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, u32 id)
202 return &vs->vni_list[hash_32(id, 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, u32 id)
247 struct vxlan_dev *vxlan;
249 hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
250 if (vxlan->default_dst.remote_vni == id)
257 /* Look up VNI in a per net namespace table */
258 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
259 sa_family_t family, __be16 port,
262 struct vxlan_sock *vs;
264 vs = vxlan_find_sock(net, family, port, flags);
268 return vxlan_vs_find_vni(vs, id);
271 /* Fill in neighbour message in skbuff. */
272 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
273 const struct vxlan_fdb *fdb,
274 u32 portid, u32 seq, int type, unsigned int flags,
275 const struct vxlan_rdst *rdst)
277 unsigned long now = jiffies;
278 struct nda_cacheinfo ci;
279 struct nlmsghdr *nlh;
281 bool send_ip, send_eth;
283 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
287 ndm = nlmsg_data(nlh);
288 memset(ndm, 0, sizeof(*ndm));
290 send_eth = send_ip = true;
292 if (type == RTM_GETNEIGH) {
293 ndm->ndm_family = AF_INET;
294 send_ip = !vxlan_addr_any(&rdst->remote_ip);
295 send_eth = !is_zero_ether_addr(fdb->eth_addr);
297 ndm->ndm_family = AF_BRIDGE;
298 ndm->ndm_state = fdb->state;
299 ndm->ndm_ifindex = vxlan->dev->ifindex;
300 ndm->ndm_flags = fdb->flags;
301 ndm->ndm_type = RTN_UNICAST;
303 if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
304 nla_put_s32(skb, NDA_LINK_NETNSID,
305 peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
306 goto nla_put_failure;
308 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
309 goto nla_put_failure;
311 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
312 goto nla_put_failure;
314 if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
315 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
316 goto nla_put_failure;
317 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
318 nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
319 goto nla_put_failure;
320 if (rdst->remote_ifindex &&
321 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
322 goto nla_put_failure;
324 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
325 ci.ndm_confirmed = 0;
326 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
329 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
330 goto nla_put_failure;
336 nlmsg_cancel(skb, nlh);
340 static inline size_t vxlan_nlmsg_size(void)
342 return NLMSG_ALIGN(sizeof(struct ndmsg))
343 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
344 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
345 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
346 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
347 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
348 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
349 + nla_total_size(sizeof(struct nda_cacheinfo));
352 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
353 struct vxlan_rdst *rd, int type)
355 struct net *net = dev_net(vxlan->dev);
359 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
363 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
365 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
366 WARN_ON(err == -EMSGSIZE);
371 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
375 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
378 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
380 struct vxlan_dev *vxlan = netdev_priv(dev);
381 struct vxlan_fdb f = {
384 struct vxlan_rdst remote = {
385 .remote_ip = *ipa, /* goes to NDA_DST */
386 .remote_vni = VXLAN_N_VID,
389 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
392 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
394 struct vxlan_fdb f = {
397 struct vxlan_rdst remote = { };
399 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
401 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
404 /* Hash Ethernet address */
405 static u32 eth_hash(const unsigned char *addr)
407 u64 value = get_unaligned((u64 *)addr);
409 /* only want 6 bytes */
415 return hash_64(value, FDB_HASH_BITS);
418 /* Hash chain to use given mac address */
419 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
422 return &vxlan->fdb_head[eth_hash(mac)];
425 /* Look up Ethernet address in forwarding table */
426 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
429 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
432 hlist_for_each_entry_rcu(f, head, hlist) {
433 if (ether_addr_equal(mac, f->eth_addr))
440 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
445 f = __vxlan_find_mac(vxlan, mac);
452 /* caller should hold vxlan->hash_lock */
453 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
454 union vxlan_addr *ip, __be16 port,
455 __u32 vni, __u32 ifindex)
457 struct vxlan_rdst *rd;
459 list_for_each_entry(rd, &f->remotes, list) {
460 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
461 rd->remote_port == port &&
462 rd->remote_vni == vni &&
463 rd->remote_ifindex == ifindex)
470 /* Replace destination of unicast mac */
471 static int vxlan_fdb_replace(struct vxlan_fdb *f,
472 union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
474 struct vxlan_rdst *rd;
476 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
480 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
484 dst_cache_reset(&rd->dst_cache);
486 rd->remote_port = port;
487 rd->remote_vni = vni;
488 rd->remote_ifindex = ifindex;
492 /* Add/update destinations for multicast */
493 static int vxlan_fdb_append(struct vxlan_fdb *f,
494 union vxlan_addr *ip, __be16 port, __u32 vni,
495 __u32 ifindex, struct vxlan_rdst **rdp)
497 struct vxlan_rdst *rd;
499 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
503 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
507 if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
513 rd->remote_port = port;
514 rd->remote_vni = vni;
515 rd->remote_ifindex = ifindex;
517 list_add_tail_rcu(&rd->list, &f->remotes);
523 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
525 struct vxlanhdr *vh, size_t hdrlen,
526 u32 data, struct gro_remcsum *grc,
529 size_t start, offset;
531 if (skb->remcsum_offload)
534 if (!NAPI_GRO_CB(skb)->csum_valid)
537 start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
538 offset = start + ((data & VXLAN_RCO_UDP) ?
539 offsetof(struct udphdr, check) :
540 offsetof(struct tcphdr, check));
542 vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
543 start, offset, grc, nopartial);
545 skb->remcsum_offload = 1;
550 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
552 struct udp_offload *uoff)
554 struct sk_buff *p, **pp = NULL;
555 struct vxlanhdr *vh, *vh2;
556 unsigned int hlen, off_vx;
558 struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
561 struct gro_remcsum grc;
563 skb_gro_remcsum_init(&grc);
565 off_vx = skb_gro_offset(skb);
566 hlen = off_vx + sizeof(*vh);
567 vh = skb_gro_header_fast(skb, off_vx);
568 if (skb_gro_header_hard(skb, hlen)) {
569 vh = skb_gro_header_slow(skb, hlen, off_vx);
574 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
576 flags = ntohl(vh->vx_flags);
578 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
579 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
580 ntohl(vh->vx_vni), &grc,
582 VXLAN_F_REMCSUM_NOPARTIAL));
588 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
592 for (p = *head; p; p = p->next) {
593 if (!NAPI_GRO_CB(p)->same_flow)
596 vh2 = (struct vxlanhdr *)(p->data + off_vx);
597 if (vh->vx_flags != vh2->vx_flags ||
598 vh->vx_vni != vh2->vx_vni) {
599 NAPI_GRO_CB(p)->same_flow = 0;
604 pp = eth_gro_receive(head, skb);
607 skb_gro_remcsum_cleanup(skb, &grc);
608 NAPI_GRO_CB(skb)->flush |= flush;
613 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
614 struct udp_offload *uoff)
616 udp_tunnel_gro_complete(skb, nhoff);
618 return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
621 /* Notify netdevs that UDP port started listening */
622 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
624 struct net_device *dev;
625 struct sock *sk = vs->sock->sk;
626 struct net *net = sock_net(sk);
627 sa_family_t sa_family = vxlan_get_sk_family(vs);
628 __be16 port = inet_sk(sk)->inet_sport;
631 if (sa_family == AF_INET) {
632 err = udp_add_offload(net, &vs->udp_offloads);
634 pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
638 for_each_netdev_rcu(net, dev) {
639 if (dev->netdev_ops->ndo_add_vxlan_port)
640 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
646 /* Notify netdevs that UDP port is no more listening */
647 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
649 struct net_device *dev;
650 struct sock *sk = vs->sock->sk;
651 struct net *net = sock_net(sk);
652 sa_family_t sa_family = vxlan_get_sk_family(vs);
653 __be16 port = inet_sk(sk)->inet_sport;
656 for_each_netdev_rcu(net, dev) {
657 if (dev->netdev_ops->ndo_del_vxlan_port)
658 dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
663 if (sa_family == AF_INET)
664 udp_del_offload(&vs->udp_offloads);
667 /* Add new entry to forwarding table -- assumes lock held */
668 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
669 const u8 *mac, union vxlan_addr *ip,
670 __u16 state, __u16 flags,
671 __be16 port, __u32 vni, __u32 ifindex,
674 struct vxlan_rdst *rd = NULL;
678 f = __vxlan_find_mac(vxlan, mac);
680 if (flags & NLM_F_EXCL) {
681 netdev_dbg(vxlan->dev,
682 "lost race to create %pM\n", mac);
685 if (f->state != state) {
687 f->updated = jiffies;
690 if (f->flags != ndm_flags) {
691 f->flags = ndm_flags;
692 f->updated = jiffies;
695 if ((flags & NLM_F_REPLACE)) {
696 /* Only change unicasts */
697 if (!(is_multicast_ether_addr(f->eth_addr) ||
698 is_zero_ether_addr(f->eth_addr))) {
699 notify |= vxlan_fdb_replace(f, ip, port, vni,
704 if ((flags & NLM_F_APPEND) &&
705 (is_multicast_ether_addr(f->eth_addr) ||
706 is_zero_ether_addr(f->eth_addr))) {
707 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
715 if (!(flags & NLM_F_CREATE))
718 if (vxlan->cfg.addrmax &&
719 vxlan->addrcnt >= vxlan->cfg.addrmax)
722 /* Disallow replace to add a multicast entry */
723 if ((flags & NLM_F_REPLACE) &&
724 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
727 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
728 f = kmalloc(sizeof(*f), GFP_ATOMIC);
734 f->flags = ndm_flags;
735 f->updated = f->used = jiffies;
736 INIT_LIST_HEAD(&f->remotes);
737 memcpy(f->eth_addr, mac, ETH_ALEN);
739 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
742 hlist_add_head_rcu(&f->hlist,
743 vxlan_fdb_head(vxlan, mac));
748 rd = first_remote_rtnl(f);
749 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
755 static void vxlan_fdb_free(struct rcu_head *head)
757 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
758 struct vxlan_rdst *rd, *nd;
760 list_for_each_entry_safe(rd, nd, &f->remotes, list) {
761 dst_cache_destroy(&rd->dst_cache);
767 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
769 netdev_dbg(vxlan->dev,
770 "delete %pM\n", f->eth_addr);
773 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
775 hlist_del_rcu(&f->hlist);
776 call_rcu(&f->rcu, vxlan_fdb_free);
779 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
780 union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
782 struct net *net = dev_net(vxlan->dev);
786 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
790 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
791 if (remote->sa.sa_family == AF_INET) {
792 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
793 ip->sa.sa_family = AF_INET;
794 #if IS_ENABLED(CONFIG_IPV6)
796 ip->sin6.sin6_addr = in6addr_any;
797 ip->sa.sa_family = AF_INET6;
803 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
805 *port = nla_get_be16(tb[NDA_PORT]);
807 *port = vxlan->cfg.dst_port;
811 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
813 *vni = nla_get_u32(tb[NDA_VNI]);
815 *vni = vxlan->default_dst.remote_vni;
818 if (tb[NDA_IFINDEX]) {
819 struct net_device *tdev;
821 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
823 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
824 tdev = __dev_get_by_index(net, *ifindex);
826 return -EADDRNOTAVAIL;
834 /* Add static entry (via netlink) */
835 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
836 struct net_device *dev,
837 const unsigned char *addr, u16 vid, u16 flags)
839 struct vxlan_dev *vxlan = netdev_priv(dev);
840 /* struct net *net = dev_net(vxlan->dev); */
846 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
847 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
852 if (tb[NDA_DST] == NULL)
855 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
859 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
860 return -EAFNOSUPPORT;
862 spin_lock_bh(&vxlan->hash_lock);
863 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
864 port, vni, ifindex, ndm->ndm_flags);
865 spin_unlock_bh(&vxlan->hash_lock);
870 /* Delete entry (via netlink) */
871 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
872 struct net_device *dev,
873 const unsigned char *addr, u16 vid)
875 struct vxlan_dev *vxlan = netdev_priv(dev);
877 struct vxlan_rdst *rd = NULL;
883 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
889 spin_lock_bh(&vxlan->hash_lock);
890 f = vxlan_find_mac(vxlan, addr);
894 if (!vxlan_addr_any(&ip)) {
895 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
902 /* remove a destination if it's not the only one on the list,
903 * otherwise destroy the fdb entry
905 if (rd && !list_is_singular(&f->remotes)) {
906 list_del_rcu(&rd->list);
907 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
912 vxlan_fdb_destroy(vxlan, f);
915 spin_unlock_bh(&vxlan->hash_lock);
920 /* Dump forwarding table */
921 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
922 struct net_device *dev,
923 struct net_device *filter_dev, int idx)
925 struct vxlan_dev *vxlan = netdev_priv(dev);
928 for (h = 0; h < FDB_HASH_SIZE; ++h) {
932 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
933 struct vxlan_rdst *rd;
935 list_for_each_entry_rcu(rd, &f->remotes, list) {
936 if (idx < cb->args[0])
939 err = vxlan_fdb_info(skb, vxlan, f,
940 NETLINK_CB(cb->skb).portid,
955 /* Watch incoming packets to learn mapping between Ethernet address
956 * and Tunnel endpoint.
957 * Return true if packet is bogus and should be dropped.
959 static bool vxlan_snoop(struct net_device *dev,
960 union vxlan_addr *src_ip, const u8 *src_mac)
962 struct vxlan_dev *vxlan = netdev_priv(dev);
965 f = vxlan_find_mac(vxlan, src_mac);
967 struct vxlan_rdst *rdst = first_remote_rcu(f);
969 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
972 /* Don't migrate static entries, drop packets */
973 if (f->state & NUD_NOARP)
978 "%pM migrated from %pIS to %pIS\n",
979 src_mac, &rdst->remote_ip.sa, &src_ip->sa);
981 rdst->remote_ip = *src_ip;
982 f->updated = jiffies;
983 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
985 /* learned new entry */
986 spin_lock(&vxlan->hash_lock);
988 /* close off race between vxlan_flush and incoming packets */
989 if (netif_running(dev))
990 vxlan_fdb_create(vxlan, src_mac, src_ip,
992 NLM_F_EXCL|NLM_F_CREATE,
994 vxlan->default_dst.remote_vni,
996 spin_unlock(&vxlan->hash_lock);
1002 /* See if multicast group is already in use by other ID */
1003 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1005 struct vxlan_dev *vxlan;
1006 unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1008 /* The vxlan_sock is only used by dev, leaving group has
1009 * no effect on other vxlan devices.
1011 if (family == AF_INET && dev->vn4_sock &&
1012 atomic_read(&dev->vn4_sock->refcnt) == 1)
1014 #if IS_ENABLED(CONFIG_IPV6)
1015 if (family == AF_INET6 && dev->vn6_sock &&
1016 atomic_read(&dev->vn6_sock->refcnt) == 1)
1020 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1021 if (!netif_running(vxlan->dev) || vxlan == dev)
1024 if (family == AF_INET && vxlan->vn4_sock != dev->vn4_sock)
1026 #if IS_ENABLED(CONFIG_IPV6)
1027 if (family == AF_INET6 && vxlan->vn6_sock != dev->vn6_sock)
1031 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1032 &dev->default_dst.remote_ip))
1035 if (vxlan->default_dst.remote_ifindex !=
1036 dev->default_dst.remote_ifindex)
1045 static void __vxlan_sock_release(struct vxlan_sock *vs)
1047 struct vxlan_net *vn;
1051 if (!atomic_dec_and_test(&vs->refcnt))
1054 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1055 spin_lock(&vn->sock_lock);
1056 hlist_del_rcu(&vs->hlist);
1057 vxlan_notify_del_rx_port(vs);
1058 spin_unlock(&vn->sock_lock);
1060 queue_work(vxlan_wq, &vs->del_work);
1063 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1065 __vxlan_sock_release(vxlan->vn4_sock);
1066 #if IS_ENABLED(CONFIG_IPV6)
1067 __vxlan_sock_release(vxlan->vn6_sock);
1071 /* Update multicast group membership when first VNI on
1072 * multicast address is brought up
1074 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1077 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1078 int ifindex = vxlan->default_dst.remote_ifindex;
1081 if (ip->sa.sa_family == AF_INET) {
1082 struct ip_mreqn mreq = {
1083 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1084 .imr_ifindex = ifindex,
1087 sk = vxlan->vn4_sock->sock->sk;
1089 ret = ip_mc_join_group(sk, &mreq);
1091 #if IS_ENABLED(CONFIG_IPV6)
1093 sk = vxlan->vn6_sock->sock->sk;
1095 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1096 &ip->sin6.sin6_addr);
1104 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1105 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1108 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1109 int ifindex = vxlan->default_dst.remote_ifindex;
1112 if (ip->sa.sa_family == AF_INET) {
1113 struct ip_mreqn mreq = {
1114 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1115 .imr_ifindex = ifindex,
1118 sk = vxlan->vn4_sock->sock->sk;
1120 ret = ip_mc_leave_group(sk, &mreq);
1122 #if IS_ENABLED(CONFIG_IPV6)
1124 sk = vxlan->vn6_sock->sock->sk;
1126 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1127 &ip->sin6.sin6_addr);
1135 static struct vxlanhdr *vxlan_remcsum(struct sk_buff *skb, struct vxlanhdr *vh,
1136 size_t hdrlen, u32 data, bool nopartial)
1138 size_t start, offset, plen;
1140 if (skb->remcsum_offload)
1143 start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
1144 offset = start + ((data & VXLAN_RCO_UDP) ?
1145 offsetof(struct udphdr, check) :
1146 offsetof(struct tcphdr, check));
1148 plen = hdrlen + offset + sizeof(u16);
1150 if (!pskb_may_pull(skb, plen))
1153 vh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1155 skb_remcsum_process(skb, (void *)vh + hdrlen, start, offset,
1161 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1162 struct vxlan_metadata *md, u32 vni,
1163 struct metadata_dst *tun_dst)
1165 struct iphdr *oip = NULL;
1166 struct ipv6hdr *oip6 = NULL;
1167 struct vxlan_dev *vxlan;
1168 struct pcpu_sw_netstats *stats;
1169 union vxlan_addr saddr;
1172 /* For flow based devices, map all packets to VNI 0 */
1173 if (vs->flags & VXLAN_F_COLLECT_METADATA)
1176 /* Is this VNI defined? */
1177 vxlan = vxlan_vs_find_vni(vs, vni);
1181 skb_reset_mac_header(skb);
1182 skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1183 skb->protocol = eth_type_trans(skb, vxlan->dev);
1184 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1186 /* Ignore packet loops (and multicast echo) */
1187 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1190 /* Get data from the outer IP header */
1191 if (vxlan_get_sk_family(vs) == AF_INET) {
1193 saddr.sin.sin_addr.s_addr = oip->saddr;
1194 saddr.sa.sa_family = AF_INET;
1195 #if IS_ENABLED(CONFIG_IPV6)
1197 oip6 = ipv6_hdr(skb);
1198 saddr.sin6.sin6_addr = oip6->saddr;
1199 saddr.sa.sa_family = AF_INET6;
1204 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1208 if ((vxlan->flags & VXLAN_F_LEARN) &&
1209 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1212 skb_reset_network_header(skb);
1213 /* In flow-based mode, GBP is carried in dst_metadata */
1214 if (!(vs->flags & VXLAN_F_COLLECT_METADATA))
1215 skb->mark = md->gbp;
1218 err = IP6_ECN_decapsulate(oip6, skb);
1220 err = IP_ECN_decapsulate(oip, skb);
1222 if (unlikely(err)) {
1223 if (log_ecn_error) {
1225 net_info_ratelimited("non-ECT from %pI6\n",
1228 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1229 &oip->saddr, oip->tos);
1232 ++vxlan->dev->stats.rx_frame_errors;
1233 ++vxlan->dev->stats.rx_errors;
1238 stats = this_cpu_ptr(vxlan->dev->tstats);
1239 u64_stats_update_begin(&stats->syncp);
1240 stats->rx_packets++;
1241 stats->rx_bytes += skb->len;
1242 u64_stats_update_end(&stats->syncp);
1244 gro_cells_receive(&vxlan->gro_cells, skb);
1249 dst_release((struct dst_entry *)tun_dst);
1251 /* Consume bad packet */
1255 /* Callback from net/ipv4/udp.c to receive packets */
1256 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1258 struct metadata_dst *tun_dst = NULL;
1259 struct vxlan_sock *vs;
1260 struct vxlanhdr *vxh;
1262 struct vxlan_metadata _md;
1263 struct vxlan_metadata *md = &_md;
1265 /* Need Vxlan and inner Ethernet header to be present */
1266 if (!pskb_may_pull(skb, VXLAN_HLEN))
1269 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1270 flags = ntohl(vxh->vx_flags);
1271 vni = ntohl(vxh->vx_vni);
1273 if (flags & VXLAN_HF_VNI) {
1274 flags &= ~VXLAN_HF_VNI;
1276 /* VNI flag always required to be set */
1280 if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1282 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1284 vs = rcu_dereference_sk_user_data(sk);
1288 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1289 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni,
1290 !!(vs->flags & VXLAN_F_REMCSUM_NOPARTIAL));
1294 flags &= ~VXLAN_HF_RCO;
1295 vni &= VXLAN_VNI_MASK;
1298 if (vxlan_collect_metadata(vs)) {
1299 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1300 cpu_to_be64(vni >> 8), sizeof(*md));
1305 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1307 memset(md, 0, sizeof(*md));
1310 /* For backwards compatibility, only allow reserved fields to be
1311 * used by VXLAN extensions if explicitly requested.
1313 if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1314 struct vxlanhdr_gbp *gbp;
1316 gbp = (struct vxlanhdr_gbp *)vxh;
1317 md->gbp = ntohs(gbp->policy_id);
1320 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1322 if (gbp->dont_learn)
1323 md->gbp |= VXLAN_GBP_DONT_LEARN;
1325 if (gbp->policy_applied)
1326 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1328 flags &= ~VXLAN_GBP_USED_BITS;
1331 if (flags || vni & ~VXLAN_VNI_MASK) {
1332 /* If there are any unprocessed flags remaining treat
1333 * this as a malformed packet. This behavior diverges from
1334 * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1335 * in reserved fields are to be ignored. The approach here
1336 * maintains compatibility with previous stack code, and also
1337 * is more robust and provides a little more security in
1338 * adding extensions to VXLAN.
1344 vxlan_rcv(vs, skb, md, vni >> 8, tun_dst);
1348 /* Consume bad packet */
1353 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1354 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1358 dst_release((struct dst_entry *)tun_dst);
1360 /* Return non vxlan pkt */
1364 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1366 struct vxlan_dev *vxlan = netdev_priv(dev);
1367 struct arphdr *parp;
1370 struct neighbour *n;
1372 if (dev->flags & IFF_NOARP)
1375 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1376 dev->stats.tx_dropped++;
1379 parp = arp_hdr(skb);
1381 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1382 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1383 parp->ar_pro != htons(ETH_P_IP) ||
1384 parp->ar_op != htons(ARPOP_REQUEST) ||
1385 parp->ar_hln != dev->addr_len ||
1388 arpptr = (u8 *)parp + sizeof(struct arphdr);
1390 arpptr += dev->addr_len; /* sha */
1391 memcpy(&sip, arpptr, sizeof(sip));
1392 arpptr += sizeof(sip);
1393 arpptr += dev->addr_len; /* tha */
1394 memcpy(&tip, arpptr, sizeof(tip));
1396 if (ipv4_is_loopback(tip) ||
1397 ipv4_is_multicast(tip))
1400 n = neigh_lookup(&arp_tbl, &tip, dev);
1403 struct vxlan_fdb *f;
1404 struct sk_buff *reply;
1406 if (!(n->nud_state & NUD_CONNECTED)) {
1411 f = vxlan_find_mac(vxlan, n->ha);
1412 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1413 /* bridge-local neighbor */
1418 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1426 skb_reset_mac_header(reply);
1427 __skb_pull(reply, skb_network_offset(reply));
1428 reply->ip_summed = CHECKSUM_UNNECESSARY;
1429 reply->pkt_type = PACKET_HOST;
1431 if (netif_rx_ni(reply) == NET_RX_DROP)
1432 dev->stats.rx_dropped++;
1433 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1434 union vxlan_addr ipa = {
1435 .sin.sin_addr.s_addr = tip,
1436 .sin.sin_family = AF_INET,
1439 vxlan_ip_miss(dev, &ipa);
1443 return NETDEV_TX_OK;
1446 #if IS_ENABLED(CONFIG_IPV6)
1447 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1448 struct neighbour *n, bool isrouter)
1450 struct net_device *dev = request->dev;
1451 struct sk_buff *reply;
1452 struct nd_msg *ns, *na;
1453 struct ipv6hdr *pip6;
1455 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1462 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1463 sizeof(*na) + na_olen + dev->needed_tailroom;
1464 reply = alloc_skb(len, GFP_ATOMIC);
1468 reply->protocol = htons(ETH_P_IPV6);
1470 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1471 skb_push(reply, sizeof(struct ethhdr));
1472 skb_set_mac_header(reply, 0);
1474 ns = (struct nd_msg *)skb_transport_header(request);
1476 daddr = eth_hdr(request)->h_source;
1477 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1478 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1479 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1480 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1485 /* Ethernet header */
1486 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1487 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1488 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1489 reply->protocol = htons(ETH_P_IPV6);
1491 skb_pull(reply, sizeof(struct ethhdr));
1492 skb_set_network_header(reply, 0);
1493 skb_put(reply, sizeof(struct ipv6hdr));
1497 pip6 = ipv6_hdr(reply);
1498 memset(pip6, 0, sizeof(struct ipv6hdr));
1500 pip6->priority = ipv6_hdr(request)->priority;
1501 pip6->nexthdr = IPPROTO_ICMPV6;
1502 pip6->hop_limit = 255;
1503 pip6->daddr = ipv6_hdr(request)->saddr;
1504 pip6->saddr = *(struct in6_addr *)n->primary_key;
1506 skb_pull(reply, sizeof(struct ipv6hdr));
1507 skb_set_transport_header(reply, 0);
1509 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1511 /* Neighbor Advertisement */
1512 memset(na, 0, sizeof(*na)+na_olen);
1513 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1514 na->icmph.icmp6_router = isrouter;
1515 na->icmph.icmp6_override = 1;
1516 na->icmph.icmp6_solicited = 1;
1517 na->target = ns->target;
1518 ether_addr_copy(&na->opt[2], n->ha);
1519 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1520 na->opt[1] = na_olen >> 3;
1522 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1523 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1524 csum_partial(na, sizeof(*na)+na_olen, 0));
1526 pip6->payload_len = htons(sizeof(*na)+na_olen);
1528 skb_push(reply, sizeof(struct ipv6hdr));
1530 reply->ip_summed = CHECKSUM_UNNECESSARY;
1535 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1537 struct vxlan_dev *vxlan = netdev_priv(dev);
1539 const struct ipv6hdr *iphdr;
1540 const struct in6_addr *saddr, *daddr;
1541 struct neighbour *n;
1542 struct inet6_dev *in6_dev;
1544 in6_dev = __in6_dev_get(dev);
1548 iphdr = ipv6_hdr(skb);
1549 saddr = &iphdr->saddr;
1550 daddr = &iphdr->daddr;
1552 msg = (struct nd_msg *)skb_transport_header(skb);
1553 if (msg->icmph.icmp6_code != 0 ||
1554 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1557 if (ipv6_addr_loopback(daddr) ||
1558 ipv6_addr_is_multicast(&msg->target))
1561 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1564 struct vxlan_fdb *f;
1565 struct sk_buff *reply;
1567 if (!(n->nud_state & NUD_CONNECTED)) {
1572 f = vxlan_find_mac(vxlan, n->ha);
1573 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1574 /* bridge-local neighbor */
1579 reply = vxlan_na_create(skb, n,
1580 !!(f ? f->flags & NTF_ROUTER : 0));
1587 if (netif_rx_ni(reply) == NET_RX_DROP)
1588 dev->stats.rx_dropped++;
1590 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1591 union vxlan_addr ipa = {
1592 .sin6.sin6_addr = msg->target,
1593 .sin6.sin6_family = AF_INET6,
1596 vxlan_ip_miss(dev, &ipa);
1601 return NETDEV_TX_OK;
1605 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1607 struct vxlan_dev *vxlan = netdev_priv(dev);
1608 struct neighbour *n;
1610 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1614 switch (ntohs(eth_hdr(skb)->h_proto)) {
1619 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1622 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1623 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1624 union vxlan_addr ipa = {
1625 .sin.sin_addr.s_addr = pip->daddr,
1626 .sin.sin_family = AF_INET,
1629 vxlan_ip_miss(dev, &ipa);
1635 #if IS_ENABLED(CONFIG_IPV6)
1638 struct ipv6hdr *pip6;
1640 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1642 pip6 = ipv6_hdr(skb);
1643 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1644 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1645 union vxlan_addr ipa = {
1646 .sin6.sin6_addr = pip6->daddr,
1647 .sin6.sin6_family = AF_INET6,
1650 vxlan_ip_miss(dev, &ipa);
1664 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1666 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1668 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1677 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1678 struct vxlan_metadata *md)
1680 struct vxlanhdr_gbp *gbp;
1685 gbp = (struct vxlanhdr_gbp *)vxh;
1686 vxh->vx_flags |= htonl(VXLAN_HF_GBP);
1688 if (md->gbp & VXLAN_GBP_DONT_LEARN)
1689 gbp->dont_learn = 1;
1691 if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1692 gbp->policy_applied = 1;
1694 gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1697 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1698 int iphdr_len, __be32 vni,
1699 struct vxlan_metadata *md, u32 vxflags,
1702 struct vxlanhdr *vxh;
1705 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1706 u16 hdrlen = sizeof(struct vxlanhdr);
1708 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1709 skb->ip_summed == CHECKSUM_PARTIAL) {
1710 int csum_start = skb_checksum_start_offset(skb);
1712 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1713 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1714 (skb->csum_offset == offsetof(struct udphdr, check) ||
1715 skb->csum_offset == offsetof(struct tcphdr, check)))
1716 type |= SKB_GSO_TUNNEL_REMCSUM;
1719 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1720 + VXLAN_HLEN + iphdr_len
1721 + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1723 /* Need space for new headers (invalidates iph ptr) */
1724 err = skb_cow_head(skb, min_headroom);
1725 if (unlikely(err)) {
1730 skb = vlan_hwaccel_push_inside(skb);
1734 skb = iptunnel_handle_offloads(skb, type);
1736 return PTR_ERR(skb);
1738 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1739 vxh->vx_flags = htonl(VXLAN_HF_VNI);
1742 if (type & SKB_GSO_TUNNEL_REMCSUM) {
1743 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1746 if (skb->csum_offset == offsetof(struct udphdr, check))
1747 data |= VXLAN_RCO_UDP;
1749 vxh->vx_vni |= htonl(data);
1750 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1752 if (!skb_is_gso(skb)) {
1753 skb->ip_summed = CHECKSUM_NONE;
1754 skb->encapsulation = 0;
1758 if (vxflags & VXLAN_F_GBP)
1759 vxlan_build_gbp_hdr(vxh, vxflags, md);
1761 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1765 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1766 struct sk_buff *skb, int oif, u8 tos,
1767 __be32 daddr, __be32 *saddr,
1768 struct dst_cache *dst_cache,
1769 struct ip_tunnel_info *info)
1771 struct rtable *rt = NULL;
1772 bool use_cache = false;
1775 /* when the ip_tunnel_info is availble, the tos used for lookup is
1776 * packet independent, so we can use the cache
1778 if (dst_cache && !skb->mark && (!tos || info)) {
1780 rt = dst_cache_get_ip4(dst_cache, saddr);
1785 memset(&fl4, 0, sizeof(fl4));
1786 fl4.flowi4_oif = oif;
1787 fl4.flowi4_tos = RT_TOS(tos);
1788 fl4.flowi4_mark = skb->mark;
1789 fl4.flowi4_proto = IPPROTO_UDP;
1791 fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
1793 rt = ip_route_output_key(vxlan->net, &fl4);
1797 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1802 #if IS_ENABLED(CONFIG_IPV6)
1803 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1804 struct sk_buff *skb, int oif,
1805 const struct in6_addr *daddr,
1806 struct in6_addr *saddr,
1807 struct dst_cache *dst_cache)
1809 bool use_cache = false;
1810 struct dst_entry *ndst;
1814 if (dst_cache && !skb->mark) {
1816 ndst = dst_cache_get_ip6(dst_cache, saddr);
1821 memset(&fl6, 0, sizeof(fl6));
1822 fl6.flowi6_oif = oif;
1824 fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
1825 fl6.flowi6_mark = skb->mark;
1826 fl6.flowi6_proto = IPPROTO_UDP;
1828 err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1829 vxlan->vn6_sock->sock->sk,
1832 return ERR_PTR(err);
1836 dst_cache_set_ip6(dst_cache, ndst, saddr);
1841 /* Bypass encapsulation if the destination is local */
1842 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1843 struct vxlan_dev *dst_vxlan)
1845 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1846 union vxlan_addr loopback;
1847 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1848 struct net_device *dev = skb->dev;
1851 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1852 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1853 skb->pkt_type = PACKET_HOST;
1854 skb->encapsulation = 0;
1855 skb->dev = dst_vxlan->dev;
1856 __skb_pull(skb, skb_network_offset(skb));
1858 if (remote_ip->sa.sa_family == AF_INET) {
1859 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1860 loopback.sa.sa_family = AF_INET;
1861 #if IS_ENABLED(CONFIG_IPV6)
1863 loopback.sin6.sin6_addr = in6addr_loopback;
1864 loopback.sa.sa_family = AF_INET6;
1868 if (dst_vxlan->flags & VXLAN_F_LEARN)
1869 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1871 u64_stats_update_begin(&tx_stats->syncp);
1872 tx_stats->tx_packets++;
1873 tx_stats->tx_bytes += len;
1874 u64_stats_update_end(&tx_stats->syncp);
1876 if (netif_rx(skb) == NET_RX_SUCCESS) {
1877 u64_stats_update_begin(&rx_stats->syncp);
1878 rx_stats->rx_packets++;
1879 rx_stats->rx_bytes += len;
1880 u64_stats_update_end(&rx_stats->syncp);
1882 dev->stats.rx_dropped++;
1886 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1887 struct vxlan_rdst *rdst, bool did_rsc)
1889 struct ip_tunnel_info *info;
1890 struct vxlan_dev *vxlan = netdev_priv(dev);
1892 struct rtable *rt = NULL;
1893 const struct iphdr *old_iph;
1894 union vxlan_addr *dst;
1895 union vxlan_addr remote_ip;
1896 struct vxlan_metadata _md;
1897 struct vxlan_metadata *md = &_md;
1898 __be16 src_port = 0, dst_port;
1903 u32 flags = vxlan->flags;
1904 bool udp_sum = false;
1905 bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1907 info = skb_tunnel_info(skb);
1910 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1911 vni = rdst->remote_vni;
1912 dst = &rdst->remote_ip;
1915 WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1919 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1920 vni = be64_to_cpu(info->key.tun_id);
1921 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1922 if (remote_ip.sa.sa_family == AF_INET)
1923 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1925 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1929 if (vxlan_addr_any(dst)) {
1931 /* short-circuited back to local bridge */
1932 vxlan_encap_bypass(skb, vxlan, vxlan);
1938 old_iph = ip_hdr(skb);
1940 ttl = vxlan->cfg.ttl;
1941 if (!ttl && vxlan_addr_multicast(dst))
1944 tos = vxlan->cfg.tos;
1946 tos = ip_tunnel_get_dsfield(old_iph, skb);
1948 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1949 vxlan->cfg.port_max, true);
1952 ttl = info->key.ttl;
1953 tos = info->key.tos;
1954 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
1956 if (info->options_len)
1957 md = ip_tunnel_info_opts(info);
1959 md->gbp = skb->mark;
1962 if (dst->sa.sa_family == AF_INET) {
1965 if (!vxlan->vn4_sock)
1967 sk = vxlan->vn4_sock->sock->sk;
1970 if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
1973 udp_sum = !!(flags & VXLAN_F_UDP_CSUM);
1976 rt = vxlan_get_route(vxlan, skb,
1977 rdst ? rdst->remote_ifindex : 0, tos,
1978 dst->sin.sin_addr.s_addr, &saddr,
1979 rdst ? &rdst->dst_cache : NULL, info);
1981 netdev_dbg(dev, "no route to %pI4\n",
1982 &dst->sin.sin_addr.s_addr);
1983 dev->stats.tx_carrier_errors++;
1987 if (rt->dst.dev == dev) {
1988 netdev_dbg(dev, "circular route to %pI4\n",
1989 &dst->sin.sin_addr.s_addr);
1990 dev->stats.collisions++;
1994 /* Bypass encapsulation if the destination is local */
1995 if (rt->rt_flags & RTCF_LOCAL &&
1996 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1997 struct vxlan_dev *dst_vxlan;
2000 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2001 dst->sa.sa_family, dst_port,
2005 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2009 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2010 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2011 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2012 htonl(vni << 8), md, flags, udp_sum);
2016 udp_tunnel_xmit_skb(rt, sk, skb, saddr,
2017 dst->sin.sin_addr.s_addr, tos, ttl, df,
2018 src_port, dst_port, xnet, !udp_sum);
2019 #if IS_ENABLED(CONFIG_IPV6)
2021 struct dst_entry *ndst;
2022 struct in6_addr saddr;
2025 if (!vxlan->vn6_sock)
2027 sk = vxlan->vn6_sock->sock->sk;
2029 ndst = vxlan6_get_route(vxlan, skb,
2030 rdst ? rdst->remote_ifindex : 0,
2031 &dst->sin6.sin6_addr, &saddr,
2032 rdst ? &rdst->dst_cache : NULL);
2034 netdev_dbg(dev, "no route to %pI6\n",
2035 &dst->sin6.sin6_addr);
2036 dev->stats.tx_carrier_errors++;
2040 if (ndst->dev == dev) {
2041 netdev_dbg(dev, "circular route to %pI6\n",
2042 &dst->sin6.sin6_addr);
2044 dev->stats.collisions++;
2048 /* Bypass encapsulation if the destination is local */
2049 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2050 if (rt6i_flags & RTF_LOCAL &&
2051 !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2052 struct vxlan_dev *dst_vxlan;
2055 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2056 dst->sa.sa_family, dst_port,
2060 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2065 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2067 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2068 skb_scrub_packet(skb, xnet);
2069 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2070 htonl(vni << 8), md, flags, udp_sum);
2075 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2076 &saddr, &dst->sin6.sin6_addr,
2077 0, ttl, src_port, dst_port, !udp_sum);
2084 dev->stats.tx_dropped++;
2088 /* skb is already freed. */
2093 dev->stats.tx_errors++;
2098 /* Transmit local packets over Vxlan
2100 * Outer IP header inherits ECN and DF from inner header.
2101 * Outer UDP destination is the VXLAN assigned port.
2102 * source port is based on hash of flow
2104 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2106 struct vxlan_dev *vxlan = netdev_priv(dev);
2107 const struct ip_tunnel_info *info;
2109 bool did_rsc = false;
2110 struct vxlan_rdst *rdst, *fdst = NULL;
2111 struct vxlan_fdb *f;
2113 info = skb_tunnel_info(skb);
2115 skb_reset_mac_header(skb);
2118 if ((vxlan->flags & VXLAN_F_PROXY)) {
2119 if (ntohs(eth->h_proto) == ETH_P_ARP)
2120 return arp_reduce(dev, skb);
2121 #if IS_ENABLED(CONFIG_IPV6)
2122 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2123 pskb_may_pull(skb, sizeof(struct ipv6hdr)
2124 + sizeof(struct nd_msg)) &&
2125 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2128 msg = (struct nd_msg *)skb_transport_header(skb);
2129 if (msg->icmph.icmp6_code == 0 &&
2130 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2131 return neigh_reduce(dev, skb);
2137 if (vxlan->flags & VXLAN_F_COLLECT_METADATA &&
2138 info && info->mode & IP_TUNNEL_INFO_TX) {
2139 vxlan_xmit_one(skb, dev, NULL, false);
2140 return NETDEV_TX_OK;
2143 f = vxlan_find_mac(vxlan, eth->h_dest);
2146 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2147 (ntohs(eth->h_proto) == ETH_P_IP ||
2148 ntohs(eth->h_proto) == ETH_P_IPV6)) {
2149 did_rsc = route_shortcircuit(dev, skb);
2151 f = vxlan_find_mac(vxlan, eth->h_dest);
2155 f = vxlan_find_mac(vxlan, all_zeros_mac);
2157 if ((vxlan->flags & VXLAN_F_L2MISS) &&
2158 !is_multicast_ether_addr(eth->h_dest))
2159 vxlan_fdb_miss(vxlan, eth->h_dest);
2161 dev->stats.tx_dropped++;
2163 return NETDEV_TX_OK;
2167 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2168 struct sk_buff *skb1;
2174 skb1 = skb_clone(skb, GFP_ATOMIC);
2176 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2180 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2183 return NETDEV_TX_OK;
2186 /* Walk the forwarding table and purge stale entries */
2187 static void vxlan_cleanup(unsigned long arg)
2189 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2190 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2193 if (!netif_running(vxlan->dev))
2196 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2197 struct hlist_node *p, *n;
2199 spin_lock_bh(&vxlan->hash_lock);
2200 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2202 = container_of(p, struct vxlan_fdb, hlist);
2203 unsigned long timeout;
2205 if (f->state & NUD_PERMANENT)
2208 timeout = f->used + vxlan->cfg.age_interval * HZ;
2209 if (time_before_eq(timeout, jiffies)) {
2210 netdev_dbg(vxlan->dev,
2211 "garbage collect %pM\n",
2213 f->state = NUD_STALE;
2214 vxlan_fdb_destroy(vxlan, f);
2215 } else if (time_before(timeout, next_timer))
2216 next_timer = timeout;
2218 spin_unlock_bh(&vxlan->hash_lock);
2221 mod_timer(&vxlan->age_timer, next_timer);
2224 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2226 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2227 __u32 vni = vxlan->default_dst.remote_vni;
2229 spin_lock(&vn->sock_lock);
2230 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2231 spin_unlock(&vn->sock_lock);
2234 /* Setup stats when device is created */
2235 static int vxlan_init(struct net_device *dev)
2237 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2244 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2246 struct vxlan_fdb *f;
2248 spin_lock_bh(&vxlan->hash_lock);
2249 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2251 vxlan_fdb_destroy(vxlan, f);
2252 spin_unlock_bh(&vxlan->hash_lock);
2255 static void vxlan_uninit(struct net_device *dev)
2257 struct vxlan_dev *vxlan = netdev_priv(dev);
2259 vxlan_fdb_delete_default(vxlan);
2261 free_percpu(dev->tstats);
2264 /* Start ageing timer and join group when device is brought up */
2265 static int vxlan_open(struct net_device *dev)
2267 struct vxlan_dev *vxlan = netdev_priv(dev);
2270 ret = vxlan_sock_add(vxlan);
2274 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2275 ret = vxlan_igmp_join(vxlan);
2276 if (ret == -EADDRINUSE)
2279 vxlan_sock_release(vxlan);
2284 if (vxlan->cfg.age_interval)
2285 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2290 /* Purge the forwarding table */
2291 static void vxlan_flush(struct vxlan_dev *vxlan)
2295 spin_lock_bh(&vxlan->hash_lock);
2296 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2297 struct hlist_node *p, *n;
2298 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2300 = container_of(p, struct vxlan_fdb, hlist);
2301 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2302 if (!is_zero_ether_addr(f->eth_addr))
2303 vxlan_fdb_destroy(vxlan, f);
2306 spin_unlock_bh(&vxlan->hash_lock);
2309 /* Cleanup timer and forwarding table on shutdown */
2310 static int vxlan_stop(struct net_device *dev)
2312 struct vxlan_dev *vxlan = netdev_priv(dev);
2313 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2316 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2317 !vxlan_group_used(vn, vxlan))
2318 ret = vxlan_igmp_leave(vxlan);
2320 del_timer_sync(&vxlan->age_timer);
2323 vxlan_sock_release(vxlan);
2328 /* Stub, nothing needs to be done. */
2329 static void vxlan_set_multicast_list(struct net_device *dev)
2333 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2335 struct vxlan_dev *vxlan = netdev_priv(dev);
2336 struct vxlan_rdst *dst = &vxlan->default_dst;
2337 struct net_device *lowerdev;
2340 lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2341 if (lowerdev == NULL)
2342 return eth_change_mtu(dev, new_mtu);
2344 if (dst->remote_ip.sa.sa_family == AF_INET6)
2345 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2347 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2349 if (new_mtu < 68 || new_mtu > max_mtu)
2356 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2358 struct vxlan_dev *vxlan = netdev_priv(dev);
2359 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2360 __be16 sport, dport;
2362 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2363 vxlan->cfg.port_max, true);
2364 dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2366 if (ip_tunnel_info_af(info) == AF_INET) {
2369 if (!vxlan->vn4_sock)
2371 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2372 info->key.u.ipv4.dst,
2373 &info->key.u.ipv4.src, NULL, info);
2378 #if IS_ENABLED(CONFIG_IPV6)
2379 struct dst_entry *ndst;
2381 if (!vxlan->vn6_sock)
2383 ndst = vxlan6_get_route(vxlan, skb, 0,
2384 &info->key.u.ipv6.dst,
2385 &info->key.u.ipv6.src, NULL);
2387 return PTR_ERR(ndst);
2389 #else /* !CONFIG_IPV6 */
2390 return -EPFNOSUPPORT;
2393 info->key.tp_src = sport;
2394 info->key.tp_dst = dport;
2398 static const struct net_device_ops vxlan_netdev_ops = {
2399 .ndo_init = vxlan_init,
2400 .ndo_uninit = vxlan_uninit,
2401 .ndo_open = vxlan_open,
2402 .ndo_stop = vxlan_stop,
2403 .ndo_start_xmit = vxlan_xmit,
2404 .ndo_get_stats64 = ip_tunnel_get_stats64,
2405 .ndo_set_rx_mode = vxlan_set_multicast_list,
2406 .ndo_change_mtu = vxlan_change_mtu,
2407 .ndo_validate_addr = eth_validate_addr,
2408 .ndo_set_mac_address = eth_mac_addr,
2409 .ndo_fdb_add = vxlan_fdb_add,
2410 .ndo_fdb_del = vxlan_fdb_delete,
2411 .ndo_fdb_dump = vxlan_fdb_dump,
2412 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2415 /* Info for udev, that this is a virtual tunnel endpoint */
2416 static struct device_type vxlan_type = {
2420 /* Calls the ndo_add_vxlan_port of the caller in order to
2421 * supply the listening VXLAN udp ports. Callers are expected
2422 * to implement the ndo_add_vxlan_port.
2424 void vxlan_get_rx_port(struct net_device *dev)
2426 struct vxlan_sock *vs;
2427 struct net *net = dev_net(dev);
2428 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2429 sa_family_t sa_family;
2433 spin_lock(&vn->sock_lock);
2434 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2435 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2436 port = inet_sk(vs->sock->sk)->inet_sport;
2437 sa_family = vxlan_get_sk_family(vs);
2438 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2442 spin_unlock(&vn->sock_lock);
2444 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2446 /* Initialize the device structure. */
2447 static void vxlan_setup(struct net_device *dev)
2449 struct vxlan_dev *vxlan = netdev_priv(dev);
2452 eth_hw_addr_random(dev);
2455 dev->netdev_ops = &vxlan_netdev_ops;
2456 dev->destructor = free_netdev;
2457 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2459 dev->features |= NETIF_F_LLTX;
2460 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2461 dev->features |= NETIF_F_RXCSUM;
2462 dev->features |= NETIF_F_GSO_SOFTWARE;
2464 dev->vlan_features = dev->features;
2465 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2466 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2467 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2468 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2469 netif_keep_dst(dev);
2470 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
2472 INIT_LIST_HEAD(&vxlan->next);
2473 spin_lock_init(&vxlan->hash_lock);
2475 init_timer_deferrable(&vxlan->age_timer);
2476 vxlan->age_timer.function = vxlan_cleanup;
2477 vxlan->age_timer.data = (unsigned long) vxlan;
2479 vxlan->cfg.dst_port = htons(vxlan_port);
2483 gro_cells_init(&vxlan->gro_cells, dev);
2485 for (h = 0; h < FDB_HASH_SIZE; ++h)
2486 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2489 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2490 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2491 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2492 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2493 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2494 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2495 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2496 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2497 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2498 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2499 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2500 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2501 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2502 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2503 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2504 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2505 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2506 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 },
2507 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2508 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 },
2509 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
2510 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
2511 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2512 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2513 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, },
2514 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG },
2517 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2519 if (tb[IFLA_ADDRESS]) {
2520 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2521 pr_debug("invalid link address (not ethernet)\n");
2525 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2526 pr_debug("invalid all zero ethernet address\n");
2527 return -EADDRNOTAVAIL;
2534 if (data[IFLA_VXLAN_ID]) {
2535 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2536 if (id >= VXLAN_VID_MASK)
2540 if (data[IFLA_VXLAN_PORT_RANGE]) {
2541 const struct ifla_vxlan_port_range *p
2542 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2544 if (ntohs(p->high) < ntohs(p->low)) {
2545 pr_debug("port range %u .. %u not valid\n",
2546 ntohs(p->low), ntohs(p->high));
2554 static void vxlan_get_drvinfo(struct net_device *netdev,
2555 struct ethtool_drvinfo *drvinfo)
2557 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2558 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2561 static const struct ethtool_ops vxlan_ethtool_ops = {
2562 .get_drvinfo = vxlan_get_drvinfo,
2563 .get_link = ethtool_op_get_link,
2566 static void vxlan_del_work(struct work_struct *work)
2568 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2569 udp_tunnel_sock_release(vs->sock);
2573 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2574 __be16 port, u32 flags)
2576 struct socket *sock;
2577 struct udp_port_cfg udp_conf;
2580 memset(&udp_conf, 0, sizeof(udp_conf));
2583 udp_conf.family = AF_INET6;
2584 udp_conf.use_udp6_rx_checksums =
2585 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2586 udp_conf.ipv6_v6only = 1;
2588 udp_conf.family = AF_INET;
2591 udp_conf.local_udp_port = port;
2593 /* Open UDP socket */
2594 err = udp_sock_create(net, &udp_conf, &sock);
2596 return ERR_PTR(err);
2601 /* Create new listen socket if needed */
2602 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2603 __be16 port, u32 flags)
2605 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2606 struct vxlan_sock *vs;
2607 struct socket *sock;
2609 struct udp_tunnel_sock_cfg tunnel_cfg;
2611 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2613 return ERR_PTR(-ENOMEM);
2615 for (h = 0; h < VNI_HASH_SIZE; ++h)
2616 INIT_HLIST_HEAD(&vs->vni_list[h]);
2618 INIT_WORK(&vs->del_work, vxlan_del_work);
2620 sock = vxlan_create_sock(net, ipv6, port, flags);
2622 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2625 return ERR_CAST(sock);
2629 atomic_set(&vs->refcnt, 1);
2630 vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2632 /* Initialize the vxlan udp offloads structure */
2633 vs->udp_offloads.port = port;
2634 vs->udp_offloads.callbacks.gro_receive = vxlan_gro_receive;
2635 vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2637 spin_lock(&vn->sock_lock);
2638 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2639 vxlan_notify_add_rx_port(vs);
2640 spin_unlock(&vn->sock_lock);
2642 /* Mark socket as an encapsulation socket. */
2643 tunnel_cfg.sk_user_data = vs;
2644 tunnel_cfg.encap_type = 1;
2645 tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2646 tunnel_cfg.encap_destroy = NULL;
2648 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2653 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2655 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2656 struct vxlan_sock *vs = NULL;
2658 if (!vxlan->cfg.no_share) {
2659 spin_lock(&vn->sock_lock);
2660 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2661 vxlan->cfg.dst_port, vxlan->flags);
2662 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2663 spin_unlock(&vn->sock_lock);
2666 spin_unlock(&vn->sock_lock);
2669 vs = vxlan_socket_create(vxlan->net, ipv6,
2670 vxlan->cfg.dst_port, vxlan->flags);
2673 #if IS_ENABLED(CONFIG_IPV6)
2675 vxlan->vn6_sock = vs;
2678 vxlan->vn4_sock = vs;
2679 vxlan_vs_add_dev(vs, vxlan);
2683 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2685 bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2686 bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2689 vxlan->vn4_sock = NULL;
2690 #if IS_ENABLED(CONFIG_IPV6)
2691 vxlan->vn6_sock = NULL;
2692 if (ipv6 || metadata)
2693 ret = __vxlan_sock_add(vxlan, true);
2695 if (!ret && (!ipv6 || metadata))
2696 ret = __vxlan_sock_add(vxlan, false);
2698 vxlan_sock_release(vxlan);
2702 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2703 struct vxlan_config *conf)
2705 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2706 struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2707 struct vxlan_rdst *dst = &vxlan->default_dst;
2708 unsigned short needed_headroom = ETH_HLEN;
2710 bool use_ipv6 = false;
2711 __be16 default_port = vxlan->cfg.dst_port;
2713 vxlan->net = src_net;
2715 dst->remote_vni = conf->vni;
2717 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2719 /* Unless IPv6 is explicitly requested, assume IPv4 */
2720 if (!dst->remote_ip.sa.sa_family)
2721 dst->remote_ip.sa.sa_family = AF_INET;
2723 if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2724 vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2725 if (!IS_ENABLED(CONFIG_IPV6))
2726 return -EPFNOSUPPORT;
2728 vxlan->flags |= VXLAN_F_IPV6;
2731 if (conf->remote_ifindex) {
2732 struct net_device *lowerdev
2733 = __dev_get_by_index(src_net, conf->remote_ifindex);
2735 dst->remote_ifindex = conf->remote_ifindex;
2738 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2742 #if IS_ENABLED(CONFIG_IPV6)
2744 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2745 if (idev && idev->cnf.disable_ipv6) {
2746 pr_info("IPv6 is disabled via sysctl\n");
2753 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2755 needed_headroom = lowerdev->hard_header_len;
2758 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2759 needed_headroom += VXLAN6_HEADROOM;
2761 needed_headroom += VXLAN_HEADROOM;
2762 dev->needed_headroom = needed_headroom;
2764 memcpy(&vxlan->cfg, conf, sizeof(*conf));
2765 if (!vxlan->cfg.dst_port)
2766 vxlan->cfg.dst_port = default_port;
2767 vxlan->flags |= conf->flags;
2769 if (!vxlan->cfg.age_interval)
2770 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2772 list_for_each_entry(tmp, &vn->vxlan_list, next) {
2773 if (tmp->cfg.vni == conf->vni &&
2774 (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2775 tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2776 tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2777 (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2778 (vxlan->flags & VXLAN_F_RCV_FLAGS))
2782 dev->ethtool_ops = &vxlan_ethtool_ops;
2784 /* create an fdb entry for a valid default destination */
2785 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2786 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2787 &vxlan->default_dst.remote_ip,
2788 NUD_REACHABLE|NUD_PERMANENT,
2789 NLM_F_EXCL|NLM_F_CREATE,
2790 vxlan->cfg.dst_port,
2791 vxlan->default_dst.remote_vni,
2792 vxlan->default_dst.remote_ifindex,
2798 err = register_netdevice(dev);
2800 vxlan_fdb_delete_default(vxlan);
2804 list_add(&vxlan->next, &vn->vxlan_list);
2809 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2810 u8 name_assign_type, struct vxlan_config *conf)
2812 struct nlattr *tb[IFLA_MAX+1];
2813 struct net_device *dev;
2816 memset(&tb, 0, sizeof(tb));
2818 dev = rtnl_create_link(net, name, name_assign_type,
2819 &vxlan_link_ops, tb);
2823 err = vxlan_dev_configure(net, dev, conf);
2826 return ERR_PTR(err);
2831 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2833 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2834 struct nlattr *tb[], struct nlattr *data[])
2836 struct vxlan_config conf;
2839 memset(&conf, 0, sizeof(conf));
2841 if (data[IFLA_VXLAN_ID])
2842 conf.vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2844 if (data[IFLA_VXLAN_GROUP]) {
2845 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2846 } else if (data[IFLA_VXLAN_GROUP6]) {
2847 if (!IS_ENABLED(CONFIG_IPV6))
2848 return -EPFNOSUPPORT;
2850 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2851 conf.remote_ip.sa.sa_family = AF_INET6;
2854 if (data[IFLA_VXLAN_LOCAL]) {
2855 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2856 conf.saddr.sa.sa_family = AF_INET;
2857 } else if (data[IFLA_VXLAN_LOCAL6]) {
2858 if (!IS_ENABLED(CONFIG_IPV6))
2859 return -EPFNOSUPPORT;
2861 /* TODO: respect scope id */
2862 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2863 conf.saddr.sa.sa_family = AF_INET6;
2866 if (data[IFLA_VXLAN_LINK])
2867 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2869 if (data[IFLA_VXLAN_TOS])
2870 conf.tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2872 if (data[IFLA_VXLAN_TTL])
2873 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2875 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2876 conf.flags |= VXLAN_F_LEARN;
2878 if (data[IFLA_VXLAN_AGEING])
2879 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2881 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2882 conf.flags |= VXLAN_F_PROXY;
2884 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2885 conf.flags |= VXLAN_F_RSC;
2887 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2888 conf.flags |= VXLAN_F_L2MISS;
2890 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2891 conf.flags |= VXLAN_F_L3MISS;
2893 if (data[IFLA_VXLAN_LIMIT])
2894 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2896 if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2897 nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2898 conf.flags |= VXLAN_F_COLLECT_METADATA;
2900 if (data[IFLA_VXLAN_PORT_RANGE]) {
2901 const struct ifla_vxlan_port_range *p
2902 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2903 conf.port_min = ntohs(p->low);
2904 conf.port_max = ntohs(p->high);
2907 if (data[IFLA_VXLAN_PORT])
2908 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2910 if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2911 conf.flags |= VXLAN_F_UDP_CSUM;
2913 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2914 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2915 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2917 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2918 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2919 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2921 if (data[IFLA_VXLAN_REMCSUM_TX] &&
2922 nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2923 conf.flags |= VXLAN_F_REMCSUM_TX;
2925 if (data[IFLA_VXLAN_REMCSUM_RX] &&
2926 nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2927 conf.flags |= VXLAN_F_REMCSUM_RX;
2929 if (data[IFLA_VXLAN_GBP])
2930 conf.flags |= VXLAN_F_GBP;
2932 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2933 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2935 err = vxlan_dev_configure(src_net, dev, &conf);
2938 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
2942 pr_info("IPv6 is disabled via sysctl\n");
2946 pr_info("duplicate VNI %u\n", conf.vni);
2953 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2955 struct vxlan_dev *vxlan = netdev_priv(dev);
2956 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2958 spin_lock(&vn->sock_lock);
2959 if (!hlist_unhashed(&vxlan->hlist))
2960 hlist_del_rcu(&vxlan->hlist);
2961 spin_unlock(&vn->sock_lock);
2963 gro_cells_destroy(&vxlan->gro_cells);
2964 list_del(&vxlan->next);
2965 unregister_netdevice_queue(dev, head);
2968 static size_t vxlan_get_size(const struct net_device *dev)
2971 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
2972 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2973 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2974 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2975 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
2976 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
2977 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
2978 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
2979 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
2980 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
2981 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
2982 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */
2983 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2984 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2985 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2986 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2987 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2988 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2989 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2990 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
2991 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
2995 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2997 const struct vxlan_dev *vxlan = netdev_priv(dev);
2998 const struct vxlan_rdst *dst = &vxlan->default_dst;
2999 struct ifla_vxlan_port_range ports = {
3000 .low = htons(vxlan->cfg.port_min),
3001 .high = htons(vxlan->cfg.port_max),
3004 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
3005 goto nla_put_failure;
3007 if (!vxlan_addr_any(&dst->remote_ip)) {
3008 if (dst->remote_ip.sa.sa_family == AF_INET) {
3009 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3010 dst->remote_ip.sin.sin_addr.s_addr))
3011 goto nla_put_failure;
3012 #if IS_ENABLED(CONFIG_IPV6)
3014 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3015 &dst->remote_ip.sin6.sin6_addr))
3016 goto nla_put_failure;
3021 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3022 goto nla_put_failure;
3024 if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3025 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3026 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3027 vxlan->cfg.saddr.sin.sin_addr.s_addr))
3028 goto nla_put_failure;
3029 #if IS_ENABLED(CONFIG_IPV6)
3031 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3032 &vxlan->cfg.saddr.sin6.sin6_addr))
3033 goto nla_put_failure;
3038 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3039 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3040 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3041 !!(vxlan->flags & VXLAN_F_LEARN)) ||
3042 nla_put_u8(skb, IFLA_VXLAN_PROXY,
3043 !!(vxlan->flags & VXLAN_F_PROXY)) ||
3044 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3045 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3046 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3047 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3048 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3049 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3050 !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3051 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3052 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3053 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3054 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3055 !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
3056 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3057 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3058 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3059 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3060 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3061 !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3062 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3063 !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3064 goto nla_put_failure;
3066 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3067 goto nla_put_failure;
3069 if (vxlan->flags & VXLAN_F_GBP &&
3070 nla_put_flag(skb, IFLA_VXLAN_GBP))
3071 goto nla_put_failure;
3073 if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3074 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3075 goto nla_put_failure;
3083 static struct net *vxlan_get_link_net(const struct net_device *dev)
3085 struct vxlan_dev *vxlan = netdev_priv(dev);
3090 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3092 .maxtype = IFLA_VXLAN_MAX,
3093 .policy = vxlan_policy,
3094 .priv_size = sizeof(struct vxlan_dev),
3095 .setup = vxlan_setup,
3096 .validate = vxlan_validate,
3097 .newlink = vxlan_newlink,
3098 .dellink = vxlan_dellink,
3099 .get_size = vxlan_get_size,
3100 .fill_info = vxlan_fill_info,
3101 .get_link_net = vxlan_get_link_net,
3104 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3105 struct net_device *dev)
3107 struct vxlan_dev *vxlan, *next;
3108 LIST_HEAD(list_kill);
3110 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3111 struct vxlan_rdst *dst = &vxlan->default_dst;
3113 /* In case we created vxlan device with carrier
3114 * and we loose the carrier due to module unload
3115 * we also need to remove vxlan device. In other
3116 * cases, it's not necessary and remote_ifindex
3117 * is 0 here, so no matches.
3119 if (dst->remote_ifindex == dev->ifindex)
3120 vxlan_dellink(vxlan->dev, &list_kill);
3123 unregister_netdevice_many(&list_kill);
3126 static int vxlan_lowerdev_event(struct notifier_block *unused,
3127 unsigned long event, void *ptr)
3129 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3130 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3132 if (event == NETDEV_UNREGISTER)
3133 vxlan_handle_lowerdev_unregister(vn, dev);
3138 static struct notifier_block vxlan_notifier_block __read_mostly = {
3139 .notifier_call = vxlan_lowerdev_event,
3142 static __net_init int vxlan_init_net(struct net *net)
3144 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3147 INIT_LIST_HEAD(&vn->vxlan_list);
3148 spin_lock_init(&vn->sock_lock);
3150 for (h = 0; h < PORT_HASH_SIZE; ++h)
3151 INIT_HLIST_HEAD(&vn->sock_list[h]);
3156 static void __net_exit vxlan_exit_net(struct net *net)
3158 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3159 struct vxlan_dev *vxlan, *next;
3160 struct net_device *dev, *aux;
3164 for_each_netdev_safe(net, dev, aux)
3165 if (dev->rtnl_link_ops == &vxlan_link_ops)
3166 unregister_netdevice_queue(dev, &list);
3168 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3169 /* If vxlan->dev is in the same netns, it has already been added
3170 * to the list by the previous loop.
3172 if (!net_eq(dev_net(vxlan->dev), net)) {
3173 gro_cells_destroy(&vxlan->gro_cells);
3174 unregister_netdevice_queue(vxlan->dev, &list);
3178 unregister_netdevice_many(&list);
3182 static struct pernet_operations vxlan_net_ops = {
3183 .init = vxlan_init_net,
3184 .exit = vxlan_exit_net,
3185 .id = &vxlan_net_id,
3186 .size = sizeof(struct vxlan_net),
3189 static int __init vxlan_init_module(void)
3193 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3197 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3199 rc = register_pernet_subsys(&vxlan_net_ops);
3203 rc = register_netdevice_notifier(&vxlan_notifier_block);
3207 rc = rtnl_link_register(&vxlan_link_ops);
3213 unregister_netdevice_notifier(&vxlan_notifier_block);
3215 unregister_pernet_subsys(&vxlan_net_ops);
3217 destroy_workqueue(vxlan_wq);
3220 late_initcall(vxlan_init_module);
3222 static void __exit vxlan_cleanup_module(void)
3224 rtnl_link_unregister(&vxlan_link_ops);
3225 unregister_netdevice_notifier(&vxlan_notifier_block);
3226 destroy_workqueue(vxlan_wq);
3227 unregister_pernet_subsys(&vxlan_net_ops);
3228 /* rcu_barrier() is called by netns */
3230 module_exit(vxlan_cleanup_module);
3232 MODULE_LICENSE("GPL");
3233 MODULE_VERSION(VXLAN_VERSION);
3234 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3235 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3236 MODULE_ALIAS_RTNL_LINK("vxlan");