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>
45 #include <net/udp_tunnel.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];
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 rd->remote_port = port;
485 rd->remote_vni = vni;
486 rd->remote_ifindex = ifindex;
490 /* Add/update destinations for multicast */
491 static int vxlan_fdb_append(struct vxlan_fdb *f,
492 union vxlan_addr *ip, __be16 port, __u32 vni,
493 __u32 ifindex, struct vxlan_rdst **rdp)
495 struct vxlan_rdst *rd;
497 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
501 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
505 rd->remote_port = port;
506 rd->remote_vni = vni;
507 rd->remote_ifindex = ifindex;
509 list_add_tail_rcu(&rd->list, &f->remotes);
515 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
517 struct vxlanhdr *vh, size_t hdrlen,
518 u32 data, struct gro_remcsum *grc,
521 size_t start, offset;
523 if (skb->remcsum_offload)
526 if (!NAPI_GRO_CB(skb)->csum_valid)
529 start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
530 offset = start + ((data & VXLAN_RCO_UDP) ?
531 offsetof(struct udphdr, check) :
532 offsetof(struct tcphdr, check));
534 vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
535 start, offset, grc, nopartial);
537 skb->remcsum_offload = 1;
542 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
544 struct udp_offload *uoff)
546 struct sk_buff *p, **pp = NULL;
547 struct vxlanhdr *vh, *vh2;
548 unsigned int hlen, off_vx;
550 struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
553 struct gro_remcsum grc;
555 skb_gro_remcsum_init(&grc);
557 off_vx = skb_gro_offset(skb);
558 hlen = off_vx + sizeof(*vh);
559 vh = skb_gro_header_fast(skb, off_vx);
560 if (skb_gro_header_hard(skb, hlen)) {
561 vh = skb_gro_header_slow(skb, hlen, off_vx);
566 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
568 flags = ntohl(vh->vx_flags);
570 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
571 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
572 ntohl(vh->vx_vni), &grc,
574 VXLAN_F_REMCSUM_NOPARTIAL));
580 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
584 for (p = *head; p; p = p->next) {
585 if (!NAPI_GRO_CB(p)->same_flow)
588 vh2 = (struct vxlanhdr *)(p->data + off_vx);
589 if (vh->vx_flags != vh2->vx_flags ||
590 vh->vx_vni != vh2->vx_vni) {
591 NAPI_GRO_CB(p)->same_flow = 0;
596 pp = call_gro_receive(eth_gro_receive, head, skb);
599 skb_gro_remcsum_cleanup(skb, &grc);
600 NAPI_GRO_CB(skb)->flush |= flush;
605 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
606 struct udp_offload *uoff)
608 udp_tunnel_gro_complete(skb, nhoff);
610 return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
613 /* Notify netdevs that UDP port started listening */
614 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
616 struct net_device *dev;
617 struct sock *sk = vs->sock->sk;
618 struct net *net = sock_net(sk);
619 sa_family_t sa_family = vxlan_get_sk_family(vs);
620 __be16 port = inet_sk(sk)->inet_sport;
623 if (sa_family == AF_INET) {
624 err = udp_add_offload(&vs->udp_offloads);
626 pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
630 for_each_netdev_rcu(net, dev) {
631 if (dev->netdev_ops->ndo_add_vxlan_port)
632 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
638 /* Notify netdevs that UDP port is no more listening */
639 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
641 struct net_device *dev;
642 struct sock *sk = vs->sock->sk;
643 struct net *net = sock_net(sk);
644 sa_family_t sa_family = vxlan_get_sk_family(vs);
645 __be16 port = inet_sk(sk)->inet_sport;
648 for_each_netdev_rcu(net, dev) {
649 if (dev->netdev_ops->ndo_del_vxlan_port)
650 dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
655 if (sa_family == AF_INET)
656 udp_del_offload(&vs->udp_offloads);
659 /* Add new entry to forwarding table -- assumes lock held */
660 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
661 const u8 *mac, union vxlan_addr *ip,
662 __u16 state, __u16 flags,
663 __be16 port, __u32 vni, __u32 ifindex,
666 struct vxlan_rdst *rd = NULL;
670 f = __vxlan_find_mac(vxlan, mac);
672 if (flags & NLM_F_EXCL) {
673 netdev_dbg(vxlan->dev,
674 "lost race to create %pM\n", mac);
677 if (f->state != state) {
679 f->updated = jiffies;
682 if (f->flags != ndm_flags) {
683 f->flags = ndm_flags;
684 f->updated = jiffies;
687 if ((flags & NLM_F_REPLACE)) {
688 /* Only change unicasts */
689 if (!(is_multicast_ether_addr(f->eth_addr) ||
690 is_zero_ether_addr(f->eth_addr))) {
691 notify |= vxlan_fdb_replace(f, ip, port, vni,
696 if ((flags & NLM_F_APPEND) &&
697 (is_multicast_ether_addr(f->eth_addr) ||
698 is_zero_ether_addr(f->eth_addr))) {
699 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
707 if (!(flags & NLM_F_CREATE))
710 if (vxlan->cfg.addrmax &&
711 vxlan->addrcnt >= vxlan->cfg.addrmax)
714 /* Disallow replace to add a multicast entry */
715 if ((flags & NLM_F_REPLACE) &&
716 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
719 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
720 f = kmalloc(sizeof(*f), GFP_ATOMIC);
726 f->flags = ndm_flags;
727 f->updated = f->used = jiffies;
728 INIT_LIST_HEAD(&f->remotes);
729 memcpy(f->eth_addr, mac, ETH_ALEN);
731 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
734 hlist_add_head_rcu(&f->hlist,
735 vxlan_fdb_head(vxlan, mac));
740 rd = first_remote_rtnl(f);
741 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
747 static void vxlan_fdb_free(struct rcu_head *head)
749 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
750 struct vxlan_rdst *rd, *nd;
752 list_for_each_entry_safe(rd, nd, &f->remotes, list)
757 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
759 netdev_dbg(vxlan->dev,
760 "delete %pM\n", f->eth_addr);
763 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
765 hlist_del_rcu(&f->hlist);
766 call_rcu(&f->rcu, vxlan_fdb_free);
769 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
770 union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
772 struct net *net = dev_net(vxlan->dev);
776 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
780 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
781 if (remote->sa.sa_family == AF_INET) {
782 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
783 ip->sa.sa_family = AF_INET;
784 #if IS_ENABLED(CONFIG_IPV6)
786 ip->sin6.sin6_addr = in6addr_any;
787 ip->sa.sa_family = AF_INET6;
793 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
795 *port = nla_get_be16(tb[NDA_PORT]);
797 *port = vxlan->cfg.dst_port;
801 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
803 *vni = nla_get_u32(tb[NDA_VNI]);
805 *vni = vxlan->default_dst.remote_vni;
808 if (tb[NDA_IFINDEX]) {
809 struct net_device *tdev;
811 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
813 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
814 tdev = __dev_get_by_index(net, *ifindex);
816 return -EADDRNOTAVAIL;
824 /* Add static entry (via netlink) */
825 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
826 struct net_device *dev,
827 const unsigned char *addr, u16 vid, u16 flags)
829 struct vxlan_dev *vxlan = netdev_priv(dev);
830 /* struct net *net = dev_net(vxlan->dev); */
836 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
837 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
842 if (tb[NDA_DST] == NULL)
845 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
849 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
850 return -EAFNOSUPPORT;
852 spin_lock_bh(&vxlan->hash_lock);
853 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
854 port, vni, ifindex, ndm->ndm_flags);
855 spin_unlock_bh(&vxlan->hash_lock);
860 /* Delete entry (via netlink) */
861 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
862 struct net_device *dev,
863 const unsigned char *addr, u16 vid)
865 struct vxlan_dev *vxlan = netdev_priv(dev);
867 struct vxlan_rdst *rd = NULL;
873 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
879 spin_lock_bh(&vxlan->hash_lock);
880 f = vxlan_find_mac(vxlan, addr);
884 if (!vxlan_addr_any(&ip)) {
885 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
892 /* remove a destination if it's not the only one on the list,
893 * otherwise destroy the fdb entry
895 if (rd && !list_is_singular(&f->remotes)) {
896 list_del_rcu(&rd->list);
897 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
902 vxlan_fdb_destroy(vxlan, f);
905 spin_unlock_bh(&vxlan->hash_lock);
910 /* Dump forwarding table */
911 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
912 struct net_device *dev,
913 struct net_device *filter_dev, int idx)
915 struct vxlan_dev *vxlan = netdev_priv(dev);
918 for (h = 0; h < FDB_HASH_SIZE; ++h) {
922 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
923 struct vxlan_rdst *rd;
925 list_for_each_entry_rcu(rd, &f->remotes, list) {
926 if (idx < cb->args[0])
929 err = vxlan_fdb_info(skb, vxlan, f,
930 NETLINK_CB(cb->skb).portid,
945 /* Watch incoming packets to learn mapping between Ethernet address
946 * and Tunnel endpoint.
947 * Return true if packet is bogus and should be dropped.
949 static bool vxlan_snoop(struct net_device *dev,
950 union vxlan_addr *src_ip, const u8 *src_mac)
952 struct vxlan_dev *vxlan = netdev_priv(dev);
955 f = vxlan_find_mac(vxlan, src_mac);
957 struct vxlan_rdst *rdst = first_remote_rcu(f);
959 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
962 /* Don't migrate static entries, drop packets */
963 if (f->state & NUD_NOARP)
968 "%pM migrated from %pIS to %pIS\n",
969 src_mac, &rdst->remote_ip.sa, &src_ip->sa);
971 rdst->remote_ip = *src_ip;
972 f->updated = jiffies;
973 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
975 /* learned new entry */
976 spin_lock(&vxlan->hash_lock);
978 /* close off race between vxlan_flush and incoming packets */
979 if (netif_running(dev))
980 vxlan_fdb_create(vxlan, src_mac, src_ip,
982 NLM_F_EXCL|NLM_F_CREATE,
984 vxlan->default_dst.remote_vni,
986 spin_unlock(&vxlan->hash_lock);
992 /* See if multicast group is already in use by other ID */
993 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
995 struct vxlan_dev *vxlan;
996 unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
998 /* The vxlan_sock is only used by dev, leaving group has
999 * no effect on other vxlan devices.
1001 if (family == AF_INET && dev->vn4_sock &&
1002 atomic_read(&dev->vn4_sock->refcnt) == 1)
1004 #if IS_ENABLED(CONFIG_IPV6)
1005 if (family == AF_INET6 && dev->vn6_sock &&
1006 atomic_read(&dev->vn6_sock->refcnt) == 1)
1010 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1011 if (!netif_running(vxlan->dev) || vxlan == dev)
1014 if (family == AF_INET && vxlan->vn4_sock != dev->vn4_sock)
1016 #if IS_ENABLED(CONFIG_IPV6)
1017 if (family == AF_INET6 && vxlan->vn6_sock != dev->vn6_sock)
1021 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1022 &dev->default_dst.remote_ip))
1025 if (vxlan->default_dst.remote_ifindex !=
1026 dev->default_dst.remote_ifindex)
1035 static void __vxlan_sock_release(struct vxlan_sock *vs)
1037 struct vxlan_net *vn;
1041 if (!atomic_dec_and_test(&vs->refcnt))
1044 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1045 spin_lock(&vn->sock_lock);
1046 hlist_del_rcu(&vs->hlist);
1047 vxlan_notify_del_rx_port(vs);
1048 spin_unlock(&vn->sock_lock);
1050 queue_work(vxlan_wq, &vs->del_work);
1053 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1055 __vxlan_sock_release(vxlan->vn4_sock);
1056 #if IS_ENABLED(CONFIG_IPV6)
1057 __vxlan_sock_release(vxlan->vn6_sock);
1061 /* Update multicast group membership when first VNI on
1062 * multicast address is brought up
1064 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1067 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1068 int ifindex = vxlan->default_dst.remote_ifindex;
1071 if (ip->sa.sa_family == AF_INET) {
1072 struct ip_mreqn mreq = {
1073 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1074 .imr_ifindex = ifindex,
1077 sk = vxlan->vn4_sock->sock->sk;
1079 ret = ip_mc_join_group(sk, &mreq);
1081 #if IS_ENABLED(CONFIG_IPV6)
1083 sk = vxlan->vn6_sock->sock->sk;
1085 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1086 &ip->sin6.sin6_addr);
1094 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1095 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1098 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1099 int ifindex = vxlan->default_dst.remote_ifindex;
1102 if (ip->sa.sa_family == AF_INET) {
1103 struct ip_mreqn mreq = {
1104 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1105 .imr_ifindex = ifindex,
1108 sk = vxlan->vn4_sock->sock->sk;
1110 ret = ip_mc_leave_group(sk, &mreq);
1112 #if IS_ENABLED(CONFIG_IPV6)
1114 sk = vxlan->vn6_sock->sock->sk;
1116 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1117 &ip->sin6.sin6_addr);
1125 static struct vxlanhdr *vxlan_remcsum(struct sk_buff *skb, struct vxlanhdr *vh,
1126 size_t hdrlen, u32 data, bool nopartial)
1128 size_t start, offset, plen;
1130 if (skb->remcsum_offload)
1133 start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
1134 offset = start + ((data & VXLAN_RCO_UDP) ?
1135 offsetof(struct udphdr, check) :
1136 offsetof(struct tcphdr, check));
1138 plen = hdrlen + offset + sizeof(u16);
1140 if (!pskb_may_pull(skb, plen))
1143 vh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1145 skb_remcsum_process(skb, (void *)vh + hdrlen, start, offset,
1151 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1152 struct vxlan_metadata *md, u32 vni,
1153 struct metadata_dst *tun_dst)
1155 struct iphdr *oip = NULL;
1156 struct ipv6hdr *oip6 = NULL;
1157 struct vxlan_dev *vxlan;
1158 struct pcpu_sw_netstats *stats;
1159 union vxlan_addr saddr;
1162 /* For flow based devices, map all packets to VNI 0 */
1163 if (vs->flags & VXLAN_F_COLLECT_METADATA)
1166 /* Is this VNI defined? */
1167 vxlan = vxlan_vs_find_vni(vs, vni);
1171 skb_reset_mac_header(skb);
1172 skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1173 skb->protocol = eth_type_trans(skb, vxlan->dev);
1174 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1176 /* Ignore packet loops (and multicast echo) */
1177 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1180 /* Get data from the outer IP header */
1181 if (vxlan_get_sk_family(vs) == AF_INET) {
1183 saddr.sin.sin_addr.s_addr = oip->saddr;
1184 saddr.sa.sa_family = AF_INET;
1185 #if IS_ENABLED(CONFIG_IPV6)
1187 oip6 = ipv6_hdr(skb);
1188 saddr.sin6.sin6_addr = oip6->saddr;
1189 saddr.sa.sa_family = AF_INET6;
1194 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1198 if ((vxlan->flags & VXLAN_F_LEARN) &&
1199 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1202 skb_reset_network_header(skb);
1203 /* In flow-based mode, GBP is carried in dst_metadata */
1204 if (!(vs->flags & VXLAN_F_COLLECT_METADATA))
1205 skb->mark = md->gbp;
1208 err = IP6_ECN_decapsulate(oip6, skb);
1210 err = IP_ECN_decapsulate(oip, skb);
1212 if (unlikely(err)) {
1213 if (log_ecn_error) {
1215 net_info_ratelimited("non-ECT from %pI6\n",
1218 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1219 &oip->saddr, oip->tos);
1222 ++vxlan->dev->stats.rx_frame_errors;
1223 ++vxlan->dev->stats.rx_errors;
1228 stats = this_cpu_ptr(vxlan->dev->tstats);
1229 u64_stats_update_begin(&stats->syncp);
1230 stats->rx_packets++;
1231 stats->rx_bytes += skb->len;
1232 u64_stats_update_end(&stats->syncp);
1234 gro_cells_receive(&vxlan->gro_cells, skb);
1239 dst_release((struct dst_entry *)tun_dst);
1241 /* Consume bad packet */
1245 /* Callback from net/ipv4/udp.c to receive packets */
1246 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1248 struct metadata_dst *tun_dst = NULL;
1249 struct vxlan_sock *vs;
1250 struct vxlanhdr *vxh;
1252 struct vxlan_metadata _md;
1253 struct vxlan_metadata *md = &_md;
1255 /* Need Vxlan and inner Ethernet header to be present */
1256 if (!pskb_may_pull(skb, VXLAN_HLEN))
1259 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1260 flags = ntohl(vxh->vx_flags);
1261 vni = ntohl(vxh->vx_vni);
1263 if (flags & VXLAN_HF_VNI) {
1264 flags &= ~VXLAN_HF_VNI;
1266 /* VNI flag always required to be set */
1270 if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1272 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1274 vs = rcu_dereference_sk_user_data(sk);
1278 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1279 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni,
1280 !!(vs->flags & VXLAN_F_REMCSUM_NOPARTIAL));
1284 flags &= ~VXLAN_HF_RCO;
1285 vni &= VXLAN_VNI_MASK;
1288 if (vxlan_collect_metadata(vs)) {
1289 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1290 cpu_to_be64(vni >> 8), sizeof(*md));
1295 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1297 memset(md, 0, sizeof(*md));
1300 /* For backwards compatibility, only allow reserved fields to be
1301 * used by VXLAN extensions if explicitly requested.
1303 if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1304 struct vxlanhdr_gbp *gbp;
1306 gbp = (struct vxlanhdr_gbp *)vxh;
1307 md->gbp = ntohs(gbp->policy_id);
1310 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1311 tun_dst->u.tun_info.options_len = sizeof(*md);
1314 if (gbp->dont_learn)
1315 md->gbp |= VXLAN_GBP_DONT_LEARN;
1317 if (gbp->policy_applied)
1318 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1320 flags &= ~VXLAN_GBP_USED_BITS;
1323 if (flags || vni & ~VXLAN_VNI_MASK) {
1324 /* If there are any unprocessed flags remaining treat
1325 * this as a malformed packet. This behavior diverges from
1326 * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1327 * in reserved fields are to be ignored. The approach here
1328 * maintains compatibility with previous stack code, and also
1329 * is more robust and provides a little more security in
1330 * adding extensions to VXLAN.
1336 vxlan_rcv(vs, skb, md, vni >> 8, tun_dst);
1340 /* Consume bad packet */
1345 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1346 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1350 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1352 struct vxlan_dev *vxlan = netdev_priv(dev);
1353 struct arphdr *parp;
1356 struct neighbour *n;
1358 if (dev->flags & IFF_NOARP)
1361 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1362 dev->stats.tx_dropped++;
1365 parp = arp_hdr(skb);
1367 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1368 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1369 parp->ar_pro != htons(ETH_P_IP) ||
1370 parp->ar_op != htons(ARPOP_REQUEST) ||
1371 parp->ar_hln != dev->addr_len ||
1374 arpptr = (u8 *)parp + sizeof(struct arphdr);
1376 arpptr += dev->addr_len; /* sha */
1377 memcpy(&sip, arpptr, sizeof(sip));
1378 arpptr += sizeof(sip);
1379 arpptr += dev->addr_len; /* tha */
1380 memcpy(&tip, arpptr, sizeof(tip));
1382 if (ipv4_is_loopback(tip) ||
1383 ipv4_is_multicast(tip))
1386 n = neigh_lookup(&arp_tbl, &tip, dev);
1389 struct vxlan_fdb *f;
1390 struct sk_buff *reply;
1392 if (!(n->nud_state & NUD_CONNECTED)) {
1397 f = vxlan_find_mac(vxlan, n->ha);
1398 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1399 /* bridge-local neighbor */
1404 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1412 skb_reset_mac_header(reply);
1413 __skb_pull(reply, skb_network_offset(reply));
1414 reply->ip_summed = CHECKSUM_UNNECESSARY;
1415 reply->pkt_type = PACKET_HOST;
1417 if (netif_rx_ni(reply) == NET_RX_DROP)
1418 dev->stats.rx_dropped++;
1419 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1420 union vxlan_addr ipa = {
1421 .sin.sin_addr.s_addr = tip,
1422 .sin.sin_family = AF_INET,
1425 vxlan_ip_miss(dev, &ipa);
1429 return NETDEV_TX_OK;
1432 #if IS_ENABLED(CONFIG_IPV6)
1433 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1434 struct neighbour *n, bool isrouter)
1436 struct net_device *dev = request->dev;
1437 struct sk_buff *reply;
1438 struct nd_msg *ns, *na;
1439 struct ipv6hdr *pip6;
1441 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1448 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1449 sizeof(*na) + na_olen + dev->needed_tailroom;
1450 reply = alloc_skb(len, GFP_ATOMIC);
1454 reply->protocol = htons(ETH_P_IPV6);
1456 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1457 skb_push(reply, sizeof(struct ethhdr));
1458 skb_set_mac_header(reply, 0);
1460 ns = (struct nd_msg *)skb_transport_header(request);
1462 daddr = eth_hdr(request)->h_source;
1463 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1464 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1465 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1466 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1471 /* Ethernet header */
1472 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1473 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1474 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1475 reply->protocol = htons(ETH_P_IPV6);
1477 skb_pull(reply, sizeof(struct ethhdr));
1478 skb_set_network_header(reply, 0);
1479 skb_put(reply, sizeof(struct ipv6hdr));
1483 pip6 = ipv6_hdr(reply);
1484 memset(pip6, 0, sizeof(struct ipv6hdr));
1486 pip6->priority = ipv6_hdr(request)->priority;
1487 pip6->nexthdr = IPPROTO_ICMPV6;
1488 pip6->hop_limit = 255;
1489 pip6->daddr = ipv6_hdr(request)->saddr;
1490 pip6->saddr = *(struct in6_addr *)n->primary_key;
1492 skb_pull(reply, sizeof(struct ipv6hdr));
1493 skb_set_transport_header(reply, 0);
1495 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1497 /* Neighbor Advertisement */
1498 memset(na, 0, sizeof(*na)+na_olen);
1499 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1500 na->icmph.icmp6_router = isrouter;
1501 na->icmph.icmp6_override = 1;
1502 na->icmph.icmp6_solicited = 1;
1503 na->target = ns->target;
1504 ether_addr_copy(&na->opt[2], n->ha);
1505 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1506 na->opt[1] = na_olen >> 3;
1508 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1509 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1510 csum_partial(na, sizeof(*na)+na_olen, 0));
1512 pip6->payload_len = htons(sizeof(*na)+na_olen);
1514 skb_push(reply, sizeof(struct ipv6hdr));
1516 reply->ip_summed = CHECKSUM_UNNECESSARY;
1521 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1523 struct vxlan_dev *vxlan = netdev_priv(dev);
1525 const struct ipv6hdr *iphdr;
1526 const struct in6_addr *saddr, *daddr;
1527 struct neighbour *n;
1528 struct inet6_dev *in6_dev;
1530 in6_dev = __in6_dev_get(dev);
1534 iphdr = ipv6_hdr(skb);
1535 saddr = &iphdr->saddr;
1536 daddr = &iphdr->daddr;
1538 msg = (struct nd_msg *)skb_transport_header(skb);
1539 if (msg->icmph.icmp6_code != 0 ||
1540 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1543 if (ipv6_addr_loopback(daddr) ||
1544 ipv6_addr_is_multicast(&msg->target))
1547 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1550 struct vxlan_fdb *f;
1551 struct sk_buff *reply;
1553 if (!(n->nud_state & NUD_CONNECTED)) {
1558 f = vxlan_find_mac(vxlan, n->ha);
1559 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1560 /* bridge-local neighbor */
1565 reply = vxlan_na_create(skb, n,
1566 !!(f ? f->flags & NTF_ROUTER : 0));
1573 if (netif_rx_ni(reply) == NET_RX_DROP)
1574 dev->stats.rx_dropped++;
1576 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1577 union vxlan_addr ipa = {
1578 .sin6.sin6_addr = msg->target,
1579 .sin6.sin6_family = AF_INET6,
1582 vxlan_ip_miss(dev, &ipa);
1587 return NETDEV_TX_OK;
1591 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1593 struct vxlan_dev *vxlan = netdev_priv(dev);
1594 struct neighbour *n;
1596 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1600 switch (ntohs(eth_hdr(skb)->h_proto)) {
1605 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1608 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1609 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1610 union vxlan_addr ipa = {
1611 .sin.sin_addr.s_addr = pip->daddr,
1612 .sin.sin_family = AF_INET,
1615 vxlan_ip_miss(dev, &ipa);
1621 #if IS_ENABLED(CONFIG_IPV6)
1624 struct ipv6hdr *pip6;
1626 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1628 pip6 = ipv6_hdr(skb);
1629 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1630 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1631 union vxlan_addr ipa = {
1632 .sin6.sin6_addr = pip6->daddr,
1633 .sin6.sin6_family = AF_INET6,
1636 vxlan_ip_miss(dev, &ipa);
1650 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1652 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1654 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1663 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1664 struct vxlan_metadata *md)
1666 struct vxlanhdr_gbp *gbp;
1671 gbp = (struct vxlanhdr_gbp *)vxh;
1672 vxh->vx_flags |= htonl(VXLAN_HF_GBP);
1674 if (md->gbp & VXLAN_GBP_DONT_LEARN)
1675 gbp->dont_learn = 1;
1677 if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1678 gbp->policy_applied = 1;
1680 gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1683 #if IS_ENABLED(CONFIG_IPV6)
1684 static int vxlan6_xmit_skb(struct dst_entry *dst, struct sock *sk,
1685 struct sk_buff *skb,
1686 struct net_device *dev, struct in6_addr *saddr,
1687 struct in6_addr *daddr, __u8 prio, __u8 ttl,
1688 __be16 src_port, __be16 dst_port, __be32 vni,
1689 struct vxlan_metadata *md, bool xnet, u32 vxflags)
1691 struct vxlanhdr *vxh;
1694 bool udp_sum = !(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX);
1695 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1696 u16 hdrlen = sizeof(struct vxlanhdr);
1698 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1699 skb->ip_summed == CHECKSUM_PARTIAL) {
1700 int csum_start = skb_checksum_start_offset(skb);
1702 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1703 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1704 (skb->csum_offset == offsetof(struct udphdr, check) ||
1705 skb->csum_offset == offsetof(struct tcphdr, check))) {
1707 type |= SKB_GSO_TUNNEL_REMCSUM;
1711 skb_scrub_packet(skb, xnet);
1713 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1714 + VXLAN_HLEN + sizeof(struct ipv6hdr)
1715 + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1717 /* Need space for new headers (invalidates iph ptr) */
1718 err = skb_cow_head(skb, min_headroom);
1719 if (unlikely(err)) {
1724 skb = vlan_hwaccel_push_inside(skb);
1725 if (WARN_ON(!skb)) {
1730 skb = iptunnel_handle_offloads(skb, udp_sum, type);
1736 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1737 vxh->vx_flags = htonl(VXLAN_HF_VNI);
1740 if (type & SKB_GSO_TUNNEL_REMCSUM) {
1741 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1744 if (skb->csum_offset == offsetof(struct udphdr, check))
1745 data |= VXLAN_RCO_UDP;
1747 vxh->vx_vni |= htonl(data);
1748 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1750 if (!skb_is_gso(skb)) {
1751 skb->ip_summed = CHECKSUM_NONE;
1752 skb->encapsulation = 0;
1756 if (vxflags & VXLAN_F_GBP)
1757 vxlan_build_gbp_hdr(vxh, vxflags, md);
1759 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1761 udp_tunnel6_xmit_skb(dst, sk, skb, dev, saddr, daddr, prio,
1762 ttl, src_port, dst_port,
1763 !!(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX));
1771 static int vxlan_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb,
1772 __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1773 __be16 src_port, __be16 dst_port, __be32 vni,
1774 struct vxlan_metadata *md, bool xnet, u32 vxflags)
1776 struct vxlanhdr *vxh;
1779 bool udp_sum = !!(vxflags & VXLAN_F_UDP_CSUM);
1780 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1781 u16 hdrlen = sizeof(struct vxlanhdr);
1783 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1784 skb->ip_summed == CHECKSUM_PARTIAL) {
1785 int csum_start = skb_checksum_start_offset(skb);
1787 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1788 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1789 (skb->csum_offset == offsetof(struct udphdr, check) ||
1790 skb->csum_offset == offsetof(struct tcphdr, check))) {
1792 type |= SKB_GSO_TUNNEL_REMCSUM;
1796 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1797 + VXLAN_HLEN + sizeof(struct iphdr)
1798 + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1800 /* Need space for new headers (invalidates iph ptr) */
1801 err = skb_cow_head(skb, min_headroom);
1802 if (unlikely(err)) {
1807 skb = vlan_hwaccel_push_inside(skb);
1811 skb = iptunnel_handle_offloads(skb, udp_sum, type);
1813 return PTR_ERR(skb);
1815 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1816 vxh->vx_flags = htonl(VXLAN_HF_VNI);
1819 if (type & SKB_GSO_TUNNEL_REMCSUM) {
1820 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1823 if (skb->csum_offset == offsetof(struct udphdr, check))
1824 data |= VXLAN_RCO_UDP;
1826 vxh->vx_vni |= htonl(data);
1827 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1829 if (!skb_is_gso(skb)) {
1830 skb->ip_summed = CHECKSUM_NONE;
1831 skb->encapsulation = 0;
1835 if (vxflags & VXLAN_F_GBP)
1836 vxlan_build_gbp_hdr(vxh, vxflags, md);
1838 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1840 return udp_tunnel_xmit_skb(rt, sk, skb, src, dst, tos,
1841 ttl, df, src_port, dst_port, xnet,
1842 !(vxflags & VXLAN_F_UDP_CSUM));
1845 #if IS_ENABLED(CONFIG_IPV6)
1846 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1847 struct sk_buff *skb, int oif,
1848 const struct in6_addr *daddr,
1849 struct in6_addr *saddr)
1851 struct dst_entry *ndst;
1855 memset(&fl6, 0, sizeof(fl6));
1856 fl6.flowi6_oif = oif;
1858 fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
1859 fl6.flowi6_mark = skb->mark;
1860 fl6.flowi6_proto = IPPROTO_UDP;
1862 err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1863 vxlan->vn6_sock->sock->sk,
1866 return ERR_PTR(err);
1873 /* Bypass encapsulation if the destination is local */
1874 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1875 struct vxlan_dev *dst_vxlan)
1877 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1878 union vxlan_addr loopback;
1879 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1880 struct net_device *dev = skb->dev;
1883 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1884 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1885 skb->pkt_type = PACKET_HOST;
1886 skb->encapsulation = 0;
1887 skb->dev = dst_vxlan->dev;
1888 __skb_pull(skb, skb_network_offset(skb));
1890 if (remote_ip->sa.sa_family == AF_INET) {
1891 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1892 loopback.sa.sa_family = AF_INET;
1893 #if IS_ENABLED(CONFIG_IPV6)
1895 loopback.sin6.sin6_addr = in6addr_loopback;
1896 loopback.sa.sa_family = AF_INET6;
1900 if (dst_vxlan->flags & VXLAN_F_LEARN)
1901 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1903 u64_stats_update_begin(&tx_stats->syncp);
1904 tx_stats->tx_packets++;
1905 tx_stats->tx_bytes += len;
1906 u64_stats_update_end(&tx_stats->syncp);
1908 if (netif_rx(skb) == NET_RX_SUCCESS) {
1909 u64_stats_update_begin(&rx_stats->syncp);
1910 rx_stats->rx_packets++;
1911 rx_stats->rx_bytes += len;
1912 u64_stats_update_end(&rx_stats->syncp);
1914 dev->stats.rx_dropped++;
1918 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1919 struct vxlan_rdst *rdst, bool did_rsc)
1921 struct ip_tunnel_info *info;
1922 struct vxlan_dev *vxlan = netdev_priv(dev);
1924 struct rtable *rt = NULL;
1925 const struct iphdr *old_iph;
1927 union vxlan_addr *dst;
1928 union vxlan_addr remote_ip;
1929 struct vxlan_metadata _md;
1930 struct vxlan_metadata *md = &_md;
1931 __be16 src_port = 0, dst_port;
1936 u32 flags = vxlan->flags;
1938 info = skb_tunnel_info(skb);
1941 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1942 vni = rdst->remote_vni;
1943 dst = &rdst->remote_ip;
1946 WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1950 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1951 vni = be64_to_cpu(info->key.tun_id);
1952 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1953 if (remote_ip.sa.sa_family == AF_INET)
1954 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1956 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1960 if (vxlan_addr_any(dst)) {
1962 /* short-circuited back to local bridge */
1963 vxlan_encap_bypass(skb, vxlan, vxlan);
1969 old_iph = ip_hdr(skb);
1971 ttl = vxlan->cfg.ttl;
1972 if (!ttl && vxlan_addr_multicast(dst))
1975 tos = vxlan->cfg.tos;
1977 tos = ip_tunnel_get_dsfield(old_iph, skb);
1979 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1980 vxlan->cfg.port_max, true);
1983 ttl = info->key.ttl;
1984 tos = info->key.tos;
1986 if (info->options_len)
1987 md = ip_tunnel_info_opts(info);
1989 md->gbp = skb->mark;
1992 if (dst->sa.sa_family == AF_INET) {
1993 if (!vxlan->vn4_sock)
1995 sk = vxlan->vn4_sock->sock->sk;
1998 if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2001 if (info->key.tun_flags & TUNNEL_CSUM)
2002 flags |= VXLAN_F_UDP_CSUM;
2004 flags &= ~VXLAN_F_UDP_CSUM;
2007 memset(&fl4, 0, sizeof(fl4));
2008 fl4.flowi4_oif = rdst ? rdst->remote_ifindex : 0;
2009 fl4.flowi4_tos = RT_TOS(tos);
2010 fl4.flowi4_mark = skb->mark;
2011 fl4.flowi4_proto = IPPROTO_UDP;
2012 fl4.daddr = dst->sin.sin_addr.s_addr;
2013 fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
2015 rt = ip_route_output_key(vxlan->net, &fl4);
2017 netdev_dbg(dev, "no route to %pI4\n",
2018 &dst->sin.sin_addr.s_addr);
2019 dev->stats.tx_carrier_errors++;
2023 if (rt->dst.dev == dev) {
2024 netdev_dbg(dev, "circular route to %pI4\n",
2025 &dst->sin.sin_addr.s_addr);
2026 dev->stats.collisions++;
2030 /* Bypass encapsulation if the destination is local */
2031 if (rt->rt_flags & RTCF_LOCAL &&
2032 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2033 struct vxlan_dev *dst_vxlan;
2036 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2037 dst->sa.sa_family, dst_port,
2041 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2045 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2046 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2047 err = vxlan_xmit_skb(rt, sk, skb, fl4.saddr,
2048 dst->sin.sin_addr.s_addr, tos, ttl, df,
2049 src_port, dst_port, htonl(vni << 8), md,
2050 !net_eq(vxlan->net, dev_net(vxlan->dev)),
2053 /* skb is already freed. */
2058 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
2059 #if IS_ENABLED(CONFIG_IPV6)
2061 struct dst_entry *ndst;
2062 struct in6_addr saddr;
2065 if (!vxlan->vn6_sock)
2067 sk = vxlan->vn6_sock->sock->sk;
2069 ndst = vxlan6_get_route(vxlan, skb,
2070 rdst ? rdst->remote_ifindex : 0,
2071 &dst->sin6.sin6_addr, &saddr);
2073 netdev_dbg(dev, "no route to %pI6\n",
2074 &dst->sin6.sin6_addr);
2075 dev->stats.tx_carrier_errors++;
2079 if (ndst->dev == dev) {
2080 netdev_dbg(dev, "circular route to %pI6\n",
2081 &dst->sin6.sin6_addr);
2083 dev->stats.collisions++;
2087 /* Bypass encapsulation if the destination is local */
2088 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2089 if (rt6i_flags & RTF_LOCAL &&
2090 !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2091 struct vxlan_dev *dst_vxlan;
2094 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2095 dst->sa.sa_family, dst_port,
2099 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2104 if (info->key.tun_flags & TUNNEL_CSUM)
2105 flags &= ~VXLAN_F_UDP_ZERO_CSUM6_TX;
2107 flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2110 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2111 err = vxlan6_xmit_skb(ndst, sk, skb, dev, &saddr, &dst->sin6.sin6_addr,
2112 0, ttl, src_port, dst_port, htonl(vni << 8), md,
2113 !net_eq(vxlan->net, dev_net(vxlan->dev)),
2121 dev->stats.tx_dropped++;
2127 dev->stats.tx_errors++;
2132 /* Transmit local packets over Vxlan
2134 * Outer IP header inherits ECN and DF from inner header.
2135 * Outer UDP destination is the VXLAN assigned port.
2136 * source port is based on hash of flow
2138 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2140 struct vxlan_dev *vxlan = netdev_priv(dev);
2141 const struct ip_tunnel_info *info;
2143 bool did_rsc = false;
2144 struct vxlan_rdst *rdst, *fdst = NULL;
2145 struct vxlan_fdb *f;
2147 info = skb_tunnel_info(skb);
2149 skb_reset_mac_header(skb);
2152 if ((vxlan->flags & VXLAN_F_PROXY)) {
2153 if (ntohs(eth->h_proto) == ETH_P_ARP)
2154 return arp_reduce(dev, skb);
2155 #if IS_ENABLED(CONFIG_IPV6)
2156 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2157 pskb_may_pull(skb, sizeof(struct ipv6hdr)
2158 + sizeof(struct nd_msg)) &&
2159 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2162 msg = (struct nd_msg *)skb_transport_header(skb);
2163 if (msg->icmph.icmp6_code == 0 &&
2164 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2165 return neigh_reduce(dev, skb);
2171 if (vxlan->flags & VXLAN_F_COLLECT_METADATA &&
2172 info && info->mode & IP_TUNNEL_INFO_TX) {
2173 vxlan_xmit_one(skb, dev, NULL, false);
2174 return NETDEV_TX_OK;
2177 f = vxlan_find_mac(vxlan, eth->h_dest);
2180 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2181 (ntohs(eth->h_proto) == ETH_P_IP ||
2182 ntohs(eth->h_proto) == ETH_P_IPV6)) {
2183 did_rsc = route_shortcircuit(dev, skb);
2185 f = vxlan_find_mac(vxlan, eth->h_dest);
2189 f = vxlan_find_mac(vxlan, all_zeros_mac);
2191 if ((vxlan->flags & VXLAN_F_L2MISS) &&
2192 !is_multicast_ether_addr(eth->h_dest))
2193 vxlan_fdb_miss(vxlan, eth->h_dest);
2195 dev->stats.tx_dropped++;
2197 return NETDEV_TX_OK;
2201 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2202 struct sk_buff *skb1;
2208 skb1 = skb_clone(skb, GFP_ATOMIC);
2210 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2214 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2217 return NETDEV_TX_OK;
2220 /* Walk the forwarding table and purge stale entries */
2221 static void vxlan_cleanup(unsigned long arg)
2223 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2224 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2227 if (!netif_running(vxlan->dev))
2230 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2231 struct hlist_node *p, *n;
2233 spin_lock_bh(&vxlan->hash_lock);
2234 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2236 = container_of(p, struct vxlan_fdb, hlist);
2237 unsigned long timeout;
2239 if (f->state & NUD_PERMANENT)
2242 timeout = f->used + vxlan->cfg.age_interval * HZ;
2243 if (time_before_eq(timeout, jiffies)) {
2244 netdev_dbg(vxlan->dev,
2245 "garbage collect %pM\n",
2247 f->state = NUD_STALE;
2248 vxlan_fdb_destroy(vxlan, f);
2249 } else if (time_before(timeout, next_timer))
2250 next_timer = timeout;
2252 spin_unlock_bh(&vxlan->hash_lock);
2255 mod_timer(&vxlan->age_timer, next_timer);
2258 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2260 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2261 __u32 vni = vxlan->default_dst.remote_vni;
2263 spin_lock(&vn->sock_lock);
2264 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2265 spin_unlock(&vn->sock_lock);
2268 /* Setup stats when device is created */
2269 static int vxlan_init(struct net_device *dev)
2271 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2278 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2280 struct vxlan_fdb *f;
2282 spin_lock_bh(&vxlan->hash_lock);
2283 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2285 vxlan_fdb_destroy(vxlan, f);
2286 spin_unlock_bh(&vxlan->hash_lock);
2289 static void vxlan_uninit(struct net_device *dev)
2291 struct vxlan_dev *vxlan = netdev_priv(dev);
2293 vxlan_fdb_delete_default(vxlan);
2295 free_percpu(dev->tstats);
2298 /* Start ageing timer and join group when device is brought up */
2299 static int vxlan_open(struct net_device *dev)
2301 struct vxlan_dev *vxlan = netdev_priv(dev);
2304 ret = vxlan_sock_add(vxlan);
2308 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2309 ret = vxlan_igmp_join(vxlan);
2310 if (ret == -EADDRINUSE)
2313 vxlan_sock_release(vxlan);
2318 if (vxlan->cfg.age_interval)
2319 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2324 /* Purge the forwarding table */
2325 static void vxlan_flush(struct vxlan_dev *vxlan)
2329 spin_lock_bh(&vxlan->hash_lock);
2330 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2331 struct hlist_node *p, *n;
2332 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2334 = container_of(p, struct vxlan_fdb, hlist);
2335 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2336 if (!is_zero_ether_addr(f->eth_addr))
2337 vxlan_fdb_destroy(vxlan, f);
2340 spin_unlock_bh(&vxlan->hash_lock);
2343 /* Cleanup timer and forwarding table on shutdown */
2344 static int vxlan_stop(struct net_device *dev)
2346 struct vxlan_dev *vxlan = netdev_priv(dev);
2347 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2350 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2351 !vxlan_group_used(vn, vxlan))
2352 ret = vxlan_igmp_leave(vxlan);
2354 del_timer_sync(&vxlan->age_timer);
2357 vxlan_sock_release(vxlan);
2362 /* Stub, nothing needs to be done. */
2363 static void vxlan_set_multicast_list(struct net_device *dev)
2367 static int __vxlan_change_mtu(struct net_device *dev,
2368 struct net_device *lowerdev,
2369 struct vxlan_rdst *dst, int new_mtu, bool strict)
2371 int max_mtu = IP_MAX_MTU;
2374 max_mtu = lowerdev->mtu;
2376 if (dst->remote_ip.sa.sa_family == AF_INET6)
2377 max_mtu -= VXLAN6_HEADROOM;
2379 max_mtu -= VXLAN_HEADROOM;
2384 if (new_mtu > max_mtu) {
2395 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2397 struct vxlan_dev *vxlan = netdev_priv(dev);
2398 struct vxlan_rdst *dst = &vxlan->default_dst;
2399 struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2400 dst->remote_ifindex);
2401 return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2404 static int egress_ipv4_tun_info(struct net_device *dev, struct sk_buff *skb,
2405 struct ip_tunnel_info *info,
2406 __be16 sport, __be16 dport)
2408 struct vxlan_dev *vxlan = netdev_priv(dev);
2412 memset(&fl4, 0, sizeof(fl4));
2413 fl4.flowi4_tos = RT_TOS(info->key.tos);
2414 fl4.flowi4_mark = skb->mark;
2415 fl4.flowi4_proto = IPPROTO_UDP;
2416 fl4.daddr = info->key.u.ipv4.dst;
2418 rt = ip_route_output_key(vxlan->net, &fl4);
2423 info->key.u.ipv4.src = fl4.saddr;
2424 info->key.tp_src = sport;
2425 info->key.tp_dst = dport;
2429 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2431 struct vxlan_dev *vxlan = netdev_priv(dev);
2432 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2433 __be16 sport, dport;
2435 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2436 vxlan->cfg.port_max, true);
2437 dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2439 if (ip_tunnel_info_af(info) == AF_INET) {
2440 if (!vxlan->vn4_sock)
2442 return egress_ipv4_tun_info(dev, skb, info, sport, dport);
2444 #if IS_ENABLED(CONFIG_IPV6)
2445 struct dst_entry *ndst;
2447 if (!vxlan->vn6_sock)
2449 ndst = vxlan6_get_route(vxlan, skb, 0,
2450 &info->key.u.ipv6.dst,
2451 &info->key.u.ipv6.src);
2453 return PTR_ERR(ndst);
2456 info->key.tp_src = sport;
2457 info->key.tp_dst = dport;
2458 #else /* !CONFIG_IPV6 */
2459 return -EPFNOSUPPORT;
2465 static const struct net_device_ops vxlan_netdev_ops = {
2466 .ndo_init = vxlan_init,
2467 .ndo_uninit = vxlan_uninit,
2468 .ndo_open = vxlan_open,
2469 .ndo_stop = vxlan_stop,
2470 .ndo_start_xmit = vxlan_xmit,
2471 .ndo_get_stats64 = ip_tunnel_get_stats64,
2472 .ndo_set_rx_mode = vxlan_set_multicast_list,
2473 .ndo_change_mtu = vxlan_change_mtu,
2474 .ndo_validate_addr = eth_validate_addr,
2475 .ndo_set_mac_address = eth_mac_addr,
2476 .ndo_fdb_add = vxlan_fdb_add,
2477 .ndo_fdb_del = vxlan_fdb_delete,
2478 .ndo_fdb_dump = vxlan_fdb_dump,
2479 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2482 /* Info for udev, that this is a virtual tunnel endpoint */
2483 static struct device_type vxlan_type = {
2487 /* Calls the ndo_add_vxlan_port of the caller in order to
2488 * supply the listening VXLAN udp ports. Callers are expected
2489 * to implement the ndo_add_vxlan_port.
2491 void vxlan_get_rx_port(struct net_device *dev)
2493 struct vxlan_sock *vs;
2494 struct net *net = dev_net(dev);
2495 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2496 sa_family_t sa_family;
2500 spin_lock(&vn->sock_lock);
2501 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2502 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2503 port = inet_sk(vs->sock->sk)->inet_sport;
2504 sa_family = vxlan_get_sk_family(vs);
2505 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2509 spin_unlock(&vn->sock_lock);
2511 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2513 /* Initialize the device structure. */
2514 static void vxlan_setup(struct net_device *dev)
2516 struct vxlan_dev *vxlan = netdev_priv(dev);
2519 eth_hw_addr_random(dev);
2522 dev->netdev_ops = &vxlan_netdev_ops;
2523 dev->destructor = free_netdev;
2524 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2526 dev->features |= NETIF_F_LLTX;
2527 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2528 dev->features |= NETIF_F_RXCSUM;
2529 dev->features |= NETIF_F_GSO_SOFTWARE;
2531 dev->vlan_features = dev->features;
2532 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2533 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2534 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2535 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2536 netif_keep_dst(dev);
2537 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
2539 INIT_LIST_HEAD(&vxlan->next);
2540 spin_lock_init(&vxlan->hash_lock);
2542 init_timer_deferrable(&vxlan->age_timer);
2543 vxlan->age_timer.function = vxlan_cleanup;
2544 vxlan->age_timer.data = (unsigned long) vxlan;
2546 vxlan->cfg.dst_port = htons(vxlan_port);
2550 gro_cells_init(&vxlan->gro_cells, dev);
2552 for (h = 0; h < FDB_HASH_SIZE; ++h)
2553 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2556 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2557 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2558 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2559 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2560 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2561 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2562 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2563 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2564 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2565 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2566 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2567 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2568 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2569 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2570 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2571 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2572 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2573 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 },
2574 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2575 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 },
2576 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
2577 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
2578 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2579 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2580 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, },
2581 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG },
2584 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2586 if (tb[IFLA_ADDRESS]) {
2587 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2588 pr_debug("invalid link address (not ethernet)\n");
2592 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2593 pr_debug("invalid all zero ethernet address\n");
2594 return -EADDRNOTAVAIL;
2601 if (data[IFLA_VXLAN_ID]) {
2602 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2603 if (id >= VXLAN_N_VID)
2607 if (data[IFLA_VXLAN_PORT_RANGE]) {
2608 const struct ifla_vxlan_port_range *p
2609 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2611 if (ntohs(p->high) < ntohs(p->low)) {
2612 pr_debug("port range %u .. %u not valid\n",
2613 ntohs(p->low), ntohs(p->high));
2621 static void vxlan_get_drvinfo(struct net_device *netdev,
2622 struct ethtool_drvinfo *drvinfo)
2624 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2625 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2628 static const struct ethtool_ops vxlan_ethtool_ops = {
2629 .get_drvinfo = vxlan_get_drvinfo,
2630 .get_link = ethtool_op_get_link,
2633 static void vxlan_del_work(struct work_struct *work)
2635 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2636 udp_tunnel_sock_release(vs->sock);
2640 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2641 __be16 port, u32 flags)
2643 struct socket *sock;
2644 struct udp_port_cfg udp_conf;
2647 memset(&udp_conf, 0, sizeof(udp_conf));
2650 udp_conf.family = AF_INET6;
2651 udp_conf.use_udp6_rx_checksums =
2652 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2653 udp_conf.ipv6_v6only = 1;
2655 udp_conf.family = AF_INET;
2658 udp_conf.local_udp_port = port;
2660 /* Open UDP socket */
2661 err = udp_sock_create(net, &udp_conf, &sock);
2663 return ERR_PTR(err);
2668 /* Create new listen socket if needed */
2669 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2670 __be16 port, u32 flags)
2672 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2673 struct vxlan_sock *vs;
2674 struct socket *sock;
2676 struct udp_tunnel_sock_cfg tunnel_cfg;
2678 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2680 return ERR_PTR(-ENOMEM);
2682 for (h = 0; h < VNI_HASH_SIZE; ++h)
2683 INIT_HLIST_HEAD(&vs->vni_list[h]);
2685 INIT_WORK(&vs->del_work, vxlan_del_work);
2687 sock = vxlan_create_sock(net, ipv6, port, flags);
2689 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2692 return ERR_CAST(sock);
2696 atomic_set(&vs->refcnt, 1);
2697 vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2699 /* Initialize the vxlan udp offloads structure */
2700 vs->udp_offloads.port = port;
2701 vs->udp_offloads.callbacks.gro_receive = vxlan_gro_receive;
2702 vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2704 spin_lock(&vn->sock_lock);
2705 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2706 vxlan_notify_add_rx_port(vs);
2707 spin_unlock(&vn->sock_lock);
2709 /* Mark socket as an encapsulation socket. */
2710 tunnel_cfg.sk_user_data = vs;
2711 tunnel_cfg.encap_type = 1;
2712 tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2713 tunnel_cfg.encap_destroy = NULL;
2715 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2720 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2722 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2723 struct vxlan_sock *vs = NULL;
2725 if (!vxlan->cfg.no_share) {
2726 spin_lock(&vn->sock_lock);
2727 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2728 vxlan->cfg.dst_port, vxlan->flags);
2729 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2730 spin_unlock(&vn->sock_lock);
2733 spin_unlock(&vn->sock_lock);
2736 vs = vxlan_socket_create(vxlan->net, ipv6,
2737 vxlan->cfg.dst_port, vxlan->flags);
2740 #if IS_ENABLED(CONFIG_IPV6)
2742 vxlan->vn6_sock = vs;
2745 vxlan->vn4_sock = vs;
2746 vxlan_vs_add_dev(vs, vxlan);
2750 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2752 bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2753 bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2756 vxlan->vn4_sock = NULL;
2757 #if IS_ENABLED(CONFIG_IPV6)
2758 vxlan->vn6_sock = NULL;
2759 if (ipv6 || metadata)
2760 ret = __vxlan_sock_add(vxlan, true);
2762 if (!ret && (!ipv6 || metadata))
2763 ret = __vxlan_sock_add(vxlan, false);
2765 vxlan_sock_release(vxlan);
2769 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2770 struct vxlan_config *conf)
2772 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2773 struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2774 struct vxlan_rdst *dst = &vxlan->default_dst;
2775 unsigned short needed_headroom = ETH_HLEN;
2777 bool use_ipv6 = false;
2778 __be16 default_port = vxlan->cfg.dst_port;
2779 struct net_device *lowerdev = NULL;
2781 vxlan->net = src_net;
2783 dst->remote_vni = conf->vni;
2785 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2787 /* Unless IPv6 is explicitly requested, assume IPv4 */
2788 if (!dst->remote_ip.sa.sa_family)
2789 dst->remote_ip.sa.sa_family = AF_INET;
2791 if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2792 vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2793 if (!IS_ENABLED(CONFIG_IPV6))
2794 return -EPFNOSUPPORT;
2796 vxlan->flags |= VXLAN_F_IPV6;
2799 if (conf->remote_ifindex) {
2800 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2801 dst->remote_ifindex = conf->remote_ifindex;
2804 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2808 #if IS_ENABLED(CONFIG_IPV6)
2810 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2811 if (idev && idev->cnf.disable_ipv6) {
2812 pr_info("IPv6 is disabled via sysctl\n");
2819 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2821 needed_headroom = lowerdev->hard_header_len;
2825 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2830 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2831 needed_headroom += VXLAN6_HEADROOM;
2833 needed_headroom += VXLAN_HEADROOM;
2834 dev->needed_headroom = needed_headroom;
2836 memcpy(&vxlan->cfg, conf, sizeof(*conf));
2837 if (!vxlan->cfg.dst_port)
2838 vxlan->cfg.dst_port = default_port;
2839 vxlan->flags |= conf->flags;
2841 if (!vxlan->cfg.age_interval)
2842 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2844 list_for_each_entry(tmp, &vn->vxlan_list, next) {
2845 if (tmp->cfg.vni == conf->vni &&
2846 (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2847 tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2848 tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2849 (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2850 (vxlan->flags & VXLAN_F_RCV_FLAGS))
2854 dev->ethtool_ops = &vxlan_ethtool_ops;
2856 /* create an fdb entry for a valid default destination */
2857 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2858 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2859 &vxlan->default_dst.remote_ip,
2860 NUD_REACHABLE|NUD_PERMANENT,
2861 NLM_F_EXCL|NLM_F_CREATE,
2862 vxlan->cfg.dst_port,
2863 vxlan->default_dst.remote_vni,
2864 vxlan->default_dst.remote_ifindex,
2870 err = register_netdevice(dev);
2872 vxlan_fdb_delete_default(vxlan);
2876 list_add(&vxlan->next, &vn->vxlan_list);
2881 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2882 u8 name_assign_type, struct vxlan_config *conf)
2884 struct nlattr *tb[IFLA_MAX+1];
2885 struct net_device *dev;
2888 memset(&tb, 0, sizeof(tb));
2890 dev = rtnl_create_link(net, name, name_assign_type,
2891 &vxlan_link_ops, tb);
2895 err = vxlan_dev_configure(net, dev, conf);
2898 return ERR_PTR(err);
2903 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2905 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2906 struct nlattr *tb[], struct nlattr *data[])
2908 struct vxlan_config conf;
2911 memset(&conf, 0, sizeof(conf));
2913 if (data[IFLA_VXLAN_ID])
2914 conf.vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2916 if (data[IFLA_VXLAN_GROUP]) {
2917 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2918 } else if (data[IFLA_VXLAN_GROUP6]) {
2919 if (!IS_ENABLED(CONFIG_IPV6))
2920 return -EPFNOSUPPORT;
2922 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2923 conf.remote_ip.sa.sa_family = AF_INET6;
2926 if (data[IFLA_VXLAN_LOCAL]) {
2927 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2928 conf.saddr.sa.sa_family = AF_INET;
2929 } else if (data[IFLA_VXLAN_LOCAL6]) {
2930 if (!IS_ENABLED(CONFIG_IPV6))
2931 return -EPFNOSUPPORT;
2933 /* TODO: respect scope id */
2934 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2935 conf.saddr.sa.sa_family = AF_INET6;
2938 if (data[IFLA_VXLAN_LINK])
2939 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2941 if (data[IFLA_VXLAN_TOS])
2942 conf.tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2944 if (data[IFLA_VXLAN_TTL])
2945 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2947 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2948 conf.flags |= VXLAN_F_LEARN;
2950 if (data[IFLA_VXLAN_AGEING])
2951 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2953 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2954 conf.flags |= VXLAN_F_PROXY;
2956 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2957 conf.flags |= VXLAN_F_RSC;
2959 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2960 conf.flags |= VXLAN_F_L2MISS;
2962 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2963 conf.flags |= VXLAN_F_L3MISS;
2965 if (data[IFLA_VXLAN_LIMIT])
2966 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2968 if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2969 nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2970 conf.flags |= VXLAN_F_COLLECT_METADATA;
2972 if (data[IFLA_VXLAN_PORT_RANGE]) {
2973 const struct ifla_vxlan_port_range *p
2974 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2975 conf.port_min = ntohs(p->low);
2976 conf.port_max = ntohs(p->high);
2979 if (data[IFLA_VXLAN_PORT])
2980 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2982 if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2983 conf.flags |= VXLAN_F_UDP_CSUM;
2985 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2986 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2987 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2989 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2990 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2991 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2993 if (data[IFLA_VXLAN_REMCSUM_TX] &&
2994 nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2995 conf.flags |= VXLAN_F_REMCSUM_TX;
2997 if (data[IFLA_VXLAN_REMCSUM_RX] &&
2998 nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2999 conf.flags |= VXLAN_F_REMCSUM_RX;
3001 if (data[IFLA_VXLAN_GBP])
3002 conf.flags |= VXLAN_F_GBP;
3004 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3005 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3008 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3010 err = vxlan_dev_configure(src_net, dev, &conf);
3013 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
3017 pr_info("IPv6 is disabled via sysctl\n");
3021 pr_info("duplicate VNI %u\n", conf.vni);
3028 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3030 struct vxlan_dev *vxlan = netdev_priv(dev);
3031 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3033 spin_lock(&vn->sock_lock);
3034 if (!hlist_unhashed(&vxlan->hlist))
3035 hlist_del_rcu(&vxlan->hlist);
3036 spin_unlock(&vn->sock_lock);
3038 gro_cells_destroy(&vxlan->gro_cells);
3039 list_del(&vxlan->next);
3040 unregister_netdevice_queue(dev, head);
3043 static size_t vxlan_get_size(const struct net_device *dev)
3046 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
3047 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3048 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3049 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3050 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
3051 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
3052 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
3053 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
3054 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
3055 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
3056 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
3057 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */
3058 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3059 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3060 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3061 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3062 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3063 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3064 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3065 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3066 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3070 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3072 const struct vxlan_dev *vxlan = netdev_priv(dev);
3073 const struct vxlan_rdst *dst = &vxlan->default_dst;
3074 struct ifla_vxlan_port_range ports = {
3075 .low = htons(vxlan->cfg.port_min),
3076 .high = htons(vxlan->cfg.port_max),
3079 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
3080 goto nla_put_failure;
3082 if (!vxlan_addr_any(&dst->remote_ip)) {
3083 if (dst->remote_ip.sa.sa_family == AF_INET) {
3084 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3085 dst->remote_ip.sin.sin_addr.s_addr))
3086 goto nla_put_failure;
3087 #if IS_ENABLED(CONFIG_IPV6)
3089 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3090 &dst->remote_ip.sin6.sin6_addr))
3091 goto nla_put_failure;
3096 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3097 goto nla_put_failure;
3099 if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3100 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3101 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3102 vxlan->cfg.saddr.sin.sin_addr.s_addr))
3103 goto nla_put_failure;
3104 #if IS_ENABLED(CONFIG_IPV6)
3106 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3107 &vxlan->cfg.saddr.sin6.sin6_addr))
3108 goto nla_put_failure;
3113 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3114 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3115 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3116 !!(vxlan->flags & VXLAN_F_LEARN)) ||
3117 nla_put_u8(skb, IFLA_VXLAN_PROXY,
3118 !!(vxlan->flags & VXLAN_F_PROXY)) ||
3119 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3120 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3121 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3122 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3123 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3124 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3125 !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3126 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3127 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3128 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3129 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3130 !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
3131 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3132 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3133 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3134 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3135 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3136 !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3137 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3138 !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3139 goto nla_put_failure;
3141 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3142 goto nla_put_failure;
3144 if (vxlan->flags & VXLAN_F_GBP &&
3145 nla_put_flag(skb, IFLA_VXLAN_GBP))
3146 goto nla_put_failure;
3148 if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3149 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3150 goto nla_put_failure;
3158 static struct net *vxlan_get_link_net(const struct net_device *dev)
3160 struct vxlan_dev *vxlan = netdev_priv(dev);
3165 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3167 .maxtype = IFLA_VXLAN_MAX,
3168 .policy = vxlan_policy,
3169 .priv_size = sizeof(struct vxlan_dev),
3170 .setup = vxlan_setup,
3171 .validate = vxlan_validate,
3172 .newlink = vxlan_newlink,
3173 .dellink = vxlan_dellink,
3174 .get_size = vxlan_get_size,
3175 .fill_info = vxlan_fill_info,
3176 .get_link_net = vxlan_get_link_net,
3179 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3180 struct net_device *dev)
3182 struct vxlan_dev *vxlan, *next;
3183 LIST_HEAD(list_kill);
3185 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3186 struct vxlan_rdst *dst = &vxlan->default_dst;
3188 /* In case we created vxlan device with carrier
3189 * and we loose the carrier due to module unload
3190 * we also need to remove vxlan device. In other
3191 * cases, it's not necessary and remote_ifindex
3192 * is 0 here, so no matches.
3194 if (dst->remote_ifindex == dev->ifindex)
3195 vxlan_dellink(vxlan->dev, &list_kill);
3198 unregister_netdevice_many(&list_kill);
3201 static int vxlan_lowerdev_event(struct notifier_block *unused,
3202 unsigned long event, void *ptr)
3204 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3205 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3207 if (event == NETDEV_UNREGISTER)
3208 vxlan_handle_lowerdev_unregister(vn, dev);
3213 static struct notifier_block vxlan_notifier_block __read_mostly = {
3214 .notifier_call = vxlan_lowerdev_event,
3217 static __net_init int vxlan_init_net(struct net *net)
3219 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3222 INIT_LIST_HEAD(&vn->vxlan_list);
3223 spin_lock_init(&vn->sock_lock);
3225 for (h = 0; h < PORT_HASH_SIZE; ++h)
3226 INIT_HLIST_HEAD(&vn->sock_list[h]);
3231 static void __net_exit vxlan_exit_net(struct net *net)
3233 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3234 struct vxlan_dev *vxlan, *next;
3235 struct net_device *dev, *aux;
3239 for_each_netdev_safe(net, dev, aux)
3240 if (dev->rtnl_link_ops == &vxlan_link_ops)
3241 unregister_netdevice_queue(dev, &list);
3243 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3244 /* If vxlan->dev is in the same netns, it has already been added
3245 * to the list by the previous loop.
3247 if (!net_eq(dev_net(vxlan->dev), net)) {
3248 gro_cells_destroy(&vxlan->gro_cells);
3249 unregister_netdevice_queue(vxlan->dev, &list);
3253 unregister_netdevice_many(&list);
3257 static struct pernet_operations vxlan_net_ops = {
3258 .init = vxlan_init_net,
3259 .exit = vxlan_exit_net,
3260 .id = &vxlan_net_id,
3261 .size = sizeof(struct vxlan_net),
3264 static int __init vxlan_init_module(void)
3268 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3272 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3274 rc = register_pernet_subsys(&vxlan_net_ops);
3278 rc = register_netdevice_notifier(&vxlan_notifier_block);
3282 rc = rtnl_link_register(&vxlan_link_ops);
3288 unregister_netdevice_notifier(&vxlan_notifier_block);
3290 unregister_pernet_subsys(&vxlan_net_ops);
3292 destroy_workqueue(vxlan_wq);
3295 late_initcall(vxlan_init_module);
3297 static void __exit vxlan_cleanup_module(void)
3299 rtnl_link_unregister(&vxlan_link_ops);
3300 unregister_netdevice_notifier(&vxlan_notifier_block);
3301 destroy_workqueue(vxlan_wq);
3302 unregister_pernet_subsys(&vxlan_net_ops);
3303 /* rcu_barrier() is called by netns */
3305 module_exit(vxlan_cleanup_module);
3307 MODULE_LICENSE("GPL");
3308 MODULE_VERSION(VXLAN_VERSION);
3309 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3310 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3311 MODULE_ALIAS_RTNL_LINK("vxlan");