev,
866 const unsigned char *addr, u16 vid)
868 struct vxlan_dev *vxlan = netdev_priv(dev);
870 struct vxlan_rdst *rd = NULL;
877 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
883 spin_lock_bh(&vxlan->hash_lock);
884 f = vxlan_find_mac(vxlan, addr);
888 if (!vxlan_addr_any(&ip)) {
889 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
896 /* remove a destination if it's not the only one on the list,
897 * otherwise destroy the fdb entry
899 if (rd && !list_is_singular(&f->remotes)) {
900 list_del_rcu(&rd->list);
901 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
906 vxlan_fdb_destroy(vxlan, f);
909 spin_unlock_bh(&vxlan->hash_lock);
914 /* Dump forwarding table */
915 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
916 struct net_device *dev,
917 struct net_device *filter_dev, int idx)
919 struct vxlan_dev *vxlan = netdev_priv(dev);
922 for (h = 0; h < FDB_HASH_SIZE; ++h) {
926 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
927 struct vxlan_rdst *rd;
929 list_for_each_entry_rcu(rd, &f->remotes, list) {
930 if (idx < cb->args[0])
933 err = vxlan_fdb_info(skb, vxlan, f,
934 NETLINK_CB(cb->skb).portid,
951 /* Watch incoming packets to learn mapping between Ethernet address
952 * and Tunnel endpoint.
953 * Return true if packet is bogus and should be dropped.
955 static bool vxlan_snoop(struct net_device *dev,
956 union vxlan_addr *src_ip, const u8 *src_mac)
958 struct vxlan_dev *vxlan = netdev_priv(dev);
961 f = vxlan_find_mac(vxlan, src_mac);
963 struct vxlan_rdst *rdst = first_remote_rcu(f);
965 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
968 /* Don't migrate static entries, drop packets */
969 if (f->state & NUD_NOARP)
974 "%pM migrated from %pIS to %pIS\n",
975 src_mac, &rdst->remote_ip.sa, &src_ip->sa);
977 rdst->remote_ip = *src_ip;
978 f->updated = jiffies;
979 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
981 /* learned new entry */
982 spin_lock(&vxlan->hash_lock);
984 /* close off race between vxlan_flush and incoming packets */
985 if (netif_running(dev))
986 vxlan_fdb_create(vxlan, src_mac, src_ip,
988 NLM_F_EXCL|NLM_F_CREATE,
990 vxlan->default_dst.remote_vni,
992 spin_unlock(&vxlan->hash_lock);
998 /* See if multicast group is already in use by other ID */
999 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1001 struct vxlan_dev *vxlan;
1002 unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1004 /* The vxlan_sock is only used by dev, leaving group has
1005 * no effect on other vxlan devices.
1007 if (family == AF_INET && dev->vn4_sock &&
1008 atomic_read(&dev->vn4_sock->refcnt) == 1)
1010 #if IS_ENABLED(CONFIG_IPV6)
1011 if (family == AF_INET6 && dev->vn6_sock &&
1012 atomic_read(&dev->vn6_sock->refcnt) == 1)
1016 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1017 if (!netif_running(vxlan->dev) || vxlan == dev)
1020 if (family == AF_INET && vxlan->vn4_sock != dev->vn4_sock)
1022 #if IS_ENABLED(CONFIG_IPV6)
1023 if (family == AF_INET6 && vxlan->vn6_sock != dev->vn6_sock)
1027 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1028 &dev->default_dst.remote_ip))
1031 if (vxlan->default_dst.remote_ifindex !=
1032 dev->default_dst.remote_ifindex)
1041 static void __vxlan_sock_release(struct vxlan_sock *vs)
1043 struct vxlan_net *vn;
1047 if (!atomic_dec_and_test(&vs->refcnt))
1050 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1051 spin_lock(&vn->sock_lock);
1052 hlist_del_rcu(&vs->hlist);
1053 vxlan_notify_del_rx_port(vs);
1054 spin_unlock(&vn->sock_lock);
1056 queue_work(vxlan_wq, &vs->del_work);
1059 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1061 __vxlan_sock_release(vxlan->vn4_sock);
1062 #if IS_ENABLED(CONFIG_IPV6)
1063 __vxlan_sock_release(vxlan->vn6_sock);
1067 /* Update multicast group membership when first VNI on
1068 * multicast address is brought up
1070 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1073 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1074 int ifindex = vxlan->default_dst.remote_ifindex;
1077 if (ip->sa.sa_family == AF_INET) {
1078 struct ip_mreqn mreq = {
1079 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1080 .imr_ifindex = ifindex,
1083 sk = vxlan->vn4_sock->sock->sk;
1085 ret = ip_mc_join_group(sk, &mreq);
1087 #if IS_ENABLED(CONFIG_IPV6)
1089 sk = vxlan->vn6_sock->sock->sk;
1091 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1092 &ip->sin6.sin6_addr);
1100 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1101 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1104 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1105 int ifindex = vxlan->default_dst.remote_ifindex;
1108 if (ip->sa.sa_family == AF_INET) {
1109 struct ip_mreqn mreq = {
1110 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1111 .imr_ifindex = ifindex,
1114 sk = vxlan->vn4_sock->sock->sk;
1116 ret = ip_mc_leave_group(sk, &mreq);
1118 #if IS_ENABLED(CONFIG_IPV6)
1120 sk = vxlan->vn6_sock->sock->sk;
1122 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1123 &ip->sin6.sin6_addr);
1131 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1132 struct sk_buff *skb, u32 vxflags)
1134 size_t start, offset;
1136 if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1139 start = vxlan_rco_start(unparsed->vx_vni);
1140 offset = start + vxlan_rco_offset(unparsed->vx_vni);
1142 if (!pskb_may_pull(skb, offset + sizeof(u16)))
1145 skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1146 !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1148 unparsed->vx_flags &= ~VXLAN_HF_RCO;
1149 unparsed->vx_vni &= VXLAN_VNI_MASK;
1153 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1154 struct sk_buff *skb, u32 vxflags,
1155 struct vxlan_metadata *md)
1157 struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1158 struct metadata_dst *tun_dst;
1160 if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1163 md->gbp = ntohs(gbp->policy_id);
1165 tun_dst = (struct metadata_dst *)skb_dst(skb);
1167 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1168 tun_dst->u.tun_info.options_len = sizeof(*md);
1170 if (gbp->dont_learn)
1171 md->gbp |= VXLAN_GBP_DONT_LEARN;
1173 if (gbp->policy_applied)
1174 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1176 /* In flow-based mode, GBP is carried in dst_metadata */
1177 if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1178 skb->mark = md->gbp;
1180 unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1183 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1185 struct sk_buff *skb, u32 vxflags)
1187 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1189 /* Need to have Next Protocol set for interfaces in GPE mode. */
1190 if (!gpe->np_applied)
1192 /* "The initial version is 0. If a receiver does not support the
1193 * version indicated it MUST drop the packet.
1195 if (gpe->version != 0)
1197 /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1198 * processing MUST occur." However, we don't implement OAM
1199 * processing, thus drop the packet.
1204 switch (gpe->next_protocol) {
1205 case VXLAN_GPE_NP_IPV4:
1206 *protocol = htons(ETH_P_IP);
1208 case VXLAN_GPE_NP_IPV6:
1209 *protocol = htons(ETH_P_IPV6);
1211 case VXLAN_GPE_NP_ETHERNET:
1212 *protocol = htons(ETH_P_TEB);
1218 unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1222 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1223 struct vxlan_sock *vs,
1224 struct sk_buff *skb)
1226 union vxlan_addr saddr;
1228 skb_reset_mac_header(skb);
1229 skb->protocol = eth_type_trans(skb, vxlan->dev);
1230 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1232 /* Ignore packet loops (and multicast echo) */
1233 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1236 /* Get address from the outer IP header */
1237 if (vxlan_get_sk_family(vs) == AF_INET) {
1238 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1239 saddr.sa.sa_family = AF_INET;
1240 #if IS_ENABLED(CONFIG_IPV6)
1242 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1243 saddr.sa.sa_family = AF_INET6;
1247 if ((vxlan->flags & VXLAN_F_LEARN) &&
1248 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1254 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1255 struct sk_buff *skb)
1259 if (vxlan_get_sk_family(vs) == AF_INET)
1260 err = IP_ECN_decapsulate(oiph, skb);
1261 #if IS_ENABLED(CONFIG_IPV6)
1263 err = IP6_ECN_decapsulate(oiph, skb);
1266 if (unlikely(err) && log_ecn_error) {
1267 if (vxlan_get_sk_family(vs) == AF_INET)
1268 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1269 &((struct iphdr *)oiph)->saddr,
1270 ((struct iphdr *)oiph)->tos);
1272 net_info_ratelimited("non-ECT from %pI6\n",
1273 &((struct ipv6hdr *)oiph)->saddr);
1278 /* Callback from net/ipv4/udp.c to receive packets */
1279 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1281 struct pcpu_sw_netstats *stats;
1282 struct vxlan_dev *vxlan;
1283 struct vxlan_sock *vs;
1284 struct vxlanhdr unparsed;
1285 struct vxlan_metadata _md;
1286 struct vxlan_metadata *md = &_md;
1287 __be32 protocol = htons(ETH_P_TEB);
1288 bool raw_proto = false;
1291 /* Need UDP and VXLAN header to be present */
1292 if (!pskb_may_pull(skb, VXLAN_HLEN))
1295 unparsed = *vxlan_hdr(skb);
1296 /* VNI flag always required to be set */
1297 if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1298 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1299 ntohl(vxlan_hdr(skb)->vx_flags),
1300 ntohl(vxlan_hdr(skb)->vx_vni));
1301 /* Return non vxlan pkt */
1304 unparsed.vx_flags &= ~VXLAN_HF_VNI;
1305 unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1307 vs = rcu_dereference_sk_user_data(sk);
1311 vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1315 /* For backwards compatibility, only allow reserved fields to be
1316 * used by VXLAN extensions if explicitly requested.
1318 if (vs->flags & VXLAN_F_GPE) {
1319 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1324 if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1325 !net_eq(vxlan->net, dev_net(vxlan->dev))))
1328 if (vxlan_collect_metadata(vs)) {
1329 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1330 struct metadata_dst *tun_dst;
1332 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1333 vxlan_vni_to_tun_id(vni), sizeof(*md));
1338 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1340 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1342 memset(md, 0, sizeof(*md));
1345 if (vs->flags & VXLAN_F_REMCSUM_RX)
1346 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1348 if (vs->flags & VXLAN_F_GBP)
1349 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1350 /* Note that GBP and GPE can never be active together. This is
1351 * ensured in vxlan_dev_configure.
1354 if (unparsed.vx_flags || unparsed.vx_vni) {
1355 /* If there are any unprocessed flags remaining treat
1356 * this as a malformed packet. This behavior diverges from
1357 * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1358 * in reserved fields are to be ignored. The approach here
1359 * maintains compatibility with previous stack code, and also
1360 * is more robust and provides a little more security in
1361 * adding extensions to VXLAN.
1367 if (!vxlan_set_mac(vxlan, vs, skb))
1370 skb->dev = vxlan->dev;
1371 skb->pkt_type = PACKET_HOST;
1374 oiph = skb_network_header(skb);
1375 skb_reset_network_header(skb);
1377 if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1378 ++vxlan->dev->stats.rx_frame_errors;
1379 ++vxlan->dev->stats.rx_errors;
1383 stats = this_cpu_ptr(vxlan->dev->tstats);
1384 u64_stats_update_begin(&stats->syncp);
1385 stats->rx_packets++;
1386 stats->rx_bytes += skb->len;
1387 u64_stats_update_end(&stats->syncp);
1389 gro_cells_receive(&vxlan->gro_cells, skb);
1393 /* Consume bad packet */
1398 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1400 struct vxlan_dev *vxlan = netdev_priv(dev);
1401 struct arphdr *parp;
1404 struct neighbour *n;
1406 if (dev->flags & IFF_NOARP)
1409 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1410 dev->stats.tx_dropped++;
1413 parp = arp_hdr(skb);
1415 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1416 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1417 parp->ar_pro != htons(ETH_P_IP) ||
1418 parp->ar_op != htons(ARPOP_REQUEST) ||
1419 parp->ar_hln != dev->addr_len ||
1422 arpptr = (u8 *)parp + sizeof(struct arphdr);
1424 arpptr += dev->addr_len; /* sha */
1425 memcpy(&sip, arpptr, sizeof(sip));
1426 arpptr += sizeof(sip);
1427 arpptr += dev->addr_len; /* tha */
1428 memcpy(&tip, arpptr, sizeof(tip));
1430 if (ipv4_is_loopback(tip) ||
1431 ipv4_is_multicast(tip))
1434 n = neigh_lookup(&arp_tbl, &tip, dev);
1437 struct vxlan_fdb *f;
1438 struct sk_buff *reply;
1440 if (!(n->nud_state & NUD_CONNECTED)) {
1445 f = vxlan_find_mac(vxlan, n->ha);
1446 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1447 /* bridge-local neighbor */
1452 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1460 skb_reset_mac_header(reply);
1461 __skb_pull(reply, skb_network_offset(reply));
1462 reply->ip_summed = CHECKSUM_UNNECESSARY;
1463 reply->pkt_type = PACKET_HOST;
1465 if (netif_rx_ni(reply) == NET_RX_DROP)
1466 dev->stats.rx_dropped++;
1467 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1468 union vxlan_addr ipa = {
1469 .sin.sin_addr.s_addr = tip,
1470 .sin.sin_family = AF_INET,
1473 vxlan_ip_miss(dev, &ipa);
1477 return NETDEV_TX_OK;
1480 #if IS_ENABLED(CONFIG_IPV6)
1481 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1482 struct neighbour *n, bool isrouter)
1484 struct net_device *dev = request->dev;
1485 struct sk_buff *reply;
1486 struct nd_msg *ns, *na;
1487 struct ipv6hdr *pip6;
1489 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1496 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1497 sizeof(*na) + na_olen + dev->needed_tailroom;
1498 reply = alloc_skb(len, GFP_ATOMIC);
1502 reply->protocol = htons(ETH_P_IPV6);
1504 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1505 skb_push(reply, sizeof(struct ethhdr));
1506 skb_reset_mac_header(reply);
1508 ns = (struct nd_msg *)skb_transport_header(request);
1510 daddr = eth_hdr(request)->h_source;
1511 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1512 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1513 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1514 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1519 /* Ethernet header */
1520 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1521 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1522 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1523 reply->protocol = htons(ETH_P_IPV6);
1525 skb_pull(reply, sizeof(struct ethhdr));
1526 skb_reset_network_header(reply);
1527 skb_put(reply, sizeof(struct ipv6hdr));
1531 pip6 = ipv6_hdr(reply);
1532 memset(pip6, 0, sizeof(struct ipv6hdr));
1534 pip6->priority = ipv6_hdr(request)->priority;
1535 pip6->nexthdr = IPPROTO_ICMPV6;
1536 pip6->hop_limit = 255;
1537 pip6->daddr = ipv6_hdr(request)->saddr;
1538 pip6->saddr = *(struct in6_addr *)n->primary_key;
1540 skb_pull(reply, sizeof(struct ipv6hdr));
1541 skb_reset_transport_header(reply);
1543 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1545 /* Neighbor Advertisement */
1546 memset(na, 0, sizeof(*na)+na_olen);
1547 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1548 na->icmph.icmp6_router = isrouter;
1549 na->icmph.icmp6_override = 1;
1550 na->icmph.icmp6_solicited = 1;
1551 na->target = ns->target;
1552 ether_addr_copy(&na->opt[2], n->ha);
1553 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1554 na->opt[1] = na_olen >> 3;
1556 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1557 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1558 csum_partial(na, sizeof(*na)+na_olen, 0));
1560 pip6->payload_len = htons(sizeof(*na)+na_olen);
1562 skb_push(reply, sizeof(struct ipv6hdr));
1564 reply->ip_summed = CHECKSUM_UNNECESSARY;
1569 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1571 struct vxlan_dev *vxlan = netdev_priv(dev);
1573 const struct ipv6hdr *iphdr;
1574 const struct in6_addr *saddr, *daddr;
1575 struct neighbour *n;
1576 struct inet6_dev *in6_dev;
1578 in6_dev = __in6_dev_get(dev);
1582 iphdr = ipv6_hdr(skb);
1583 saddr = &iphdr->saddr;
1584 daddr = &iphdr->daddr;
1586 msg = (struct nd_msg *)skb_transport_header(skb);
1587 if (msg->icmph.icmp6_code != 0 ||
1588 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1591 if (ipv6_addr_loopback(daddr) ||
1592 ipv6_addr_is_multicast(&msg->target))
1595 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1598 struct vxlan_fdb *f;
1599 struct sk_buff *reply;
1601 if (!(n->nud_state & NUD_CONNECTED)) {
1606 f = vxlan_find_mac(vxlan, n->ha);
1607 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1608 /* bridge-local neighbor */
1613 reply = vxlan_na_create(skb, n,
1614 !!(f ? f->flags & NTF_ROUTER : 0));
1621 if (netif_rx_ni(reply) == NET_RX_DROP)
1622 dev->stats.rx_dropped++;
1624 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1625 union vxlan_addr ipa = {
1626 .sin6.sin6_addr = msg->target,
1627 .sin6.sin6_family = AF_INET6,
1630 vxlan_ip_miss(dev, &ipa);
1635 return NETDEV_TX_OK;
1639 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1641 struct vxlan_dev *vxlan = netdev_priv(dev);
1642 struct neighbour *n;
1644 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1648 switch (ntohs(eth_hdr(skb)->h_proto)) {
1653 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1656 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1657 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1658 union vxlan_addr ipa = {
1659 .sin.sin_addr.s_addr = pip->daddr,
1660 .sin.sin_family = AF_INET,
1663 vxlan_ip_miss(dev, &ipa);
1669 #if IS_ENABLED(CONFIG_IPV6)
1672 struct ipv6hdr *pip6;
1674 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1676 pip6 = ipv6_hdr(skb);
1677 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1678 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1679 union vxlan_addr ipa = {
1680 .sin6.sin6_addr = pip6->daddr,
1681 .sin6.sin6_family = AF_INET6,
1684 vxlan_ip_miss(dev, &ipa);
1698 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1700 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1702 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1711 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1712 struct vxlan_metadata *md)
1714 struct vxlanhdr_gbp *gbp;
1719 gbp = (struct vxlanhdr_gbp *)vxh;
1720 vxh->vx_flags |= VXLAN_HF_GBP;
1722 if (md->gbp & VXLAN_GBP_DONT_LEARN)
1723 gbp->dont_learn = 1;
1725 if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1726 gbp->policy_applied = 1;
1728 gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1731 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1734 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1736 gpe->np_applied = 1;
1739 case htons(ETH_P_IP):
1740 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1742 case htons(ETH_P_IPV6):
1743 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1745 case htons(ETH_P_TEB):
1746 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1749 return -EPFNOSUPPORT;
1752 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1753 int iphdr_len, __be32 vni,
1754 struct vxlan_metadata *md, u32 vxflags,
1757 struct vxlanhdr *vxh;
1760 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1761 __be16 inner_protocol = htons(ETH_P_TEB);
1763 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1764 skb->ip_summed == CHECKSUM_PARTIAL) {
1765 int csum_start = skb_checksum_start_offset(skb);
1767 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1768 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1769 (skb->csum_offset == offsetof(struct udphdr, check) ||
1770 skb->csum_offset == offsetof(struct tcphdr, check)))
1771 type |= SKB_GSO_TUNNEL_REMCSUM;
1774 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1775 + VXLAN_HLEN + iphdr_len
1776 + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1778 /* Need space for new headers (invalidates iph ptr) */
1779 err = skb_cow_head(skb, min_headroom);
1783 skb = vlan_hwaccel_push_inside(skb);
1787 skb = iptunnel_handle_offloads(skb, type);
1789 return PTR_ERR(skb);
1791 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1792 vxh->vx_flags = VXLAN_HF_VNI;
1793 vxh->vx_vni = vxlan_vni_field(vni);
1795 if (type & SKB_GSO_TUNNEL_REMCSUM) {
1798 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1799 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1800 vxh->vx_flags |= VXLAN_HF_RCO;
1802 if (!skb_is_gso(skb)) {
1803 skb->ip_summed = CHECKSUM_NONE;
1804 skb->encapsulation = 0;
1808 if (vxflags & VXLAN_F_GBP)
1809 vxlan_build_gbp_hdr(vxh, vxflags, md);
1810 if (vxflags & VXLAN_F_GPE) {
1811 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1814 inner_protocol = skb->protocol;
1817 skb_set_inner_protocol(skb, inner_protocol);
1825 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1826 struct sk_buff *skb, int oif, u8 tos,
1827 __be32 daddr, __be32 *saddr,
1828 struct dst_cache *dst_cache,
1829 const struct ip_tunnel_info *info)
1831 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1832 struct rtable *rt = NULL;
1838 rt = dst_cache_get_ip4(dst_cache, saddr);
1843 memset(&fl4, 0, sizeof(fl4));
1844 fl4.flowi4_oif = oif;
1845 fl4.flowi4_tos = RT_TOS(tos);
1846 fl4.flowi4_mark = skb->mark;
1847 fl4.flowi4_proto = IPPROTO_UDP;
1849 fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
1851 rt = ip_route_output_key(vxlan->net, &fl4);
1855 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1860 #if IS_ENABLED(CONFIG_IPV6)
1861 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1862 struct sk_buff *skb, int oif, u8 tos,
1864 const struct in6_addr *daddr,
1865 struct in6_addr *saddr,
1866 struct dst_cache *dst_cache,
1867 const struct ip_tunnel_info *info)
1869 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1870 struct dst_entry *ndst;
1877 ndst = dst_cache_get_ip6(dst_cache, saddr);
1882 memset(&fl6, 0, sizeof(fl6));
1883 fl6.flowi6_oif = oif;
1885 fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
1886 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1887 fl6.flowi6_mark = skb->mark;
1888 fl6.flowi6_proto = IPPROTO_UDP;
1890 err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1891 vxlan->vn6_sock->sock->sk,
1894 return ERR_PTR(err);
1898 dst_cache_set_ip6(dst_cache, ndst, saddr);
1903 /* Bypass encapsulation if the destination is local */
1904 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1905 struct vxlan_dev *dst_vxlan)
1907 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1908 union vxlan_addr loopback;
1909 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1910 struct net_device *dev = skb->dev;
1913 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1914 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1915 skb->pkt_type = PACKET_HOST;
1916 skb->encapsulation = 0;
1917 skb->dev = dst_vxlan->dev;
1918 __skb_pull(skb, skb_network_offset(skb));
1920 if (remote_ip->sa.sa_family == AF_INET) {
1921 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1922 loopback.sa.sa_family = AF_INET;
1923 #if IS_ENABLED(CONFIG_IPV6)
1925 loopback.sin6.sin6_addr = in6addr_loopback;
1926 loopback.sa.sa_family = AF_INET6;
1930 if (dst_vxlan->flags & VXLAN_F_LEARN)
1931 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1933 u64_stats_update_begin(&tx_stats->syncp);
1934 tx_stats->tx_packets++;
1935 tx_stats->tx_bytes += len;
1936 u64_stats_update_end(&tx_stats->syncp);
1938 if (netif_rx(skb) == NET_RX_SUCCESS) {
1939 u64_stats_update_begin(&rx_stats->syncp);
1940 rx_stats->rx_packets++;
1941 rx_stats->rx_bytes += len;
1942 u64_stats_update_end(&rx_stats->syncp);
1944 dev->stats.rx_dropped++;
1948 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1949 struct vxlan_rdst *rdst, bool did_rsc)
1951 struct dst_cache *dst_cache;
1952 struct ip_tunnel_info *info;
1953 struct vxlan_dev *vxlan = netdev_priv(dev);
1955 struct rtable *rt = NULL;
1956 const struct iphdr *old_iph;
1957 union vxlan_addr *dst;
1958 union vxlan_addr remote_ip;
1959 struct vxlan_metadata _md;
1960 struct vxlan_metadata *md = &_md;
1961 __be16 src_port = 0, dst_port;
1966 u32 flags = vxlan->flags;
1967 bool udp_sum = false;
1968 bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1970 info = skb_tunnel_info(skb);
1973 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1974 vni = rdst->remote_vni;
1975 dst = &rdst->remote_ip;
1976 dst_cache = &rdst->dst_cache;
1979 WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1983 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1984 vni = vxlan_tun_id_to_vni(info->key.tun_id);
1985 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1986 if (remote_ip.sa.sa_family == AF_INET)
1987 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1989 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1991 dst_cache = &info->dst_cache;
1994 if (vxlan_addr_any(dst)) {
1996 /* short-circuited back to local bridge */
1997 vxlan_encap_bypass(skb, vxlan, vxlan);
2003 old_iph = ip_hdr(skb);
2005 ttl = vxlan->cfg.ttl;
2006 if (!ttl && vxlan_addr_multicast(dst))
2009 tos = vxlan->cfg.tos;
2011 tos = ip_tunnel_get_dsfield(old_iph, skb);
2013 label = vxlan->cfg.label;
2014 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2015 vxlan->cfg.port_max, true);
2018 ttl = info->key.ttl;
2019 tos = info->key.tos;
2020 label = info->key.label;
2021 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2023 if (info->options_len)
2024 md = ip_tunnel_info_opts(info);
2026 md->gbp = skb->mark;
2029 if (dst->sa.sa_family == AF_INET) {
2032 if (!vxlan->vn4_sock)
2034 sk = vxlan->vn4_sock->sock->sk;
2036 rt = vxlan_get_route(vxlan, skb,
2037 rdst ? rdst->remote_ifindex : 0, tos,
2038 dst->sin.sin_addr.s_addr, &saddr,
2041 netdev_dbg(dev, "no route to %pI4\n",
2042 &dst->sin.sin_addr.s_addr);
2043 dev->stats.tx_carrier_errors++;
2047 if (rt->dst.dev == dev) {
2048 netdev_dbg(dev, "circular route to %pI4\n",
2049 &dst->sin.sin_addr.s_addr);
2050 dev->stats.collisions++;
2054 /* Bypass encapsulation if the destination is local */
2055 if (rt->rt_flags & RTCF_LOCAL &&
2056 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2057 struct vxlan_dev *dst_vxlan;
2060 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2061 dst->sa.sa_family, dst_port,
2065 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2070 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2071 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2074 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2075 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2076 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2077 vni, md, flags, udp_sum);
2081 udp_tunnel_xmit_skb(rt, sk, skb, saddr,
2082 dst->sin.sin_addr.s_addr, tos, ttl, df,
2083 src_port, dst_port, xnet, !udp_sum);
2084 #if IS_ENABLED(CONFIG_IPV6)
2086 struct dst_entry *ndst;
2087 struct in6_addr saddr;
2090 if (!vxlan->vn6_sock)
2092 sk = vxlan->vn6_sock->sock->sk;
2094 ndst = vxlan6_get_route(vxlan, skb,
2095 rdst ? rdst->remote_ifindex : 0, tos,
2096 label, &dst->sin6.sin6_addr, &saddr,
2099 netdev_dbg(dev, "no route to %pI6\n",
2100 &dst->sin6.sin6_addr);
2101 dev->stats.tx_carrier_errors++;
2105 if (ndst->dev == dev) {
2106 netdev_dbg(dev, "circular route to %pI6\n",
2107 &dst->sin6.sin6_addr);
2109 dev->stats.collisions++;
2113 /* Bypass encapsulation if the destination is local */
2114 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2115 if (rt6i_flags & RTF_LOCAL &&
2116 !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2117 struct vxlan_dev *dst_vxlan;
2120 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2121 dst->sa.sa_family, dst_port,
2125 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2130 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2132 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2133 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2134 skb_scrub_packet(skb, xnet);
2135 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2136 vni, md, flags, udp_sum);
2141 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2142 &saddr, &dst->sin6.sin6_addr, tos, ttl,
2143 label, src_port, dst_port, !udp_sum);
2150 dev->stats.tx_dropped++;
2154 /* skb is already freed. */
2159 dev->stats.tx_errors++;
2164 /* Transmit local packets over Vxlan
2166 * Outer IP header inherits ECN and DF from inner header.
2167 * Outer UDP destination is the VXLAN assigned port.
2168 * source port is based on hash of flow
2170 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2172 struct vxlan_dev *vxlan = netdev_priv(dev);
2173 const struct ip_tunnel_info *info;
2175 bool did_rsc = false;
2176 struct vxlan_rdst *rdst, *fdst = NULL;
2177 struct vxlan_fdb *f;
2179 info = skb_tunnel_info(skb);
2181 skb_reset_mac_header(skb);
2183 if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2184 if (info && info->mode & IP_TUNNEL_INFO_TX)
2185 vxlan_xmit_one(skb, dev, NULL, false);
2188 return NETDEV_TX_OK;
2191 if (vxlan->flags & VXLAN_F_PROXY) {
2193 if (ntohs(eth->h_proto) == ETH_P_ARP)
2194 return arp_reduce(dev, skb);
2195 #if IS_ENABLED(CONFIG_IPV6)
2196 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2197 pskb_may_pull(skb, sizeof(struct ipv6hdr)
2198 + sizeof(struct nd_msg)) &&
2199 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2202 msg = (struct nd_msg *)skb_transport_header(skb);
2203 if (msg->icmph.icmp6_code == 0 &&
2204 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2205 return neigh_reduce(dev, skb);
2211 f = vxlan_find_mac(vxlan, eth->h_dest);
2214 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2215 (ntohs(eth->h_proto) == ETH_P_IP ||
2216 ntohs(eth->h_proto) == ETH_P_IPV6)) {
2217 did_rsc = route_shortcircuit(dev, skb);
2219 f = vxlan_find_mac(vxlan, eth->h_dest);
2223 f = vxlan_find_mac(vxlan, all_zeros_mac);
2225 if ((vxlan->flags & VXLAN_F_L2MISS) &&
2226 !is_multicast_ether_addr(eth->h_dest))
2227 vxlan_fdb_miss(vxlan, eth->h_dest);
2229 dev->stats.tx_dropped++;
2231 return NETDEV_TX_OK;
2235 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2236 struct sk_buff *skb1;
2242 skb1 = skb_clone(skb, GFP_ATOMIC);
2244 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2248 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2251 return NETDEV_TX_OK;
2254 /* Walk the forwarding table and purge stale entries */
2255 static void vxlan_cleanup(unsigned long arg)
2257 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2258 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2261 if (!netif_running(vxlan->dev))
2264 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2265 struct hlist_node *p, *n;
2267 spin_lock_bh(&vxlan->hash_lock);
2268 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2270 = container_of(p, struct vxlan_fdb, hlist);
2271 unsigned long timeout;
2273 if (f->state & NUD_PERMANENT)
2276 timeout = f->used + vxlan->cfg.age_interval * HZ;
2277 if (time_before_eq(timeout, jiffies)) {
2278 netdev_dbg(vxlan->dev,
2279 "garbage collect %pM\n",
2281 f->state = NUD_STALE;
2282 vxlan_fdb_destroy(vxlan, f);
2283 } else if (time_before(timeout, next_timer))
2284 next_timer = timeout;
2286 spin_unlock_bh(&vxlan->hash_lock);
2289 mod_timer(&vxlan->age_timer, next_timer);
2292 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2294 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2295 __be32 vni = vxlan->default_dst.remote_vni;
2297 spin_lock(&vn->sock_lock);
2298 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2299 spin_unlock(&vn->sock_lock);
2302 /* Setup stats when device is created */
2303 static int vxlan_init(struct net_device *dev)
2305 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2312 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2314 struct vxlan_fdb *f;
2316 spin_lock_bh(&vxlan->hash_lock);
2317 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2319 vxlan_fdb_destroy(vxlan, f);
2320 spin_unlock_bh(&vxlan->hash_lock);
2323 static void vxlan_uninit(struct net_device *dev)
2325 struct vxlan_dev *vxlan = netdev_priv(dev);
2327 vxlan_fdb_delete_default(vxlan);
2329 free_percpu(dev->tstats);
2332 /* Start ageing timer and join group when device is brought up */
2333 static int vxlan_open(struct net_device *dev)
2335 struct vxlan_dev *vxlan = netdev_priv(dev);
2338 ret = vxlan_sock_add(vxlan);
2342 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2343 ret = vxlan_igmp_join(vxlan);
2344 if (ret == -EADDRINUSE)
2347 vxlan_sock_release(vxlan);
2352 if (vxlan->cfg.age_interval)
2353 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2358 /* Purge the forwarding table */
2359 static void vxlan_flush(struct vxlan_dev *vxlan)
2363 spin_lock_bh(&vxlan->hash_lock);
2364 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2365 struct hlist_node *p, *n;
2366 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2368 = container_of(p, struct vxlan_fdb, hlist);
2369 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2370 if (!is_zero_ether_addr(f->eth_addr))
2371 vxlan_fdb_destroy(vxlan, f);
2374 spin_unlock_bh(&vxlan->hash_lock);
2377 /* Cleanup timer and forwarding table on shutdown */
2378 static int vxlan_stop(struct net_device *dev)
2380 struct vxlan_dev *vxlan = netdev_priv(dev);
2381 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2384 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2385 !vxlan_group_used(vn, vxlan))
2386 ret = vxlan_igmp_leave(vxlan);
2388 del_timer_sync(&vxlan->age_timer);
2391 vxlan_sock_release(vxlan);
2396 /* Stub, nothing needs to be done. */
2397 static void vxlan_set_multicast_list(struct net_device *dev)
2401 static int __vxlan_change_mtu(struct net_device *dev,
2402 struct net_device *lowerdev,
2403 struct vxlan_rdst *dst, int new_mtu, bool strict)
2405 int max_mtu = IP_MAX_MTU;
2408 max_mtu = lowerdev->mtu;
2410 if (dst->remote_ip.sa.sa_family == AF_INET6)
2411 max_mtu -= VXLAN6_HEADROOM;
2413 max_mtu -= VXLAN_HEADROOM;
2418 if (new_mtu > max_mtu) {
2429 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2431 struct vxlan_dev *vxlan = netdev_priv(dev);
2432 struct vxlan_rdst *dst = &vxlan->default_dst;
2433 struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2434 dst->remote_ifindex);
2435 return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2438 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2440 struct vxlan_dev *vxlan = netdev_priv(dev);
2441 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2442 __be16 sport, dport;
2444 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2445 vxlan->cfg.port_max, true);
2446 dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2448 if (ip_tunnel_info_af(info) == AF_INET) {
2451 if (!vxlan->vn4_sock)
2453 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2454 info->key.u.ipv4.dst,
2455 &info->key.u.ipv4.src, NULL, info);
2460 #if IS_ENABLED(CONFIG_IPV6)
2461 struct dst_entry *ndst;
2463 if (!vxlan->vn6_sock)
2465 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2466 info->key.label, &info->key.u.ipv6.dst,
2467 &info->key.u.ipv6.src, NULL, info);
2469 return PTR_ERR(ndst);
2471 #else /* !CONFIG_IPV6 */
2472 return -EPFNOSUPPORT;
2475 info->key.tp_src = sport;
2476 info->key.tp_dst = dport;
2480 static const struct net_device_ops vxlan_netdev_ether_ops = {
2481 .ndo_init = vxlan_init,
2482 .ndo_uninit = vxlan_uninit,
2483 .ndo_open = vxlan_open,
2484 .ndo_stop = vxlan_stop,
2485 .ndo_start_xmit = vxlan_xmit,
2486 .ndo_get_stats64 = ip_tunnel_get_stats64,
2487 .ndo_set_rx_mode = vxlan_set_multicast_list,
2488 .ndo_change_mtu = vxlan_change_mtu,
2489 .ndo_validate_addr = eth_validate_addr,
2490 .ndo_set_mac_address = eth_mac_addr,
2491 .ndo_fdb_add = vxlan_fdb_add,
2492 .ndo_fdb_del = vxlan_fdb_delete,
2493 .ndo_fdb_dump = vxlan_fdb_dump,
2494 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2497 static const struct net_device_ops vxlan_netdev_raw_ops = {
2498 .ndo_init = vxlan_init,
2499 .ndo_uninit = vxlan_uninit,
2500 .ndo_open = vxlan_open,
2501 .ndo_stop = vxlan_stop,
2502 .ndo_start_xmit = vxlan_xmit,
2503 .ndo_get_stats64 = ip_tunnel_get_stats64,
2504 .ndo_change_mtu = vxlan_change_mtu,
2505 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2508 /* Info for udev, that this is a virtual tunnel endpoint */
2509 static struct device_type vxlan_type = {
2513 /* Calls the ndo_add_vxlan_port of the caller in order to
2514 * supply the listening VXLAN udp ports. Callers are expected
2515 * to implement the ndo_add_vxlan_port.
2517 void vxlan_get_rx_port(struct net_device *dev)
2519 struct vxlan_sock *vs;
2520 struct net *net = dev_net(dev);
2521 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2522 sa_family_t sa_family;
2526 spin_lock(&vn->sock_lock);
2527 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2528 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2529 port = inet_sk(vs->sock->sk)->inet_sport;
2530 sa_family = vxlan_get_sk_family(vs);
2531 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2535 spin_unlock(&vn->sock_lock);
2537 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2539 /* Initialize the device structure. */
2540 static void vxlan_setup(struct net_device *dev)
2542 struct vxlan_dev *vxlan = netdev_priv(dev);
2545 dev->destructor = free_netdev;
2546 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2548 dev->features |= NETIF_F_LLTX;
2549 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2550 dev->features |= NETIF_F_RXCSUM;
2551 dev->features |= NETIF_F_GSO_SOFTWARE;
2553 dev->vlan_features = dev->features;
2554 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2555 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2556 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2557 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2558 netif_keep_dst(dev);
2559 dev->priv_flags |= IFF_NO_QUEUE;
2561 INIT_LIST_HEAD(&vxlan->next);
2562 spin_lock_init(&vxlan->hash_lock);
2564 init_timer_deferrable(&vxlan->age_timer);
2565 vxlan->age_timer.function = vxlan_cleanup;
2566 vxlan->age_timer.data = (unsigned long) vxlan;
2568 vxlan->cfg.dst_port = htons(vxlan_port);
2572 gro_cells_init(&vxlan->gro_cells, dev);
2574 for (h = 0; h < FDB_HASH_SIZE; ++h)
2575 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2578 static void vxlan_ether_setup(struct net_device *dev)
2580 eth_hw_addr_random(dev);
2582 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2583 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2584 dev->netdev_ops = &vxlan_netdev_ether_ops;
2587 static void vxlan_raw_setup(struct net_device *dev)
2589 dev->type = ARPHRD_NONE;
2590 dev->hard_header_len = 0;
2592 dev->mtu = ETH_DATA_LEN;
2593 dev->tx_queue_len = 1000;
2594 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2595 dev->netdev_ops = &vxlan_netdev_raw_ops;
2598 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2599 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2600 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2601 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2602 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2603 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2604 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2605 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2606 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2607 [IFLA_VXLAN_LABEL] = { .type = NLA_U32 },
2608 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2609 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2610 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2611 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2612 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2613 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2614 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2615 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2616 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 },
2617 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2618 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 },
2619 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
2620 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
2621 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2622 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2623 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, },
2624 [IFLA_VXLAN_GPE] = { .type = NLA_FLAG, },
2625 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG },
2628 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2630 if (tb[IFLA_ADDRESS]) {
2631 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2632 pr_debug("invalid link address (not ethernet)\n");
2636 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2637 pr_debug("invalid all zero ethernet address\n");
2638 return -EADDRNOTAVAIL;
2645 if (data[IFLA_VXLAN_ID]) {
2646 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2647 if (id >= VXLAN_VID_MASK)
2651 if (data[IFLA_VXLAN_PORT_RANGE]) {
2652 const struct ifla_vxlan_port_range *p
2653 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2655 if (ntohs(p->high) < ntohs(p->low)) {
2656 pr_debug("port range %u .. %u not valid\n",
2657 ntohs(p->low), ntohs(p->high));
2665 static void vxlan_get_drvinfo(struct net_device *netdev,
2666 struct ethtool_drvinfo *drvinfo)
2668 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2669 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2672 static const struct ethtool_ops vxlan_ethtool_ops = {
2673 .get_drvinfo = vxlan_get_drvinfo,
2674 .get_link = ethtool_op_get_link,
2677 static void vxlan_del_work(struct work_struct *work)
2679 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2680 udp_tunnel_sock_release(vs->sock);
2684 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2685 __be16 port, u32 flags)
2687 struct socket *sock;
2688 struct udp_port_cfg udp_conf;
2691 memset(&udp_conf, 0, sizeof(udp_conf));
2694 udp_conf.family = AF_INET6;
2695 udp_conf.use_udp6_rx_checksums =
2696 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2697 udp_conf.ipv6_v6only = 1;
2699 udp_conf.family = AF_INET;
2702 udp_conf.local_udp_port = port;
2704 /* Open UDP socket */
2705 err = udp_sock_create(net, &udp_conf, &sock);
2707 return ERR_PTR(err);
2712 /* Create new listen socket if needed */
2713 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2714 __be16 port, u32 flags)
2716 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2717 struct vxlan_sock *vs;
2718 struct socket *sock;
2720 struct udp_tunnel_sock_cfg tunnel_cfg;
2722 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2724 return ERR_PTR(-ENOMEM);
2726 for (h = 0; h < VNI_HASH_SIZE; ++h)
2727 INIT_HLIST_HEAD(&vs->vni_list[h]);
2729 INIT_WORK(&vs->del_work, vxlan_del_work);
2731 sock = vxlan_create_sock(net, ipv6, port, flags);
2733 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2736 return ERR_CAST(sock);
2740 atomic_set(&vs->refcnt, 1);
2741 vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2743 spin_lock(&vn->sock_lock);
2744 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2745 vxlan_notify_add_rx_port(vs);
2746 spin_unlock(&vn->sock_lock);
2748 /* Mark socket as an encapsulation socket. */
2749 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2750 tunnel_cfg.sk_user_data = vs;
2751 tunnel_cfg.encap_type = 1;
2752 tunnel_cfg.encap_rcv = vxlan_rcv;
2753 tunnel_cfg.encap_destroy = NULL;
2754 tunnel_cfg.gro_receive = vxlan_gro_receive;
2755 tunnel_cfg.gro_complete = vxlan_gro_complete;
2757 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2762 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2764 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2765 struct vxlan_sock *vs = NULL;
2767 if (!vxlan->cfg.no_share) {
2768 spin_lock(&vn->sock_lock);
2769 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2770 vxlan->cfg.dst_port, vxlan->flags);
2771 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2772 spin_unlock(&vn->sock_lock);
2775 spin_unlock(&vn->sock_lock);
2778 vs = vxlan_socket_create(vxlan->net, ipv6,
2779 vxlan->cfg.dst_port, vxlan->flags);
2782 #if IS_ENABLED(CONFIG_IPV6)
2784 vxlan->vn6_sock = vs;
2787 vxlan->vn4_sock = vs;
2788 vxlan_vs_add_dev(vs, vxlan);
2792 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2794 bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2795 bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2798 vxlan->vn4_sock = NULL;
2799 #if IS_ENABLED(CONFIG_IPV6)
2800 vxlan->vn6_sock = NULL;
2801 if (ipv6 || metadata)
2802 ret = __vxlan_sock_add(vxlan, true);
2804 if (!ret && (!ipv6 || metadata))
2805 ret = __vxlan_sock_add(vxlan, false);
2807 vxlan_sock_release(vxlan);
2811 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2812 struct vxlan_config *conf)
2814 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2815 struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2816 struct vxlan_rdst *dst = &vxlan->default_dst;
2817 unsigned short needed_headroom = ETH_HLEN;
2819 bool use_ipv6 = false;
2820 __be16 default_port = vxlan->cfg.dst_port;
2821 struct net_device *lowerdev = NULL;
2823 if (conf->flags & VXLAN_F_GPE) {
2824 if (conf->flags & ~VXLAN_F_ALLOWED_GPE)
2826 /* For now, allow GPE only together with COLLECT_METADATA.
2827 * This can be relaxed later; in such case, the other side
2828 * of the PtP link will have to be provided.
2830 if (!(conf->flags & VXLAN_F_COLLECT_METADATA))
2833 vxlan_raw_setup(dev);
2835 vxlan_ether_setup(dev);
2838 vxlan->net = src_net;
2840 dst->remote_vni = conf->vni;
2842 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2844 /* Unless IPv6 is explicitly requested, assume IPv4 */
2845 if (!dst->remote_ip.sa.sa_family)
2846 dst->remote_ip.sa.sa_family = AF_INET;
2848 if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2849 vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2850 if (!IS_ENABLED(CONFIG_IPV6))
2851 return -EPFNOSUPPORT;
2853 vxlan->flags |= VXLAN_F_IPV6;
2856 if (conf->label && !use_ipv6) {
2857 pr_info("label only supported in use with IPv6\n");
2861 if (conf->remote_ifindex) {
2862 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2863 dst->remote_ifindex = conf->remote_ifindex;
2866 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2870 #if IS_ENABLED(CONFIG_IPV6)
2872 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2873 if (idev && idev->cnf.disable_ipv6) {
2874 pr_info("IPv6 is disabled via sysctl\n");
2881 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2883 needed_headroom = lowerdev->hard_header_len;
2887 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2892 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2893 needed_headroom += VXLAN6_HEADROOM;
2895 needed_headroom += VXLAN_HEADROOM;
2896 dev->needed_headroom = needed_headroom;
2898 memcpy(&vxlan->cfg, conf, sizeof(*conf));
2899 if (!vxlan->cfg.dst_port) {
2900 if (conf->flags & VXLAN_F_GPE)
2901 vxlan->cfg.dst_port = 4790; /* IANA assigned VXLAN-GPE port */
2903 vxlan->cfg.dst_port = default_port;
2905 vxlan->flags |= conf->flags;
2907 if (!vxlan->cfg.age_interval)
2908 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2910 list_for_each_entry(tmp, &vn->vxlan_list, next) {
2911 if (tmp->cfg.vni == conf->vni &&
2912 (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2913 tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2914 tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2915 (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2916 (vxlan->flags & VXLAN_F_RCV_FLAGS))
2920 dev->ethtool_ops = &vxlan_ethtool_ops;
2922 /* create an fdb entry for a valid default destination */
2923 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2924 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2925 &vxlan->default_dst.remote_ip,
2926 NUD_REACHABLE|NUD_PERMANENT,
2927 NLM_F_EXCL|NLM_F_CREATE,
2928 vxlan->cfg.dst_port,
2929 vxlan->default_dst.remote_vni,
2930 vxlan->default_dst.remote_ifindex,
2936 err = register_netdevice(dev);
2938 vxlan_fdb_delete_default(vxlan);
2942 list_add(&vxlan->next, &vn->vxlan_list);
2947 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2948 u8 name_assign_type, struct vxlan_config *conf)
2950 struct nlattr *tb[IFLA_MAX+1];
2951 struct net_device *dev;
2954 memset(&tb, 0, sizeof(tb));
2956 dev = rtnl_create_link(net, name, name_assign_type,
2957 &vxlan_link_ops, tb);
2961 err = vxlan_dev_configure(net, dev, conf);
2964 return ERR_PTR(err);
2969 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2971 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2972 struct nlattr *tb[], struct nlattr *data[])
2974 struct vxlan_config conf;
2977 memset(&conf, 0, sizeof(conf));
2979 if (data[IFLA_VXLAN_ID])
2980 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2982 if (data[IFLA_VXLAN_GROUP]) {
2983 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2984 } else if (data[IFLA_VXLAN_GROUP6]) {
2985 if (!IS_ENABLED(CONFIG_IPV6))
2986 return -EPFNOSUPPORT;
2988 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2989 conf.remote_ip.sa.sa_family = AF_INET6;
2992 if (data[IFLA_VXLAN_LOCAL]) {
2993 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2994 conf.saddr.sa.sa_family = AF_INET;
2995 } else if (data[IFLA_VXLAN_LOCAL6]) {
2996 if (!IS_ENABLED(CONFIG_IPV6))
2997 return -EPFNOSUPPORT;
2999 /* TODO: respect scope id */
3000 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3001 conf.saddr.sa.sa_family = AF_INET6;
3004 if (data[IFLA_VXLAN_LINK])
3005 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3007 if (data[IFLA_VXLAN_TOS])
3008 conf.tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
3010 if (data[IFLA_VXLAN_TTL])
3011 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3013 if (data[IFLA_VXLAN_LABEL])
3014 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3015 IPV6_FLOWLABEL_MASK;
3017 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3018 conf.flags |= VXLAN_F_LEARN;
3020 if (data[IFLA_VXLAN_AGEING])
3021 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3023 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
3024 conf.flags |= VXLAN_F_PROXY;
3026 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
3027 conf.flags |= VXLAN_F_RSC;
3029 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3030 conf.flags |= VXLAN_F_L2MISS;
3032 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3033 conf.flags |= VXLAN_F_L3MISS;
3035 if (data[IFLA_VXLAN_LIMIT])
3036 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3038 if (data[IFLA_VXLAN_COLLECT_METADATA] &&
3039 nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3040 conf.flags |= VXLAN_F_COLLECT_METADATA;
3042 if (data[IFLA_VXLAN_PORT_RANGE]) {
3043 const struct ifla_vxlan_port_range *p
3044 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3045 conf.port_min = ntohs(p->low);
3046 conf.port_max = ntohs(p->high);
3049 if (data[IFLA_VXLAN_PORT])
3050 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3052 if (data[IFLA_VXLAN_UDP_CSUM] &&
3053 !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3054 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3056 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3057 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3058 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3060 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3061 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3062 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3064 if (data[IFLA_VXLAN_REMCSUM_TX] &&
3065 nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3066 conf.flags |= VXLAN_F_REMCSUM_TX;
3068 if (data[IFLA_VXLAN_REMCSUM_RX] &&
3069 nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3070 conf.flags |= VXLAN_F_REMCSUM_RX;
3072 if (data[IFLA_VXLAN_GBP])
3073 conf.flags |= VXLAN_F_GBP;
3075 if (data[IFLA_VXLAN_GPE])
3076 conf.flags |= VXLAN_F_GPE;
3078 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3079 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3081 err = vxlan_dev_configure(src_net, dev, &conf);
3084 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
3088 pr_info("IPv6 is disabled via sysctl\n");
3092 pr_info("duplicate VNI %u\n", be32_to_cpu(conf.vni));
3096 pr_info("unsupported combination of extensions\n");
3103 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3105 struct vxlan_dev *vxlan = netdev_priv(dev);
3106 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3108 spin_lock(&vn->sock_lock);
3109 if (!hlist_unhashed(&vxlan->hlist))
3110 hlist_del_rcu(&vxlan->hlist);
3111 spin_unlock(&vn->sock_lock);
3113 gro_cells_destroy(&vxlan->gro_cells);
3114 list_del(&vxlan->next);
3115 unregister_netdevice_queue(dev, head);
3118 static size_t vxlan_get_size(const struct net_device *dev)
3121 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
3122 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3123 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3124 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3125 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
3126 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
3127 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3128 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
3129 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
3130 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
3131 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
3132 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
3133 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */
3134 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3135 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3136 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3137 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3138 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3139 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3140 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3141 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3142 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3146 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3148 const struct vxlan_dev *vxlan = netdev_priv(dev);
3149 const struct vxlan_rdst *dst = &vxlan->default_dst;
3150 struct ifla_vxlan_port_range ports = {
3151 .low = htons(vxlan->cfg.port_min),
3152 .high = htons(vxlan->cfg.port_max),
3155 if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3156 goto nla_put_failure;
3158 if (!vxlan_addr_any(&dst->remote_ip)) {
3159 if (dst->remote_ip.sa.sa_family == AF_INET) {
3160 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3161 dst->remote_ip.sin.sin_addr.s_addr))
3162 goto nla_put_failure;
3163 #if IS_ENABLED(CONFIG_IPV6)
3165 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3166 &dst->remote_ip.sin6.sin6_addr))
3167 goto nla_put_failure;
3172 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3173 goto nla_put_failure;
3175 if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3176 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3177 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3178 vxlan->cfg.saddr.sin.sin_addr.s_addr))
3179 goto nla_put_failure;
3180 #if IS_ENABLED(CONFIG_IPV6)
3182 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3183 &vxlan->cfg.saddr.sin6.sin6_addr))
3184 goto nla_put_failure;
3189 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3190 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3191 nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3192 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3193 !!(vxlan->flags & VXLAN_F_LEARN)) ||
3194 nla_put_u8(skb, IFLA_VXLAN_PROXY,
3195 !!(vxlan->flags & VXLAN_F_PROXY)) ||
3196 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3197 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3198 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3199 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3200 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3201 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3202 !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3203 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3204 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3205 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3206 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3207 !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3208 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3209 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3210 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3211 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3212 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3213 !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3214 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3215 !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3216 goto nla_put_failure;
3218 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3219 goto nla_put_failure;
3221 if (vxlan->flags & VXLAN_F_GBP &&
3222 nla_put_flag(skb, IFLA_VXLAN_GBP))
3223 goto nla_put_failure;
3225 if (vxlan->flags & VXLAN_F_GPE &&
3226 nla_put_flag(skb, IFLA_VXLAN_GPE))
3227 goto nla_put_failure;
3229 if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3230 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3231 goto nla_put_failure;
3239 static struct net *vxlan_get_link_net(const struct net_device *dev)
3241 struct vxlan_dev *vxlan = netdev_priv(dev);
3246 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3248 .maxtype = IFLA_VXLAN_MAX,
3249 .policy = vxlan_policy,
3250 .priv_size = sizeof(struct vxlan_dev),
3251 .setup = vxlan_setup,
3252 .validate = vxlan_validate,
3253 .newlink = vxlan_newlink,
3254 .dellink = vxlan_dellink,
3255 .get_size = vxlan_get_size,
3256 .fill_info = vxlan_fill_info,
3257 .get_link_net = vxlan_get_link_net,
3260 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3261 struct net_device *dev)
3263 struct vxlan_dev *vxlan, *next;
3264 LIST_HEAD(list_kill);
3266 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3267 struct vxlan_rdst *dst = &vxlan->default_dst;
3269 /* In case we created vxlan device with carrier
3270 * and we loose the carrier due to module unload
3271 * we also need to remove vxlan device. In other
3272 * cases, it's not necessary and remote_ifindex
3273 * is 0 here, so no matches.
3275 if (dst->remote_ifindex == dev->ifindex)
3276 vxlan_dellink(vxlan->dev, &list_kill);
3279 unregister_netdevice_many(&list_kill);
3282 static int vxlan_lowerdev_event(struct notifier_block *unused,
3283 unsigned long event, void *ptr)
3285 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3286 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3288 if (event == NETDEV_UNREGISTER)
3289 vxlan_handle_lowerdev_unregister(vn, dev);
3294 static struct notifier_block vxlan_notifier_block __read_mostly = {
3295 .notifier_call = vxlan_lowerdev_event,
3298 static __net_init int vxlan_init_net(struct net *net)
3300 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3303 INIT_LIST_HEAD(&vn->vxlan_list);
3304 spin_lock_init(&vn->sock_lock);
3306 for (h = 0; h < PORT_HASH_SIZE; ++h)
3307 INIT_HLIST_HEAD(&vn->sock_list[h]);
3312 static void __net_exit vxlan_exit_net(struct net *net)
3314 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3315 struct vxlan_dev *vxlan, *next;
3316 struct net_device *dev, *aux;
3320 for_each_netdev_safe(net, dev, aux)
3321 if (dev->rtnl_link_ops == &vxlan_link_ops)
3322 unregister_netdevice_queue(dev, &list);
3324 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3325 /* If vxlan->dev is in the same netns, it has already been added
3326 * to the list by the previous loop.
3328 if (!net_eq(dev_net(vxlan->dev), net)) {
3329 gro_cells_destroy(&vxlan->gro_cells);
3330 unregister_netdevice_queue(vxlan->dev, &list);
3334 unregister_netdevice_many(&list);
3338 static struct pernet_operations vxlan_net_ops = {
3339 .init = vxlan_init_net,
3340 .exit = vxlan_exit_net,
3341 .id = &vxlan_net_id,
3342 .size = sizeof(struct vxlan_net),
3345 static int __init vxlan_init_module(void)
3349 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3353 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3355 rc = register_pernet_subsys(&vxlan_net_ops);
3359 rc = register_netdevice_notifier(&vxlan_notifier_block);
3363 rc = rtnl_link_register(&vxlan_link_ops);
3369 unregister_netdevice_notifier(&vxlan_notifier_block);
3371 unregister_pernet_subsys(&vxlan_net_ops);
3373 destroy_workqueue(vxlan_wq);
3376 late_initcall(vxlan_init_module);
3378 static void __exit vxlan_cleanup_module(void)
3380 rtnl_link_unregister(&vxlan_link_ops);
3381 unregister_netdevice_notifier(&vxlan_notifier_block);
3382 destroy_workqueue(vxlan_wq);
3383 unregister_pernet_subsys(&vxlan_net_ops);
3384 /* rcu_barrier() is called by netns */
3386 module_exit(vxlan_cleanup_module);
3388 MODULE_LICENSE("GPL");
3389 MODULE_VERSION(VXLAN_VERSION);
3390 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3391 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3392 MODULE_ALIAS_RTNL_LINK("vxlan");