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/module.h>
15 #include <linux/errno.h>
16 #include <linux/slab.h>
17 #include <linux/udp.h>
18 #include <linux/igmp.h>
19 #include <linux/if_ether.h>
20 #include <linux/ethtool.h>
22 #include <net/ndisc.h>
25 #include <net/rtnetlink.h>
26 #include <net/inet_ecn.h>
27 #include <net/net_namespace.h>
28 #include <net/netns/generic.h>
29 #include <net/vxlan.h>
31 #if IS_ENABLED(CONFIG_IPV6)
32 #include <net/ip6_tunnel.h>
33 #include <net/ip6_checksum.h>
36 #define VXLAN_VERSION "0.1"
38 #define PORT_HASH_BITS 8
39 #define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
40 #define FDB_AGE_DEFAULT 300 /* 5 min */
41 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
43 /* UDP port for VXLAN traffic.
44 * The IANA assigned port is 4789, but the Linux default is 8472
45 * for compatibility with early adopters.
47 static unsigned short vxlan_port __read_mostly = 8472;
48 module_param_named(udp_port, vxlan_port, ushort, 0444);
49 MODULE_PARM_DESC(udp_port, "Destination UDP port");
51 static bool log_ecn_error = true;
52 module_param(log_ecn_error, bool, 0644);
53 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
55 static int vxlan_net_id;
56 static struct rtnl_link_ops vxlan_link_ops;
58 static const u8 all_zeros_mac[ETH_ALEN + 2];
60 static int vxlan_sock_add(struct vxlan_dev *vxlan);
62 /* per-network namespace private data for this module */
64 struct list_head vxlan_list;
65 struct hlist_head sock_list[PORT_HASH_SIZE];
69 /* Forwarding table entry */
71 struct hlist_node hlist; /* linked list of entries */
73 unsigned long updated; /* jiffies */
75 struct list_head remotes;
76 u8 eth_addr[ETH_ALEN];
77 u16 state; /* see ndm_state */
78 u8 flags; /* see ndm_flags */
81 /* salt for hash table */
82 static u32 vxlan_salt __read_mostly;
84 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
86 return vs->flags & VXLAN_F_COLLECT_METADATA ||
87 ip_tunnel_collect_metadata();
90 #if IS_ENABLED(CONFIG_IPV6)
92 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
94 if (a->sa.sa_family != b->sa.sa_family)
96 if (a->sa.sa_family == AF_INET6)
97 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
99 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
102 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
104 if (ipa->sa.sa_family == AF_INET6)
105 return ipv6_addr_any(&ipa->sin6.sin6_addr);
107 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
110 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
112 if (ipa->sa.sa_family == AF_INET6)
113 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
115 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
118 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
120 if (nla_len(nla) >= sizeof(struct in6_addr)) {
121 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
122 ip->sa.sa_family = AF_INET6;
124 } else if (nla_len(nla) >= sizeof(__be32)) {
125 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
126 ip->sa.sa_family = AF_INET;
129 return -EAFNOSUPPORT;
133 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
134 const union vxlan_addr *ip)
136 if (ip->sa.sa_family == AF_INET6)
137 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
139 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
142 #else /* !CONFIG_IPV6 */
145 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
147 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
150 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
152 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
155 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
157 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
160 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
162 if (nla_len(nla) >= sizeof(struct in6_addr)) {
163 return -EAFNOSUPPORT;
164 } else if (nla_len(nla) >= sizeof(__be32)) {
165 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
166 ip->sa.sa_family = AF_INET;
169 return -EAFNOSUPPORT;
173 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
174 const union vxlan_addr *ip)
176 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
180 /* Virtual Network hash table head */
181 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
183 return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
186 /* Socket hash table head */
187 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
189 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
191 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
194 /* First remote destination for a forwarding entry.
195 * Guaranteed to be non-NULL because remotes are never deleted.
197 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
199 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
202 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
204 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
207 /* Find VXLAN socket based on network namespace, address family and UDP port
208 * and enabled unshareable flags.
210 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
211 __be16 port, u32 flags)
213 struct vxlan_sock *vs;
215 flags &= VXLAN_F_RCV_FLAGS;
217 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
218 if (inet_sk(vs->sock->sk)->inet_sport == port &&
219 vxlan_get_sk_family(vs) == family &&
226 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
228 struct vxlan_dev *vxlan;
230 /* For flow based devices, map all packets to VNI 0 */
231 if (vs->flags & VXLAN_F_COLLECT_METADATA)
234 hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
235 if (vxlan->default_dst.remote_vni == vni)
242 /* Look up VNI in a per net namespace table */
243 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
244 sa_family_t family, __be16 port,
247 struct vxlan_sock *vs;
249 vs = vxlan_find_sock(net, family, port, flags);
253 return vxlan_vs_find_vni(vs, vni);
256 /* Fill in neighbour message in skbuff. */
257 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
258 const struct vxlan_fdb *fdb,
259 u32 portid, u32 seq, int type, unsigned int flags,
260 const struct vxlan_rdst *rdst)
262 unsigned long now = jiffies;
263 struct nda_cacheinfo ci;
264 struct nlmsghdr *nlh;
266 bool send_ip, send_eth;
268 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
272 ndm = nlmsg_data(nlh);
273 memset(ndm, 0, sizeof(*ndm));
275 send_eth = send_ip = true;
277 if (type == RTM_GETNEIGH) {
278 ndm->ndm_family = AF_INET;
279 send_ip = !vxlan_addr_any(&rdst->remote_ip);
280 send_eth = !is_zero_ether_addr(fdb->eth_addr);
282 ndm->ndm_family = AF_BRIDGE;
283 ndm->ndm_state = fdb->state;
284 ndm->ndm_ifindex = vxlan->dev->ifindex;
285 ndm->ndm_flags = fdb->flags;
286 ndm->ndm_type = RTN_UNICAST;
288 if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
289 nla_put_s32(skb, NDA_LINK_NETNSID,
290 peernet2id(dev_net(vxlan->dev), vxlan->net)))
291 goto nla_put_failure;
293 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
294 goto nla_put_failure;
296 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
297 goto nla_put_failure;
299 if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
300 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
301 goto nla_put_failure;
302 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
303 nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
304 goto nla_put_failure;
305 if (rdst->remote_ifindex &&
306 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
307 goto nla_put_failure;
309 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
310 ci.ndm_confirmed = 0;
311 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
314 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
315 goto nla_put_failure;
321 nlmsg_cancel(skb, nlh);
325 static inline size_t vxlan_nlmsg_size(void)
327 return NLMSG_ALIGN(sizeof(struct ndmsg))
328 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
329 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
330 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
331 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
332 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
333 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
334 + nla_total_size(sizeof(struct nda_cacheinfo));
337 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
338 struct vxlan_rdst *rd, int type)
340 struct net *net = dev_net(vxlan->dev);
344 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
348 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
350 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
351 WARN_ON(err == -EMSGSIZE);
356 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
360 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
363 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
365 struct vxlan_dev *vxlan = netdev_priv(dev);
366 struct vxlan_fdb f = {
369 struct vxlan_rdst remote = {
370 .remote_ip = *ipa, /* goes to NDA_DST */
371 .remote_vni = cpu_to_be32(VXLAN_N_VID),
374 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
377 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
379 struct vxlan_fdb f = {
382 struct vxlan_rdst remote = { };
384 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
386 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
389 /* Hash Ethernet address */
390 static u32 eth_hash(const unsigned char *addr)
392 u64 value = get_unaligned((u64 *)addr);
394 /* only want 6 bytes */
400 return hash_64(value, FDB_HASH_BITS);
403 /* Hash chain to use given mac address */
404 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
407 return &vxlan->fdb_head[eth_hash(mac)];
410 /* Look up Ethernet address in forwarding table */
411 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
414 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
417 hlist_for_each_entry_rcu(f, head, hlist) {
418 if (ether_addr_equal(mac, f->eth_addr))
425 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
430 f = __vxlan_find_mac(vxlan, mac);
437 /* caller should hold vxlan->hash_lock */
438 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
439 union vxlan_addr *ip, __be16 port,
440 __be32 vni, __u32 ifindex)
442 struct vxlan_rdst *rd;
444 list_for_each_entry(rd, &f->remotes, list) {
445 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
446 rd->remote_port == port &&
447 rd->remote_vni == vni &&
448 rd->remote_ifindex == ifindex)
455 /* Replace destination of unicast mac */
456 static int vxlan_fdb_replace(struct vxlan_fdb *f,
457 union vxlan_addr *ip, __be16 port, __be32 vni,
460 struct vxlan_rdst *rd;
462 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
466 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
470 dst_cache_reset(&rd->dst_cache);
472 rd->remote_port = port;
473 rd->remote_vni = vni;
474 rd->remote_ifindex = ifindex;
478 /* Add/update destinations for multicast */
479 static int vxlan_fdb_append(struct vxlan_fdb *f,
480 union vxlan_addr *ip, __be16 port, __be32 vni,
481 __u32 ifindex, struct vxlan_rdst **rdp)
483 struct vxlan_rdst *rd;
485 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
489 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
493 if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
499 rd->remote_port = port;
500 rd->remote_vni = vni;
501 rd->remote_ifindex = ifindex;
503 list_add_tail_rcu(&rd->list, &f->remotes);
509 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
511 struct vxlanhdr *vh, size_t hdrlen,
513 struct gro_remcsum *grc,
516 size_t start, offset;
518 if (skb->remcsum_offload)
521 if (!NAPI_GRO_CB(skb)->csum_valid)
524 start = vxlan_rco_start(vni_field);
525 offset = start + vxlan_rco_offset(vni_field);
527 vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
528 start, offset, grc, nopartial);
530 skb->remcsum_offload = 1;
535 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
536 struct sk_buff **head,
539 struct sk_buff *p, **pp = NULL;
540 struct vxlanhdr *vh, *vh2;
541 unsigned int hlen, off_vx;
543 struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
545 struct gro_remcsum grc;
547 skb_gro_remcsum_init(&grc);
549 off_vx = skb_gro_offset(skb);
550 hlen = off_vx + sizeof(*vh);
551 vh = skb_gro_header_fast(skb, off_vx);
552 if (skb_gro_header_hard(skb, hlen)) {
553 vh = skb_gro_header_slow(skb, hlen, off_vx);
558 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
560 flags = vh->vx_flags;
562 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
563 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
566 VXLAN_F_REMCSUM_NOPARTIAL));
572 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
574 for (p = *head; p; p = p->next) {
575 if (!NAPI_GRO_CB(p)->same_flow)
578 vh2 = (struct vxlanhdr *)(p->data + off_vx);
579 if (vh->vx_flags != vh2->vx_flags ||
580 vh->vx_vni != vh2->vx_vni) {
581 NAPI_GRO_CB(p)->same_flow = 0;
586 pp = call_gro_receive(eth_gro_receive, head, skb);
590 skb_gro_remcsum_cleanup(skb, &grc);
591 NAPI_GRO_CB(skb)->flush |= flush;
596 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
598 /* Sets 'skb->inner_mac_header' since we are always called with
599 * 'skb->encapsulation' set.
601 return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
604 /* Add new entry to forwarding table -- assumes lock held */
605 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
606 const u8 *mac, union vxlan_addr *ip,
607 __u16 state, __u16 flags,
608 __be16 port, __be32 vni, __u32 ifindex,
611 struct vxlan_rdst *rd = NULL;
616 f = __vxlan_find_mac(vxlan, mac);
618 if (flags & NLM_F_EXCL) {
619 netdev_dbg(vxlan->dev,
620 "lost race to create %pM\n", mac);
623 if (f->state != state) {
625 f->updated = jiffies;
628 if (f->flags != ndm_flags) {
629 f->flags = ndm_flags;
630 f->updated = jiffies;
633 if ((flags & NLM_F_REPLACE)) {
634 /* Only change unicasts */
635 if (!(is_multicast_ether_addr(f->eth_addr) ||
636 is_zero_ether_addr(f->eth_addr))) {
637 notify |= vxlan_fdb_replace(f, ip, port, vni,
642 if ((flags & NLM_F_APPEND) &&
643 (is_multicast_ether_addr(f->eth_addr) ||
644 is_zero_ether_addr(f->eth_addr))) {
645 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
652 if (!(flags & NLM_F_CREATE))
655 if (vxlan->cfg.addrmax &&
656 vxlan->addrcnt >= vxlan->cfg.addrmax)
659 /* Disallow replace to add a multicast entry */
660 if ((flags & NLM_F_REPLACE) &&
661 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
664 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
665 f = kmalloc(sizeof(*f), GFP_ATOMIC);
671 f->flags = ndm_flags;
672 f->updated = f->used = jiffies;
673 INIT_LIST_HEAD(&f->remotes);
674 memcpy(f->eth_addr, mac, ETH_ALEN);
676 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
683 hlist_add_head_rcu(&f->hlist,
684 vxlan_fdb_head(vxlan, mac));
689 rd = first_remote_rtnl(f);
690 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
696 static void vxlan_fdb_free(struct rcu_head *head)
698 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
699 struct vxlan_rdst *rd, *nd;
701 list_for_each_entry_safe(rd, nd, &f->remotes, list) {
702 dst_cache_destroy(&rd->dst_cache);
708 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
710 netdev_dbg(vxlan->dev,
711 "delete %pM\n", f->eth_addr);
714 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
716 hlist_del_rcu(&f->hlist);
717 call_rcu(&f->rcu, vxlan_fdb_free);
720 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
721 union vxlan_addr *ip, __be16 *port, __be32 *vni,
724 struct net *net = dev_net(vxlan->dev);
728 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
732 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
733 if (remote->sa.sa_family == AF_INET) {
734 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
735 ip->sa.sa_family = AF_INET;
736 #if IS_ENABLED(CONFIG_IPV6)
738 ip->sin6.sin6_addr = in6addr_any;
739 ip->sa.sa_family = AF_INET6;
745 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
747 *port = nla_get_be16(tb[NDA_PORT]);
749 *port = vxlan->cfg.dst_port;
753 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
755 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
757 *vni = vxlan->default_dst.remote_vni;
760 if (tb[NDA_IFINDEX]) {
761 struct net_device *tdev;
763 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
765 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
766 tdev = __dev_get_by_index(net, *ifindex);
768 return -EADDRNOTAVAIL;
776 /* Add static entry (via netlink) */
777 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
778 struct net_device *dev,
779 const unsigned char *addr, u16 vid, u16 flags)
781 struct vxlan_dev *vxlan = netdev_priv(dev);
782 /* struct net *net = dev_net(vxlan->dev); */
789 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
790 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
795 if (tb[NDA_DST] == NULL)
798 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
802 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
803 return -EAFNOSUPPORT;
805 spin_lock_bh(&vxlan->hash_lock);
806 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
807 port, vni, ifindex, ndm->ndm_flags);
808 spin_unlock_bh(&vxlan->hash_lock);
813 /* Delete entry (via netlink) */
814 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
815 struct net_device *dev,
816 const unsigned char *addr, u16 vid)
818 struct vxlan_dev *vxlan = netdev_priv(dev);
820 struct vxlan_rdst *rd = NULL;
827 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
833 spin_lock_bh(&vxlan->hash_lock);
834 f = vxlan_find_mac(vxlan, addr);
838 if (!vxlan_addr_any(&ip)) {
839 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
846 /* remove a destination if it's not the only one on the list,
847 * otherwise destroy the fdb entry
849 if (rd && !list_is_singular(&f->remotes)) {
850 list_del_rcu(&rd->list);
851 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
856 vxlan_fdb_destroy(vxlan, f);
859 spin_unlock_bh(&vxlan->hash_lock);
864 /* Dump forwarding table */
865 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
866 struct net_device *dev,
867 struct net_device *filter_dev, int *idx)
869 struct vxlan_dev *vxlan = netdev_priv(dev);
873 for (h = 0; h < FDB_HASH_SIZE; ++h) {
876 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
877 struct vxlan_rdst *rd;
879 list_for_each_entry_rcu(rd, &f->remotes, list) {
880 if (*idx < cb->args[2])
883 err = vxlan_fdb_info(skb, vxlan, f,
884 NETLINK_CB(cb->skb).portid,
899 /* Watch incoming packets to learn mapping between Ethernet address
900 * and Tunnel endpoint.
901 * Return true if packet is bogus and should be dropped.
903 static bool vxlan_snoop(struct net_device *dev,
904 union vxlan_addr *src_ip, const u8 *src_mac)
906 struct vxlan_dev *vxlan = netdev_priv(dev);
909 f = vxlan_find_mac(vxlan, src_mac);
911 struct vxlan_rdst *rdst = first_remote_rcu(f);
913 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
916 /* Don't migrate static entries, drop packets */
917 if (f->state & NUD_NOARP)
922 "%pM migrated from %pIS to %pIS\n",
923 src_mac, &rdst->remote_ip.sa, &src_ip->sa);
925 rdst->remote_ip = *src_ip;
926 f->updated = jiffies;
927 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
929 /* learned new entry */
930 spin_lock(&vxlan->hash_lock);
932 /* close off race between vxlan_flush and incoming packets */
933 if (netif_running(dev))
934 vxlan_fdb_create(vxlan, src_mac, src_ip,
936 NLM_F_EXCL|NLM_F_CREATE,
938 vxlan->default_dst.remote_vni,
940 spin_unlock(&vxlan->hash_lock);
946 /* See if multicast group is already in use by other ID */
947 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
949 struct vxlan_dev *vxlan;
950 struct vxlan_sock *sock4;
951 #if IS_ENABLED(CONFIG_IPV6)
952 struct vxlan_sock *sock6;
954 unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
956 sock4 = rtnl_dereference(dev->vn4_sock);
958 /* The vxlan_sock is only used by dev, leaving group has
959 * no effect on other vxlan devices.
961 if (family == AF_INET && sock4 && atomic_read(&sock4->refcnt) == 1)
963 #if IS_ENABLED(CONFIG_IPV6)
964 sock6 = rtnl_dereference(dev->vn6_sock);
965 if (family == AF_INET6 && sock6 && atomic_read(&sock6->refcnt) == 1)
969 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
970 if (!netif_running(vxlan->dev) || vxlan == dev)
973 if (family == AF_INET &&
974 rtnl_dereference(vxlan->vn4_sock) != sock4)
976 #if IS_ENABLED(CONFIG_IPV6)
977 if (family == AF_INET6 &&
978 rtnl_dereference(vxlan->vn6_sock) != sock6)
982 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
983 &dev->default_dst.remote_ip))
986 if (vxlan->default_dst.remote_ifindex !=
987 dev->default_dst.remote_ifindex)
996 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
998 struct vxlan_net *vn;
1002 if (!atomic_dec_and_test(&vs->refcnt))
1005 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1006 spin_lock(&vn->sock_lock);
1007 hlist_del_rcu(&vs->hlist);
1008 udp_tunnel_notify_del_rx_port(vs->sock,
1009 (vs->flags & VXLAN_F_GPE) ?
1010 UDP_TUNNEL_TYPE_VXLAN_GPE :
1011 UDP_TUNNEL_TYPE_VXLAN);
1012 spin_unlock(&vn->sock_lock);
1017 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1019 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1020 #if IS_ENABLED(CONFIG_IPV6)
1021 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1023 rcu_assign_pointer(vxlan->vn6_sock, NULL);
1026 rcu_assign_pointer(vxlan->vn4_sock, NULL);
1029 if (__vxlan_sock_release_prep(sock4)) {
1030 udp_tunnel_sock_release(sock4->sock);
1034 #if IS_ENABLED(CONFIG_IPV6)
1035 if (__vxlan_sock_release_prep(sock6)) {
1036 udp_tunnel_sock_release(sock6->sock);
1042 /* Update multicast group membership when first VNI on
1043 * multicast address is brought up
1045 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1048 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1049 int ifindex = vxlan->default_dst.remote_ifindex;
1052 if (ip->sa.sa_family == AF_INET) {
1053 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1054 struct ip_mreqn mreq = {
1055 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1056 .imr_ifindex = ifindex,
1059 sk = sock4->sock->sk;
1061 ret = ip_mc_join_group(sk, &mreq);
1063 #if IS_ENABLED(CONFIG_IPV6)
1065 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1067 sk = sock6->sock->sk;
1069 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1070 &ip->sin6.sin6_addr);
1078 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1079 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1082 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1083 int ifindex = vxlan->default_dst.remote_ifindex;
1086 if (ip->sa.sa_family == AF_INET) {
1087 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1088 struct ip_mreqn mreq = {
1089 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1090 .imr_ifindex = ifindex,
1093 sk = sock4->sock->sk;
1095 ret = ip_mc_leave_group(sk, &mreq);
1097 #if IS_ENABLED(CONFIG_IPV6)
1099 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1101 sk = sock6->sock->sk;
1103 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1104 &ip->sin6.sin6_addr);
1112 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1113 struct sk_buff *skb, u32 vxflags)
1115 size_t start, offset;
1117 if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1120 start = vxlan_rco_start(unparsed->vx_vni);
1121 offset = start + vxlan_rco_offset(unparsed->vx_vni);
1123 if (!pskb_may_pull(skb, offset + sizeof(u16)))
1126 skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1127 !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1129 unparsed->vx_flags &= ~VXLAN_HF_RCO;
1130 unparsed->vx_vni &= VXLAN_VNI_MASK;
1134 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1135 struct sk_buff *skb, u32 vxflags,
1136 struct vxlan_metadata *md)
1138 struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1139 struct metadata_dst *tun_dst;
1141 if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1144 md->gbp = ntohs(gbp->policy_id);
1146 tun_dst = (struct metadata_dst *)skb_dst(skb);
1148 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1149 tun_dst->u.tun_info.options_len = sizeof(*md);
1151 if (gbp->dont_learn)
1152 md->gbp |= VXLAN_GBP_DONT_LEARN;
1154 if (gbp->policy_applied)
1155 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1157 /* In flow-based mode, GBP is carried in dst_metadata */
1158 if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1159 skb->mark = md->gbp;
1161 unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1164 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1166 struct sk_buff *skb, u32 vxflags)
1168 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1170 /* Need to have Next Protocol set for interfaces in GPE mode. */
1171 if (!gpe->np_applied)
1173 /* "The initial version is 0. If a receiver does not support the
1174 * version indicated it MUST drop the packet.
1176 if (gpe->version != 0)
1178 /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1179 * processing MUST occur." However, we don't implement OAM
1180 * processing, thus drop the packet.
1185 switch (gpe->next_protocol) {
1186 case VXLAN_GPE_NP_IPV4:
1187 *protocol = htons(ETH_P_IP);
1189 case VXLAN_GPE_NP_IPV6:
1190 *protocol = htons(ETH_P_IPV6);
1192 case VXLAN_GPE_NP_ETHERNET:
1193 *protocol = htons(ETH_P_TEB);
1199 unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1203 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1204 struct vxlan_sock *vs,
1205 struct sk_buff *skb)
1207 union vxlan_addr saddr;
1209 skb_reset_mac_header(skb);
1210 skb->protocol = eth_type_trans(skb, vxlan->dev);
1211 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1213 /* Ignore packet loops (and multicast echo) */
1214 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1217 /* Get address from the outer IP header */
1218 if (vxlan_get_sk_family(vs) == AF_INET) {
1219 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1220 saddr.sa.sa_family = AF_INET;
1221 #if IS_ENABLED(CONFIG_IPV6)
1223 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1224 saddr.sa.sa_family = AF_INET6;
1228 if ((vxlan->flags & VXLAN_F_LEARN) &&
1229 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1235 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1236 struct sk_buff *skb)
1240 if (vxlan_get_sk_family(vs) == AF_INET)
1241 err = IP_ECN_decapsulate(oiph, skb);
1242 #if IS_ENABLED(CONFIG_IPV6)
1244 err = IP6_ECN_decapsulate(oiph, skb);
1247 if (unlikely(err) && log_ecn_error) {
1248 if (vxlan_get_sk_family(vs) == AF_INET)
1249 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1250 &((struct iphdr *)oiph)->saddr,
1251 ((struct iphdr *)oiph)->tos);
1253 net_info_ratelimited("non-ECT from %pI6\n",
1254 &((struct ipv6hdr *)oiph)->saddr);
1259 /* Callback from net/ipv4/udp.c to receive packets */
1260 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1262 struct pcpu_sw_netstats *stats;
1263 struct vxlan_dev *vxlan;
1264 struct vxlan_sock *vs;
1265 struct vxlanhdr unparsed;
1266 struct vxlan_metadata _md;
1267 struct vxlan_metadata *md = &_md;
1268 __be16 protocol = htons(ETH_P_TEB);
1269 bool raw_proto = false;
1272 /* Need UDP and VXLAN header to be present */
1273 if (!pskb_may_pull(skb, VXLAN_HLEN))
1276 unparsed = *vxlan_hdr(skb);
1277 /* VNI flag always required to be set */
1278 if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1279 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1280 ntohl(vxlan_hdr(skb)->vx_flags),
1281 ntohl(vxlan_hdr(skb)->vx_vni));
1282 /* Return non vxlan pkt */
1285 unparsed.vx_flags &= ~VXLAN_HF_VNI;
1286 unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1288 vs = rcu_dereference_sk_user_data(sk);
1292 vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1296 /* For backwards compatibility, only allow reserved fields to be
1297 * used by VXLAN extensions if explicitly requested.
1299 if (vs->flags & VXLAN_F_GPE) {
1300 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1305 if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1306 !net_eq(vxlan->net, dev_net(vxlan->dev))))
1309 if (vxlan_collect_metadata(vs)) {
1310 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1311 struct metadata_dst *tun_dst;
1313 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1314 key32_to_tunnel_id(vni), sizeof(*md));
1319 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1321 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1323 memset(md, 0, sizeof(*md));
1326 if (vs->flags & VXLAN_F_REMCSUM_RX)
1327 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1329 if (vs->flags & VXLAN_F_GBP)
1330 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1331 /* Note that GBP and GPE can never be active together. This is
1332 * ensured in vxlan_dev_configure.
1335 if (unparsed.vx_flags || unparsed.vx_vni) {
1336 /* If there are any unprocessed flags remaining treat
1337 * this as a malformed packet. This behavior diverges from
1338 * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1339 * in reserved fields are to be ignored. The approach here
1340 * maintains compatibility with previous stack code, and also
1341 * is more robust and provides a little more security in
1342 * adding extensions to VXLAN.
1348 if (!vxlan_set_mac(vxlan, vs, skb))
1351 skb_reset_mac_header(skb);
1352 skb->dev = vxlan->dev;
1353 skb->pkt_type = PACKET_HOST;
1356 oiph = skb_network_header(skb);
1357 skb_reset_network_header(skb);
1359 if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1360 ++vxlan->dev->stats.rx_frame_errors;
1361 ++vxlan->dev->stats.rx_errors;
1365 stats = this_cpu_ptr(vxlan->dev->tstats);
1366 u64_stats_update_begin(&stats->syncp);
1367 stats->rx_packets++;
1368 stats->rx_bytes += skb->len;
1369 u64_stats_update_end(&stats->syncp);
1371 gro_cells_receive(&vxlan->gro_cells, skb);
1375 /* Consume bad packet */
1380 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1382 struct vxlan_dev *vxlan = netdev_priv(dev);
1383 struct arphdr *parp;
1386 struct neighbour *n;
1388 if (dev->flags & IFF_NOARP)
1391 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1392 dev->stats.tx_dropped++;
1395 parp = arp_hdr(skb);
1397 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1398 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1399 parp->ar_pro != htons(ETH_P_IP) ||
1400 parp->ar_op != htons(ARPOP_REQUEST) ||
1401 parp->ar_hln != dev->addr_len ||
1404 arpptr = (u8 *)parp + sizeof(struct arphdr);
1406 arpptr += dev->addr_len; /* sha */
1407 memcpy(&sip, arpptr, sizeof(sip));
1408 arpptr += sizeof(sip);
1409 arpptr += dev->addr_len; /* tha */
1410 memcpy(&tip, arpptr, sizeof(tip));
1412 if (ipv4_is_loopback(tip) ||
1413 ipv4_is_multicast(tip))
1416 n = neigh_lookup(&arp_tbl, &tip, dev);
1419 struct vxlan_fdb *f;
1420 struct sk_buff *reply;
1422 if (!(n->nud_state & NUD_CONNECTED)) {
1427 f = vxlan_find_mac(vxlan, n->ha);
1428 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1429 /* bridge-local neighbor */
1434 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1442 skb_reset_mac_header(reply);
1443 __skb_pull(reply, skb_network_offset(reply));
1444 reply->ip_summed = CHECKSUM_UNNECESSARY;
1445 reply->pkt_type = PACKET_HOST;
1447 if (netif_rx_ni(reply) == NET_RX_DROP)
1448 dev->stats.rx_dropped++;
1449 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1450 union vxlan_addr ipa = {
1451 .sin.sin_addr.s_addr = tip,
1452 .sin.sin_family = AF_INET,
1455 vxlan_ip_miss(dev, &ipa);
1459 return NETDEV_TX_OK;
1462 #if IS_ENABLED(CONFIG_IPV6)
1463 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1464 struct neighbour *n, bool isrouter)
1466 struct net_device *dev = request->dev;
1467 struct sk_buff *reply;
1468 struct nd_msg *ns, *na;
1469 struct ipv6hdr *pip6;
1471 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1478 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1479 sizeof(*na) + na_olen + dev->needed_tailroom;
1480 reply = alloc_skb(len, GFP_ATOMIC);
1484 reply->protocol = htons(ETH_P_IPV6);
1486 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1487 skb_push(reply, sizeof(struct ethhdr));
1488 skb_reset_mac_header(reply);
1490 ns = (struct nd_msg *)skb_transport_header(request);
1492 daddr = eth_hdr(request)->h_source;
1493 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1494 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1495 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1496 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1501 /* Ethernet header */
1502 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1503 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1504 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1505 reply->protocol = htons(ETH_P_IPV6);
1507 skb_pull(reply, sizeof(struct ethhdr));
1508 skb_reset_network_header(reply);
1509 skb_put(reply, sizeof(struct ipv6hdr));
1513 pip6 = ipv6_hdr(reply);
1514 memset(pip6, 0, sizeof(struct ipv6hdr));
1516 pip6->priority = ipv6_hdr(request)->priority;
1517 pip6->nexthdr = IPPROTO_ICMPV6;
1518 pip6->hop_limit = 255;
1519 pip6->daddr = ipv6_hdr(request)->saddr;
1520 pip6->saddr = *(struct in6_addr *)n->primary_key;
1522 skb_pull(reply, sizeof(struct ipv6hdr));
1523 skb_reset_transport_header(reply);
1525 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1527 /* Neighbor Advertisement */
1528 memset(na, 0, sizeof(*na)+na_olen);
1529 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1530 na->icmph.icmp6_router = isrouter;
1531 na->icmph.icmp6_override = 1;
1532 na->icmph.icmp6_solicited = 1;
1533 na->target = ns->target;
1534 ether_addr_copy(&na->opt[2], n->ha);
1535 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1536 na->opt[1] = na_olen >> 3;
1538 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1539 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1540 csum_partial(na, sizeof(*na)+na_olen, 0));
1542 pip6->payload_len = htons(sizeof(*na)+na_olen);
1544 skb_push(reply, sizeof(struct ipv6hdr));
1546 reply->ip_summed = CHECKSUM_UNNECESSARY;
1551 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1553 struct vxlan_dev *vxlan = netdev_priv(dev);
1555 const struct ipv6hdr *iphdr;
1556 const struct in6_addr *saddr, *daddr;
1557 struct neighbour *n;
1558 struct inet6_dev *in6_dev;
1560 in6_dev = __in6_dev_get(dev);
1564 iphdr = ipv6_hdr(skb);
1565 saddr = &iphdr->saddr;
1566 daddr = &iphdr->daddr;
1568 msg = (struct nd_msg *)skb_transport_header(skb);
1569 if (msg->icmph.icmp6_code != 0 ||
1570 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1573 if (ipv6_addr_loopback(daddr) ||
1574 ipv6_addr_is_multicast(&msg->target))
1577 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1580 struct vxlan_fdb *f;
1581 struct sk_buff *reply;
1583 if (!(n->nud_state & NUD_CONNECTED)) {
1588 f = vxlan_find_mac(vxlan, n->ha);
1589 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1590 /* bridge-local neighbor */
1595 reply = vxlan_na_create(skb, n,
1596 !!(f ? f->flags & NTF_ROUTER : 0));
1603 if (netif_rx_ni(reply) == NET_RX_DROP)
1604 dev->stats.rx_dropped++;
1606 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1607 union vxlan_addr ipa = {
1608 .sin6.sin6_addr = msg->target,
1609 .sin6.sin6_family = AF_INET6,
1612 vxlan_ip_miss(dev, &ipa);
1617 return NETDEV_TX_OK;
1621 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1623 struct vxlan_dev *vxlan = netdev_priv(dev);
1624 struct neighbour *n;
1626 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1630 switch (ntohs(eth_hdr(skb)->h_proto)) {
1635 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1638 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1639 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1640 union vxlan_addr ipa = {
1641 .sin.sin_addr.s_addr = pip->daddr,
1642 .sin.sin_family = AF_INET,
1645 vxlan_ip_miss(dev, &ipa);
1651 #if IS_ENABLED(CONFIG_IPV6)
1654 struct ipv6hdr *pip6;
1656 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1658 pip6 = ipv6_hdr(skb);
1659 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1660 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1661 union vxlan_addr ipa = {
1662 .sin6.sin6_addr = pip6->daddr,
1663 .sin6.sin6_family = AF_INET6,
1666 vxlan_ip_miss(dev, &ipa);
1680 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1682 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1684 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1693 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1694 struct vxlan_metadata *md)
1696 struct vxlanhdr_gbp *gbp;
1701 gbp = (struct vxlanhdr_gbp *)vxh;
1702 vxh->vx_flags |= VXLAN_HF_GBP;
1704 if (md->gbp & VXLAN_GBP_DONT_LEARN)
1705 gbp->dont_learn = 1;
1707 if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1708 gbp->policy_applied = 1;
1710 gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1713 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1716 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1718 gpe->np_applied = 1;
1721 case htons(ETH_P_IP):
1722 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1724 case htons(ETH_P_IPV6):
1725 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1727 case htons(ETH_P_TEB):
1728 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1731 return -EPFNOSUPPORT;
1734 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1735 int iphdr_len, __be32 vni,
1736 struct vxlan_metadata *md, u32 vxflags,
1739 struct vxlanhdr *vxh;
1742 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1743 __be16 inner_protocol = htons(ETH_P_TEB);
1745 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1746 skb->ip_summed == CHECKSUM_PARTIAL) {
1747 int csum_start = skb_checksum_start_offset(skb);
1749 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1750 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1751 (skb->csum_offset == offsetof(struct udphdr, check) ||
1752 skb->csum_offset == offsetof(struct tcphdr, check)))
1753 type |= SKB_GSO_TUNNEL_REMCSUM;
1756 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1757 + VXLAN_HLEN + iphdr_len
1758 + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1760 /* Need space for new headers (invalidates iph ptr) */
1761 err = skb_cow_head(skb, min_headroom);
1765 skb = vlan_hwaccel_push_inside(skb);
1769 err = iptunnel_handle_offloads(skb, type);
1773 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1774 vxh->vx_flags = VXLAN_HF_VNI;
1775 vxh->vx_vni = vxlan_vni_field(vni);
1777 if (type & SKB_GSO_TUNNEL_REMCSUM) {
1780 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1781 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1782 vxh->vx_flags |= VXLAN_HF_RCO;
1784 if (!skb_is_gso(skb)) {
1785 skb->ip_summed = CHECKSUM_NONE;
1786 skb->encapsulation = 0;
1790 if (vxflags & VXLAN_F_GBP)
1791 vxlan_build_gbp_hdr(vxh, vxflags, md);
1792 if (vxflags & VXLAN_F_GPE) {
1793 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1796 inner_protocol = skb->protocol;
1799 skb_set_inner_protocol(skb, inner_protocol);
1807 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1808 struct sk_buff *skb, int oif, u8 tos,
1809 __be32 daddr, __be32 *saddr,
1810 struct dst_cache *dst_cache,
1811 const struct ip_tunnel_info *info)
1813 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1814 struct rtable *rt = NULL;
1820 rt = dst_cache_get_ip4(dst_cache, saddr);
1825 memset(&fl4, 0, sizeof(fl4));
1826 fl4.flowi4_oif = oif;
1827 fl4.flowi4_tos = RT_TOS(tos);
1828 fl4.flowi4_mark = skb->mark;
1829 fl4.flowi4_proto = IPPROTO_UDP;
1833 rt = ip_route_output_key(vxlan->net, &fl4);
1837 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1842 #if IS_ENABLED(CONFIG_IPV6)
1843 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1844 struct sk_buff *skb, int oif, u8 tos,
1846 const struct in6_addr *daddr,
1847 struct in6_addr *saddr,
1848 struct dst_cache *dst_cache,
1849 const struct ip_tunnel_info *info)
1851 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
1852 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1853 struct dst_entry *ndst;
1858 return ERR_PTR(-EIO);
1863 ndst = dst_cache_get_ip6(dst_cache, saddr);
1868 memset(&fl6, 0, sizeof(fl6));
1869 fl6.flowi6_oif = oif;
1872 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1873 fl6.flowi6_mark = skb->mark;
1874 fl6.flowi6_proto = IPPROTO_UDP;
1876 err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1880 return ERR_PTR(err);
1884 dst_cache_set_ip6(dst_cache, ndst, saddr);
1889 /* Bypass encapsulation if the destination is local */
1890 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1891 struct vxlan_dev *dst_vxlan)
1893 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1894 union vxlan_addr loopback;
1895 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1896 struct net_device *dev = skb->dev;
1899 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1900 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1901 skb->pkt_type = PACKET_HOST;
1902 skb->encapsulation = 0;
1903 skb->dev = dst_vxlan->dev;
1904 __skb_pull(skb, skb_network_offset(skb));
1906 if (remote_ip->sa.sa_family == AF_INET) {
1907 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1908 loopback.sa.sa_family = AF_INET;
1909 #if IS_ENABLED(CONFIG_IPV6)
1911 loopback.sin6.sin6_addr = in6addr_loopback;
1912 loopback.sa.sa_family = AF_INET6;
1916 if (dst_vxlan->flags & VXLAN_F_LEARN)
1917 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1919 u64_stats_update_begin(&tx_stats->syncp);
1920 tx_stats->tx_packets++;
1921 tx_stats->tx_bytes += len;
1922 u64_stats_update_end(&tx_stats->syncp);
1924 if (netif_rx(skb) == NET_RX_SUCCESS) {
1925 u64_stats_update_begin(&rx_stats->syncp);
1926 rx_stats->rx_packets++;
1927 rx_stats->rx_bytes += len;
1928 u64_stats_update_end(&rx_stats->syncp);
1930 dev->stats.rx_dropped++;
1934 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1935 struct vxlan_rdst *rdst, bool did_rsc)
1937 struct dst_cache *dst_cache;
1938 struct ip_tunnel_info *info;
1939 struct vxlan_dev *vxlan = netdev_priv(dev);
1941 struct rtable *rt = NULL;
1942 const struct iphdr *old_iph;
1943 union vxlan_addr *dst;
1944 union vxlan_addr remote_ip, local_ip;
1945 union vxlan_addr *src;
1946 struct vxlan_metadata _md;
1947 struct vxlan_metadata *md = &_md;
1948 __be16 src_port = 0, dst_port;
1953 u32 flags = vxlan->flags;
1954 bool udp_sum = false;
1955 bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1957 info = skb_tunnel_info(skb);
1960 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1961 vni = rdst->remote_vni;
1962 dst = &rdst->remote_ip;
1963 src = &vxlan->cfg.saddr;
1964 dst_cache = &rdst->dst_cache;
1967 WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1971 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1972 vni = tunnel_id_to_key32(info->key.tun_id);
1973 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1974 if (remote_ip.sa.sa_family == AF_INET) {
1975 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1976 local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
1978 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1979 local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
1983 dst_cache = &info->dst_cache;
1986 if (vxlan_addr_any(dst)) {
1988 /* short-circuited back to local bridge */
1989 vxlan_encap_bypass(skb, vxlan, vxlan);
1995 old_iph = ip_hdr(skb);
1997 ttl = vxlan->cfg.ttl;
1998 if (!ttl && vxlan_addr_multicast(dst))
2001 tos = vxlan->cfg.tos;
2003 tos = ip_tunnel_get_dsfield(old_iph, skb);
2005 label = vxlan->cfg.label;
2006 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2007 vxlan->cfg.port_max, true);
2010 ttl = info->key.ttl;
2011 tos = info->key.tos;
2012 label = info->key.label;
2013 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2015 if (info->options_len)
2016 md = ip_tunnel_info_opts(info);
2018 md->gbp = skb->mark;
2021 if (dst->sa.sa_family == AF_INET) {
2022 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2026 sk = sock4->sock->sk;
2028 rt = vxlan_get_route(vxlan, skb,
2029 rdst ? rdst->remote_ifindex : 0, tos,
2030 dst->sin.sin_addr.s_addr,
2031 &src->sin.sin_addr.s_addr,
2034 netdev_dbg(dev, "no route to %pI4\n",
2035 &dst->sin.sin_addr.s_addr);
2036 dev->stats.tx_carrier_errors++;
2040 if (rt->dst.dev == dev) {
2041 netdev_dbg(dev, "circular route to %pI4\n",
2042 &dst->sin.sin_addr.s_addr);
2043 dev->stats.collisions++;
2047 /* Bypass encapsulation if the destination is local */
2048 if (!info && rt->rt_flags & RTCF_LOCAL &&
2049 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2050 struct vxlan_dev *dst_vxlan;
2053 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2054 dst->sa.sa_family, dst_port,
2058 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2063 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2064 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2067 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2068 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2069 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2070 vni, md, flags, udp_sum);
2074 udp_tunnel_xmit_skb(rt, sk, skb, src->sin.sin_addr.s_addr,
2075 dst->sin.sin_addr.s_addr, tos, ttl, df,
2076 src_port, dst_port, xnet, !udp_sum);
2077 #if IS_ENABLED(CONFIG_IPV6)
2079 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2080 struct dst_entry *ndst;
2085 sk = sock6->sock->sk;
2087 ndst = vxlan6_get_route(vxlan, skb,
2088 rdst ? rdst->remote_ifindex : 0, tos,
2089 label, &dst->sin6.sin6_addr,
2090 &src->sin6.sin6_addr,
2093 netdev_dbg(dev, "no route to %pI6\n",
2094 &dst->sin6.sin6_addr);
2095 dev->stats.tx_carrier_errors++;
2099 if (ndst->dev == dev) {
2100 netdev_dbg(dev, "circular route to %pI6\n",
2101 &dst->sin6.sin6_addr);
2103 dev->stats.collisions++;
2107 /* Bypass encapsulation if the destination is local */
2108 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2109 if (!info && rt6i_flags & RTF_LOCAL &&
2110 !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2111 struct vxlan_dev *dst_vxlan;
2114 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2115 dst->sa.sa_family, dst_port,
2119 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2124 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2126 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2127 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2128 skb_scrub_packet(skb, xnet);
2129 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2130 vni, md, flags, udp_sum);
2133 dev->stats.tx_errors++;
2136 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2137 &src->sin6.sin6_addr,
2138 &dst->sin6.sin6_addr, tos, ttl,
2139 label, src_port, dst_port, !udp_sum);
2146 dev->stats.tx_dropped++;
2150 /* skb is already freed. */
2155 dev->stats.tx_errors++;
2160 /* Transmit local packets over Vxlan
2162 * Outer IP header inherits ECN and DF from inner header.
2163 * Outer UDP destination is the VXLAN assigned port.
2164 * source port is based on hash of flow
2166 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2168 struct vxlan_dev *vxlan = netdev_priv(dev);
2169 const struct ip_tunnel_info *info;
2171 bool did_rsc = false;
2172 struct vxlan_rdst *rdst, *fdst = NULL;
2173 struct vxlan_fdb *f;
2175 info = skb_tunnel_info(skb);
2177 skb_reset_mac_header(skb);
2179 if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2180 if (info && info->mode & IP_TUNNEL_INFO_TX)
2181 vxlan_xmit_one(skb, dev, NULL, false);
2184 return NETDEV_TX_OK;
2187 if (vxlan->flags & VXLAN_F_PROXY) {
2189 if (ntohs(eth->h_proto) == ETH_P_ARP)
2190 return arp_reduce(dev, skb);
2191 #if IS_ENABLED(CONFIG_IPV6)
2192 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2193 pskb_may_pull(skb, sizeof(struct ipv6hdr)
2194 + sizeof(struct nd_msg)) &&
2195 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2198 msg = (struct nd_msg *)skb_transport_header(skb);
2199 if (msg->icmph.icmp6_code == 0 &&
2200 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2201 return neigh_reduce(dev, skb);
2207 f = vxlan_find_mac(vxlan, eth->h_dest);
2210 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2211 (ntohs(eth->h_proto) == ETH_P_IP ||
2212 ntohs(eth->h_proto) == ETH_P_IPV6)) {
2213 did_rsc = route_shortcircuit(dev, skb);
2215 f = vxlan_find_mac(vxlan, eth->h_dest);
2219 f = vxlan_find_mac(vxlan, all_zeros_mac);
2221 if ((vxlan->flags & VXLAN_F_L2MISS) &&
2222 !is_multicast_ether_addr(eth->h_dest))
2223 vxlan_fdb_miss(vxlan, eth->h_dest);
2225 dev->stats.tx_dropped++;
2227 return NETDEV_TX_OK;
2231 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2232 struct sk_buff *skb1;
2238 skb1 = skb_clone(skb, GFP_ATOMIC);
2240 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2244 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2247 return NETDEV_TX_OK;
2250 /* Walk the forwarding table and purge stale entries */
2251 static void vxlan_cleanup(unsigned long arg)
2253 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2254 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2257 if (!netif_running(vxlan->dev))
2260 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2261 struct hlist_node *p, *n;
2263 spin_lock_bh(&vxlan->hash_lock);
2264 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2266 = container_of(p, struct vxlan_fdb, hlist);
2267 unsigned long timeout;
2269 if (f->state & NUD_PERMANENT)
2272 timeout = f->used + vxlan->cfg.age_interval * HZ;
2273 if (time_before_eq(timeout, jiffies)) {
2274 netdev_dbg(vxlan->dev,
2275 "garbage collect %pM\n",
2277 f->state = NUD_STALE;
2278 vxlan_fdb_destroy(vxlan, f);
2279 } else if (time_before(timeout, next_timer))
2280 next_timer = timeout;
2282 spin_unlock_bh(&vxlan->hash_lock);
2285 mod_timer(&vxlan->age_timer, next_timer);
2288 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2290 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2291 __be32 vni = vxlan->default_dst.remote_vni;
2293 spin_lock(&vn->sock_lock);
2294 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2295 spin_unlock(&vn->sock_lock);
2298 /* Setup stats when device is created */
2299 static int vxlan_init(struct net_device *dev)
2301 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2308 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2310 struct vxlan_fdb *f;
2312 spin_lock_bh(&vxlan->hash_lock);
2313 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2315 vxlan_fdb_destroy(vxlan, f);
2316 spin_unlock_bh(&vxlan->hash_lock);
2319 static void vxlan_uninit(struct net_device *dev)
2321 struct vxlan_dev *vxlan = netdev_priv(dev);
2323 vxlan_fdb_delete_default(vxlan);
2325 free_percpu(dev->tstats);
2328 /* Start ageing timer and join group when device is brought up */
2329 static int vxlan_open(struct net_device *dev)
2331 struct vxlan_dev *vxlan = netdev_priv(dev);
2334 ret = vxlan_sock_add(vxlan);
2338 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2339 ret = vxlan_igmp_join(vxlan);
2340 if (ret == -EADDRINUSE)
2343 vxlan_sock_release(vxlan);
2348 if (vxlan->cfg.age_interval)
2349 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2354 /* Purge the forwarding table */
2355 static void vxlan_flush(struct vxlan_dev *vxlan)
2359 spin_lock_bh(&vxlan->hash_lock);
2360 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2361 struct hlist_node *p, *n;
2362 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2364 = container_of(p, struct vxlan_fdb, hlist);
2365 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2366 if (!is_zero_ether_addr(f->eth_addr))
2367 vxlan_fdb_destroy(vxlan, f);
2370 spin_unlock_bh(&vxlan->hash_lock);
2373 /* Cleanup timer and forwarding table on shutdown */
2374 static int vxlan_stop(struct net_device *dev)
2376 struct vxlan_dev *vxlan = netdev_priv(dev);
2377 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2380 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2381 !vxlan_group_used(vn, vxlan))
2382 ret = vxlan_igmp_leave(vxlan);
2384 del_timer_sync(&vxlan->age_timer);
2387 vxlan_sock_release(vxlan);
2392 /* Stub, nothing needs to be done. */
2393 static void vxlan_set_multicast_list(struct net_device *dev)
2397 static int __vxlan_change_mtu(struct net_device *dev,
2398 struct net_device *lowerdev,
2399 struct vxlan_rdst *dst, int new_mtu, bool strict)
2401 int max_mtu = IP_MAX_MTU;
2404 max_mtu = lowerdev->mtu;
2406 if (dst->remote_ip.sa.sa_family == AF_INET6)
2407 max_mtu -= VXLAN6_HEADROOM;
2409 max_mtu -= VXLAN_HEADROOM;
2414 if (new_mtu > max_mtu) {
2425 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2427 struct vxlan_dev *vxlan = netdev_priv(dev);
2428 struct vxlan_rdst *dst = &vxlan->default_dst;
2429 struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2430 dst->remote_ifindex);
2431 return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2434 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2436 struct vxlan_dev *vxlan = netdev_priv(dev);
2437 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2438 __be16 sport, dport;
2440 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2441 vxlan->cfg.port_max, true);
2442 dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2444 if (ip_tunnel_info_af(info) == AF_INET) {
2445 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2450 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2451 info->key.u.ipv4.dst,
2452 &info->key.u.ipv4.src, NULL, info);
2457 #if IS_ENABLED(CONFIG_IPV6)
2458 struct dst_entry *ndst;
2460 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2461 info->key.label, &info->key.u.ipv6.dst,
2462 &info->key.u.ipv6.src, NULL, info);
2464 return PTR_ERR(ndst);
2466 #else /* !CONFIG_IPV6 */
2467 return -EPFNOSUPPORT;
2470 info->key.tp_src = sport;
2471 info->key.tp_dst = dport;
2475 static const struct net_device_ops vxlan_netdev_ether_ops = {
2476 .ndo_init = vxlan_init,
2477 .ndo_uninit = vxlan_uninit,
2478 .ndo_open = vxlan_open,
2479 .ndo_stop = vxlan_stop,
2480 .ndo_start_xmit = vxlan_xmit,
2481 .ndo_get_stats64 = ip_tunnel_get_stats64,
2482 .ndo_set_rx_mode = vxlan_set_multicast_list,
2483 .ndo_change_mtu = vxlan_change_mtu,
2484 .ndo_validate_addr = eth_validate_addr,
2485 .ndo_set_mac_address = eth_mac_addr,
2486 .ndo_fdb_add = vxlan_fdb_add,
2487 .ndo_fdb_del = vxlan_fdb_delete,
2488 .ndo_fdb_dump = vxlan_fdb_dump,
2489 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2492 static const struct net_device_ops vxlan_netdev_raw_ops = {
2493 .ndo_init = vxlan_init,
2494 .ndo_uninit = vxlan_uninit,
2495 .ndo_open = vxlan_open,
2496 .ndo_stop = vxlan_stop,
2497 .ndo_start_xmit = vxlan_xmit,
2498 .ndo_get_stats64 = ip_tunnel_get_stats64,
2499 .ndo_change_mtu = vxlan_change_mtu,
2500 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2503 /* Info for udev, that this is a virtual tunnel endpoint */
2504 static struct device_type vxlan_type = {
2508 /* Calls the ndo_udp_tunnel_add of the caller in order to
2509 * supply the listening VXLAN udp ports. Callers are expected
2510 * to implement the ndo_udp_tunnel_add.
2512 static void vxlan_push_rx_ports(struct net_device *dev)
2514 struct vxlan_sock *vs;
2515 struct net *net = dev_net(dev);
2516 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2519 spin_lock(&vn->sock_lock);
2520 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2521 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2522 udp_tunnel_push_rx_port(dev, vs->sock,
2523 (vs->flags & VXLAN_F_GPE) ?
2524 UDP_TUNNEL_TYPE_VXLAN_GPE :
2525 UDP_TUNNEL_TYPE_VXLAN);
2527 spin_unlock(&vn->sock_lock);
2530 /* Initialize the device structure. */
2531 static void vxlan_setup(struct net_device *dev)
2533 struct vxlan_dev *vxlan = netdev_priv(dev);
2536 eth_hw_addr_random(dev);
2539 dev->destructor = free_netdev;
2540 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2542 dev->features |= NETIF_F_LLTX;
2543 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2544 dev->features |= NETIF_F_RXCSUM;
2545 dev->features |= NETIF_F_GSO_SOFTWARE;
2547 dev->vlan_features = dev->features;
2548 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2549 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2550 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2551 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2552 netif_keep_dst(dev);
2553 dev->priv_flags |= IFF_NO_QUEUE;
2555 INIT_LIST_HEAD(&vxlan->next);
2556 spin_lock_init(&vxlan->hash_lock);
2558 init_timer_deferrable(&vxlan->age_timer);
2559 vxlan->age_timer.function = vxlan_cleanup;
2560 vxlan->age_timer.data = (unsigned long) vxlan;
2562 vxlan->cfg.dst_port = htons(vxlan_port);
2566 gro_cells_init(&vxlan->gro_cells, dev);
2568 for (h = 0; h < FDB_HASH_SIZE; ++h)
2569 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2572 static void vxlan_ether_setup(struct net_device *dev)
2574 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2575 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2576 dev->netdev_ops = &vxlan_netdev_ether_ops;
2579 static void vxlan_raw_setup(struct net_device *dev)
2581 dev->header_ops = NULL;
2582 dev->type = ARPHRD_NONE;
2583 dev->hard_header_len = 0;
2585 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2586 dev->netdev_ops = &vxlan_netdev_raw_ops;
2589 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2590 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2591 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2592 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2593 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2594 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2595 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2596 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2597 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2598 [IFLA_VXLAN_LABEL] = { .type = NLA_U32 },
2599 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2600 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2601 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2602 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2603 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2604 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2605 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2606 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2607 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 },
2608 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2609 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 },
2610 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
2611 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
2612 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2613 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2614 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, },
2615 [IFLA_VXLAN_GPE] = { .type = NLA_FLAG, },
2616 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG },
2619 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2621 if (tb[IFLA_ADDRESS]) {
2622 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2623 pr_debug("invalid link address (not ethernet)\n");
2627 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2628 pr_debug("invalid all zero ethernet address\n");
2629 return -EADDRNOTAVAIL;
2636 if (data[IFLA_VXLAN_ID]) {
2637 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2638 if (id >= VXLAN_VID_MASK)
2642 if (data[IFLA_VXLAN_PORT_RANGE]) {
2643 const struct ifla_vxlan_port_range *p
2644 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2646 if (ntohs(p->high) < ntohs(p->low)) {
2647 pr_debug("port range %u .. %u not valid\n",
2648 ntohs(p->low), ntohs(p->high));
2656 static void vxlan_get_drvinfo(struct net_device *netdev,
2657 struct ethtool_drvinfo *drvinfo)
2659 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2660 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2663 static const struct ethtool_ops vxlan_ethtool_ops = {
2664 .get_drvinfo = vxlan_get_drvinfo,
2665 .get_link = ethtool_op_get_link,
2668 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2669 __be16 port, u32 flags)
2671 struct socket *sock;
2672 struct udp_port_cfg udp_conf;
2675 memset(&udp_conf, 0, sizeof(udp_conf));
2678 udp_conf.family = AF_INET6;
2679 udp_conf.use_udp6_rx_checksums =
2680 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2681 udp_conf.ipv6_v6only = 1;
2683 udp_conf.family = AF_INET;
2686 udp_conf.local_udp_port = port;
2688 /* Open UDP socket */
2689 err = udp_sock_create(net, &udp_conf, &sock);
2691 return ERR_PTR(err);
2696 /* Create new listen socket if needed */
2697 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2698 __be16 port, u32 flags)
2700 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2701 struct vxlan_sock *vs;
2702 struct socket *sock;
2704 struct udp_tunnel_sock_cfg tunnel_cfg;
2706 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2708 return ERR_PTR(-ENOMEM);
2710 for (h = 0; h < VNI_HASH_SIZE; ++h)
2711 INIT_HLIST_HEAD(&vs->vni_list[h]);
2713 sock = vxlan_create_sock(net, ipv6, port, flags);
2715 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2718 return ERR_CAST(sock);
2722 atomic_set(&vs->refcnt, 1);
2723 vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2725 spin_lock(&vn->sock_lock);
2726 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2727 udp_tunnel_notify_add_rx_port(sock,
2728 (vs->flags & VXLAN_F_GPE) ?
2729 UDP_TUNNEL_TYPE_VXLAN_GPE :
2730 UDP_TUNNEL_TYPE_VXLAN);
2731 spin_unlock(&vn->sock_lock);
2733 /* Mark socket as an encapsulation socket. */
2734 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2735 tunnel_cfg.sk_user_data = vs;
2736 tunnel_cfg.encap_type = 1;
2737 tunnel_cfg.encap_rcv = vxlan_rcv;
2738 tunnel_cfg.encap_destroy = NULL;
2739 tunnel_cfg.gro_receive = vxlan_gro_receive;
2740 tunnel_cfg.gro_complete = vxlan_gro_complete;
2742 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2747 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2749 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2750 struct vxlan_sock *vs = NULL;
2752 if (!vxlan->cfg.no_share) {
2753 spin_lock(&vn->sock_lock);
2754 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2755 vxlan->cfg.dst_port, vxlan->flags);
2756 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2757 spin_unlock(&vn->sock_lock);
2760 spin_unlock(&vn->sock_lock);
2763 vs = vxlan_socket_create(vxlan->net, ipv6,
2764 vxlan->cfg.dst_port, vxlan->flags);
2767 #if IS_ENABLED(CONFIG_IPV6)
2769 rcu_assign_pointer(vxlan->vn6_sock, vs);
2772 rcu_assign_pointer(vxlan->vn4_sock, vs);
2773 vxlan_vs_add_dev(vs, vxlan);
2777 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2779 bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2780 bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2783 RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2784 #if IS_ENABLED(CONFIG_IPV6)
2785 RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2786 if (ipv6 || metadata)
2787 ret = __vxlan_sock_add(vxlan, true);
2789 if (!ret && (!ipv6 || metadata))
2790 ret = __vxlan_sock_add(vxlan, false);
2792 vxlan_sock_release(vxlan);
2796 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2797 struct vxlan_config *conf)
2799 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2800 struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2801 struct vxlan_rdst *dst = &vxlan->default_dst;
2802 unsigned short needed_headroom = ETH_HLEN;
2804 bool use_ipv6 = false;
2805 __be16 default_port = vxlan->cfg.dst_port;
2806 struct net_device *lowerdev = NULL;
2808 if (conf->flags & VXLAN_F_GPE) {
2809 /* For now, allow GPE only together with COLLECT_METADATA.
2810 * This can be relaxed later; in such case, the other side
2811 * of the PtP link will have to be provided.
2813 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2814 !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2815 pr_info("unsupported combination of extensions\n");
2819 vxlan_raw_setup(dev);
2821 vxlan_ether_setup(dev);
2824 vxlan->net = src_net;
2826 dst->remote_vni = conf->vni;
2828 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2830 /* Unless IPv6 is explicitly requested, assume IPv4 */
2831 if (!dst->remote_ip.sa.sa_family)
2832 dst->remote_ip.sa.sa_family = AF_INET;
2834 if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2835 vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2836 if (!IS_ENABLED(CONFIG_IPV6))
2837 return -EPFNOSUPPORT;
2839 vxlan->flags |= VXLAN_F_IPV6;
2842 if (conf->label && !use_ipv6) {
2843 pr_info("label only supported in use with IPv6\n");
2847 if (conf->remote_ifindex) {
2848 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2849 dst->remote_ifindex = conf->remote_ifindex;
2852 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2856 #if IS_ENABLED(CONFIG_IPV6)
2858 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2859 if (idev && idev->cnf.disable_ipv6) {
2860 pr_info("IPv6 is disabled via sysctl\n");
2867 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2869 needed_headroom = lowerdev->hard_header_len;
2870 } else if (vxlan_addr_multicast(&dst->remote_ip)) {
2871 pr_info("multicast destination requires interface to be specified\n");
2876 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2881 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2882 needed_headroom += VXLAN6_HEADROOM;
2884 needed_headroom += VXLAN_HEADROOM;
2885 dev->needed_headroom = needed_headroom;
2887 memcpy(&vxlan->cfg, conf, sizeof(*conf));
2888 if (!vxlan->cfg.dst_port) {
2889 if (conf->flags & VXLAN_F_GPE)
2890 vxlan->cfg.dst_port = 4790; /* IANA assigned VXLAN-GPE port */
2892 vxlan->cfg.dst_port = default_port;
2894 vxlan->flags |= conf->flags;
2896 if (!vxlan->cfg.age_interval)
2897 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2899 list_for_each_entry(tmp, &vn->vxlan_list, next) {
2900 if (tmp->cfg.vni == conf->vni &&
2901 (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2902 tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2903 tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2904 (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2905 (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2906 pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2911 dev->ethtool_ops = &vxlan_ethtool_ops;
2913 /* create an fdb entry for a valid default destination */
2914 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2915 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2916 &vxlan->default_dst.remote_ip,
2917 NUD_REACHABLE|NUD_PERMANENT,
2918 NLM_F_EXCL|NLM_F_CREATE,
2919 vxlan->cfg.dst_port,
2920 vxlan->default_dst.remote_vni,
2921 vxlan->default_dst.remote_ifindex,
2927 err = register_netdevice(dev);
2929 vxlan_fdb_delete_default(vxlan);
2933 list_add(&vxlan->next, &vn->vxlan_list);
2938 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2939 struct nlattr *tb[], struct nlattr *data[])
2941 struct vxlan_config conf;
2943 memset(&conf, 0, sizeof(conf));
2945 if (data[IFLA_VXLAN_ID])
2946 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2948 if (data[IFLA_VXLAN_GROUP]) {
2949 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2950 } else if (data[IFLA_VXLAN_GROUP6]) {
2951 if (!IS_ENABLED(CONFIG_IPV6))
2952 return -EPFNOSUPPORT;
2954 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2955 conf.remote_ip.sa.sa_family = AF_INET6;
2958 if (data[IFLA_VXLAN_LOCAL]) {
2959 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2960 conf.saddr.sa.sa_family = AF_INET;
2961 } else if (data[IFLA_VXLAN_LOCAL6]) {
2962 if (!IS_ENABLED(CONFIG_IPV6))
2963 return -EPFNOSUPPORT;
2965 /* TODO: respect scope id */
2966 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2967 conf.saddr.sa.sa_family = AF_INET6;
2970 if (data[IFLA_VXLAN_LINK])
2971 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2973 if (data[IFLA_VXLAN_TOS])
2974 conf.tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2976 if (data[IFLA_VXLAN_TTL])
2977 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2979 if (data[IFLA_VXLAN_LABEL])
2980 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
2981 IPV6_FLOWLABEL_MASK;
2983 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2984 conf.flags |= VXLAN_F_LEARN;
2986 if (data[IFLA_VXLAN_AGEING])
2987 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2989 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2990 conf.flags |= VXLAN_F_PROXY;
2992 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2993 conf.flags |= VXLAN_F_RSC;
2995 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2996 conf.flags |= VXLAN_F_L2MISS;
2998 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2999 conf.flags |= VXLAN_F_L3MISS;
3001 if (data[IFLA_VXLAN_LIMIT])
3002 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3004 if (data[IFLA_VXLAN_COLLECT_METADATA] &&
3005 nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3006 conf.flags |= VXLAN_F_COLLECT_METADATA;
3008 if (data[IFLA_VXLAN_PORT_RANGE]) {
3009 const struct ifla_vxlan_port_range *p
3010 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3011 conf.port_min = ntohs(p->low);
3012 conf.port_max = ntohs(p->high);
3015 if (data[IFLA_VXLAN_PORT])
3016 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3018 if (data[IFLA_VXLAN_UDP_CSUM] &&
3019 !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3020 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3022 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3023 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3024 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3026 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3027 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3028 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3030 if (data[IFLA_VXLAN_REMCSUM_TX] &&
3031 nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3032 conf.flags |= VXLAN_F_REMCSUM_TX;
3034 if (data[IFLA_VXLAN_REMCSUM_RX] &&
3035 nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3036 conf.flags |= VXLAN_F_REMCSUM_RX;
3038 if (data[IFLA_VXLAN_GBP])
3039 conf.flags |= VXLAN_F_GBP;
3041 if (data[IFLA_VXLAN_GPE])
3042 conf.flags |= VXLAN_F_GPE;
3044 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3045 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3048 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3050 return vxlan_dev_configure(src_net, dev, &conf);
3053 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3055 struct vxlan_dev *vxlan = netdev_priv(dev);
3056 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3058 spin_lock(&vn->sock_lock);
3059 if (!hlist_unhashed(&vxlan->hlist))
3060 hlist_del_rcu(&vxlan->hlist);
3061 spin_unlock(&vn->sock_lock);
3063 gro_cells_destroy(&vxlan->gro_cells);
3064 list_del(&vxlan->next);
3065 unregister_netdevice_queue(dev, head);
3068 static size_t vxlan_get_size(const struct net_device *dev)
3071 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
3072 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3073 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3074 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3075 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
3076 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
3077 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3078 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
3079 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
3080 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
3081 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
3082 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
3083 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */
3084 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3085 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3086 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3087 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3088 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3089 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3090 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3091 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3092 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3096 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3098 const struct vxlan_dev *vxlan = netdev_priv(dev);
3099 const struct vxlan_rdst *dst = &vxlan->default_dst;
3100 struct ifla_vxlan_port_range ports = {
3101 .low = htons(vxlan->cfg.port_min),
3102 .high = htons(vxlan->cfg.port_max),
3105 if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3106 goto nla_put_failure;
3108 if (!vxlan_addr_any(&dst->remote_ip)) {
3109 if (dst->remote_ip.sa.sa_family == AF_INET) {
3110 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3111 dst->remote_ip.sin.sin_addr.s_addr))
3112 goto nla_put_failure;
3113 #if IS_ENABLED(CONFIG_IPV6)
3115 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3116 &dst->remote_ip.sin6.sin6_addr))
3117 goto nla_put_failure;
3122 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3123 goto nla_put_failure;
3125 if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3126 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3127 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3128 vxlan->cfg.saddr.sin.sin_addr.s_addr))
3129 goto nla_put_failure;
3130 #if IS_ENABLED(CONFIG_IPV6)
3132 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3133 &vxlan->cfg.saddr.sin6.sin6_addr))
3134 goto nla_put_failure;
3139 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3140 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3141 nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3142 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3143 !!(vxlan->flags & VXLAN_F_LEARN)) ||
3144 nla_put_u8(skb, IFLA_VXLAN_PROXY,
3145 !!(vxlan->flags & VXLAN_F_PROXY)) ||
3146 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3147 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3148 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3149 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3150 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3151 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3152 !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3153 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3154 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3155 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3156 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3157 !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3158 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3159 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3160 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3161 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3162 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3163 !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3164 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3165 !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3166 goto nla_put_failure;
3168 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3169 goto nla_put_failure;
3171 if (vxlan->flags & VXLAN_F_GBP &&
3172 nla_put_flag(skb, IFLA_VXLAN_GBP))
3173 goto nla_put_failure;
3175 if (vxlan->flags & VXLAN_F_GPE &&
3176 nla_put_flag(skb, IFLA_VXLAN_GPE))
3177 goto nla_put_failure;
3179 if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3180 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3181 goto nla_put_failure;
3189 static struct net *vxlan_get_link_net(const struct net_device *dev)
3191 struct vxlan_dev *vxlan = netdev_priv(dev);
3196 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3198 .maxtype = IFLA_VXLAN_MAX,
3199 .policy = vxlan_policy,
3200 .priv_size = sizeof(struct vxlan_dev),
3201 .setup = vxlan_setup,
3202 .validate = vxlan_validate,
3203 .newlink = vxlan_newlink,
3204 .dellink = vxlan_dellink,
3205 .get_size = vxlan_get_size,
3206 .fill_info = vxlan_fill_info,
3207 .get_link_net = vxlan_get_link_net,
3210 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3211 u8 name_assign_type,
3212 struct vxlan_config *conf)
3214 struct nlattr *tb[IFLA_MAX + 1];
3215 struct net_device *dev;
3218 memset(&tb, 0, sizeof(tb));
3220 dev = rtnl_create_link(net, name, name_assign_type,
3221 &vxlan_link_ops, tb);
3225 err = vxlan_dev_configure(net, dev, conf);
3228 return ERR_PTR(err);
3231 err = rtnl_configure_link(dev, NULL);
3233 LIST_HEAD(list_kill);
3235 vxlan_dellink(dev, &list_kill);
3236 unregister_netdevice_many(&list_kill);
3237 return ERR_PTR(err);
3242 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3244 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3245 struct net_device *dev)
3247 struct vxlan_dev *vxlan, *next;
3248 LIST_HEAD(list_kill);
3250 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3251 struct vxlan_rdst *dst = &vxlan->default_dst;
3253 /* In case we created vxlan device with carrier
3254 * and we loose the carrier due to module unload
3255 * we also need to remove vxlan device. In other
3256 * cases, it's not necessary and remote_ifindex
3257 * is 0 here, so no matches.
3259 if (dst->remote_ifindex == dev->ifindex)
3260 vxlan_dellink(vxlan->dev, &list_kill);
3263 unregister_netdevice_many(&list_kill);
3266 static int vxlan_netdevice_event(struct notifier_block *unused,
3267 unsigned long event, void *ptr)
3269 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3270 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3272 if (event == NETDEV_UNREGISTER)
3273 vxlan_handle_lowerdev_unregister(vn, dev);
3274 else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3275 vxlan_push_rx_ports(dev);
3280 static struct notifier_block vxlan_notifier_block __read_mostly = {
3281 .notifier_call = vxlan_netdevice_event,
3284 static __net_init int vxlan_init_net(struct net *net)
3286 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3289 INIT_LIST_HEAD(&vn->vxlan_list);
3290 spin_lock_init(&vn->sock_lock);
3292 for (h = 0; h < PORT_HASH_SIZE; ++h)
3293 INIT_HLIST_HEAD(&vn->sock_list[h]);
3298 static void __net_exit vxlan_exit_net(struct net *net)
3300 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3301 struct vxlan_dev *vxlan, *next;
3302 struct net_device *dev, *aux;
3306 for_each_netdev_safe(net, dev, aux)
3307 if (dev->rtnl_link_ops == &vxlan_link_ops)
3308 unregister_netdevice_queue(dev, &list);
3310 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3311 /* If vxlan->dev is in the same netns, it has already been added
3312 * to the list by the previous loop.
3314 if (!net_eq(dev_net(vxlan->dev), net)) {
3315 gro_cells_destroy(&vxlan->gro_cells);
3316 unregister_netdevice_queue(vxlan->dev, &list);
3320 unregister_netdevice_many(&list);
3324 static struct pernet_operations vxlan_net_ops = {
3325 .init = vxlan_init_net,
3326 .exit = vxlan_exit_net,
3327 .id = &vxlan_net_id,
3328 .size = sizeof(struct vxlan_net),
3331 static int __init vxlan_init_module(void)
3335 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3337 rc = register_pernet_subsys(&vxlan_net_ops);
3341 rc = register_netdevice_notifier(&vxlan_notifier_block);
3345 rc = rtnl_link_register(&vxlan_link_ops);
3351 unregister_netdevice_notifier(&vxlan_notifier_block);
3353 unregister_pernet_subsys(&vxlan_net_ops);
3357 late_initcall(vxlan_init_module);
3359 static void __exit vxlan_cleanup_module(void)
3361 rtnl_link_unregister(&vxlan_link_ops);
3362 unregister_netdevice_notifier(&vxlan_notifier_block);
3363 unregister_pernet_subsys(&vxlan_net_ops);
3364 /* rcu_barrier() is called by netns */
3366 module_exit(vxlan_cleanup_module);
3368 MODULE_LICENSE("GPL");
3369 MODULE_VERSION(VXLAN_VERSION);
3370 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3371 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3372 MODULE_ALIAS_RTNL_LINK("vxlan");