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;
615 f = __vxlan_find_mac(vxlan, mac);
617 if (flags & NLM_F_EXCL) {
618 netdev_dbg(vxlan->dev,
619 "lost race to create %pM\n", mac);
622 if (f->state != state) {
624 f->updated = jiffies;
627 if (f->flags != ndm_flags) {
628 f->flags = ndm_flags;
629 f->updated = jiffies;
632 if ((flags & NLM_F_REPLACE)) {
633 /* Only change unicasts */
634 if (!(is_multicast_ether_addr(f->eth_addr) ||
635 is_zero_ether_addr(f->eth_addr))) {
636 notify |= vxlan_fdb_replace(f, ip, port, vni,
641 if ((flags & NLM_F_APPEND) &&
642 (is_multicast_ether_addr(f->eth_addr) ||
643 is_zero_ether_addr(f->eth_addr))) {
644 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
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 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
679 hlist_add_head_rcu(&f->hlist,
680 vxlan_fdb_head(vxlan, mac));
685 rd = first_remote_rtnl(f);
686 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
692 static void vxlan_fdb_free(struct rcu_head *head)
694 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
695 struct vxlan_rdst *rd, *nd;
697 list_for_each_entry_safe(rd, nd, &f->remotes, list) {
698 dst_cache_destroy(&rd->dst_cache);
704 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
706 netdev_dbg(vxlan->dev,
707 "delete %pM\n", f->eth_addr);
710 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
712 hlist_del_rcu(&f->hlist);
713 call_rcu(&f->rcu, vxlan_fdb_free);
716 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
717 union vxlan_addr *ip, __be16 *port, __be32 *vni,
720 struct net *net = dev_net(vxlan->dev);
724 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
728 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
729 if (remote->sa.sa_family == AF_INET) {
730 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
731 ip->sa.sa_family = AF_INET;
732 #if IS_ENABLED(CONFIG_IPV6)
734 ip->sin6.sin6_addr = in6addr_any;
735 ip->sa.sa_family = AF_INET6;
741 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
743 *port = nla_get_be16(tb[NDA_PORT]);
745 *port = vxlan->cfg.dst_port;
749 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
751 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
753 *vni = vxlan->default_dst.remote_vni;
756 if (tb[NDA_IFINDEX]) {
757 struct net_device *tdev;
759 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
761 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
762 tdev = __dev_get_by_index(net, *ifindex);
764 return -EADDRNOTAVAIL;
772 /* Add static entry (via netlink) */
773 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
774 struct net_device *dev,
775 const unsigned char *addr, u16 vid, u16 flags)
777 struct vxlan_dev *vxlan = netdev_priv(dev);
778 /* struct net *net = dev_net(vxlan->dev); */
785 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
786 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
791 if (tb[NDA_DST] == NULL)
794 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
798 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
799 return -EAFNOSUPPORT;
801 spin_lock_bh(&vxlan->hash_lock);
802 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
803 port, vni, ifindex, ndm->ndm_flags);
804 spin_unlock_bh(&vxlan->hash_lock);
809 /* Delete entry (via netlink) */
810 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
811 struct net_device *dev,
812 const unsigned char *addr, u16 vid)
814 struct vxlan_dev *vxlan = netdev_priv(dev);
816 struct vxlan_rdst *rd = NULL;
823 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
829 spin_lock_bh(&vxlan->hash_lock);
830 f = vxlan_find_mac(vxlan, addr);
834 if (!vxlan_addr_any(&ip)) {
835 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
842 /* remove a destination if it's not the only one on the list,
843 * otherwise destroy the fdb entry
845 if (rd && !list_is_singular(&f->remotes)) {
846 list_del_rcu(&rd->list);
847 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
852 vxlan_fdb_destroy(vxlan, f);
855 spin_unlock_bh(&vxlan->hash_lock);
860 /* Dump forwarding table */
861 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
862 struct net_device *dev,
863 struct net_device *filter_dev, int *idx)
865 struct vxlan_dev *vxlan = netdev_priv(dev);
869 for (h = 0; h < FDB_HASH_SIZE; ++h) {
872 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
873 struct vxlan_rdst *rd;
875 list_for_each_entry_rcu(rd, &f->remotes, list) {
876 if (*idx < cb->args[2])
879 err = vxlan_fdb_info(skb, vxlan, f,
880 NETLINK_CB(cb->skb).portid,
895 /* Watch incoming packets to learn mapping between Ethernet address
896 * and Tunnel endpoint.
897 * Return true if packet is bogus and should be dropped.
899 static bool vxlan_snoop(struct net_device *dev,
900 union vxlan_addr *src_ip, const u8 *src_mac)
902 struct vxlan_dev *vxlan = netdev_priv(dev);
905 f = vxlan_find_mac(vxlan, src_mac);
907 struct vxlan_rdst *rdst = first_remote_rcu(f);
909 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
912 /* Don't migrate static entries, drop packets */
913 if (f->state & NUD_NOARP)
918 "%pM migrated from %pIS to %pIS\n",
919 src_mac, &rdst->remote_ip.sa, &src_ip->sa);
921 rdst->remote_ip = *src_ip;
922 f->updated = jiffies;
923 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
925 /* learned new entry */
926 spin_lock(&vxlan->hash_lock);
928 /* close off race between vxlan_flush and incoming packets */
929 if (netif_running(dev))
930 vxlan_fdb_create(vxlan, src_mac, src_ip,
932 NLM_F_EXCL|NLM_F_CREATE,
934 vxlan->default_dst.remote_vni,
936 spin_unlock(&vxlan->hash_lock);
942 /* See if multicast group is already in use by other ID */
943 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
945 struct vxlan_dev *vxlan;
946 struct vxlan_sock *sock4;
947 #if IS_ENABLED(CONFIG_IPV6)
948 struct vxlan_sock *sock6;
950 unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
952 sock4 = rtnl_dereference(dev->vn4_sock);
954 /* The vxlan_sock is only used by dev, leaving group has
955 * no effect on other vxlan devices.
957 if (family == AF_INET && sock4 && atomic_read(&sock4->refcnt) == 1)
959 #if IS_ENABLED(CONFIG_IPV6)
960 sock6 = rtnl_dereference(dev->vn6_sock);
961 if (family == AF_INET6 && sock6 && atomic_read(&sock6->refcnt) == 1)
965 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
966 if (!netif_running(vxlan->dev) || vxlan == dev)
969 if (family == AF_INET &&
970 rtnl_dereference(vxlan->vn4_sock) != sock4)
972 #if IS_ENABLED(CONFIG_IPV6)
973 if (family == AF_INET6 &&
974 rtnl_dereference(vxlan->vn6_sock) != sock6)
978 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
979 &dev->default_dst.remote_ip))
982 if (vxlan->default_dst.remote_ifindex !=
983 dev->default_dst.remote_ifindex)
992 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
994 struct vxlan_net *vn;
998 if (!atomic_dec_and_test(&vs->refcnt))
1001 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1002 spin_lock(&vn->sock_lock);
1003 hlist_del_rcu(&vs->hlist);
1004 udp_tunnel_notify_del_rx_port(vs->sock,
1005 (vs->flags & VXLAN_F_GPE) ?
1006 UDP_TUNNEL_TYPE_VXLAN_GPE :
1007 UDP_TUNNEL_TYPE_VXLAN);
1008 spin_unlock(&vn->sock_lock);
1013 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1015 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1016 #if IS_ENABLED(CONFIG_IPV6)
1017 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1019 rcu_assign_pointer(vxlan->vn6_sock, NULL);
1022 rcu_assign_pointer(vxlan->vn4_sock, NULL);
1025 if (__vxlan_sock_release_prep(sock4)) {
1026 udp_tunnel_sock_release(sock4->sock);
1030 #if IS_ENABLED(CONFIG_IPV6)
1031 if (__vxlan_sock_release_prep(sock6)) {
1032 udp_tunnel_sock_release(sock6->sock);
1038 /* Update multicast group membership when first VNI on
1039 * multicast address is brought up
1041 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1044 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1045 int ifindex = vxlan->default_dst.remote_ifindex;
1048 if (ip->sa.sa_family == AF_INET) {
1049 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1050 struct ip_mreqn mreq = {
1051 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1052 .imr_ifindex = ifindex,
1055 sk = sock4->sock->sk;
1057 ret = ip_mc_join_group(sk, &mreq);
1059 #if IS_ENABLED(CONFIG_IPV6)
1061 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1063 sk = sock6->sock->sk;
1065 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1066 &ip->sin6.sin6_addr);
1074 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1075 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1078 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1079 int ifindex = vxlan->default_dst.remote_ifindex;
1082 if (ip->sa.sa_family == AF_INET) {
1083 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1084 struct ip_mreqn mreq = {
1085 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1086 .imr_ifindex = ifindex,
1089 sk = sock4->sock->sk;
1091 ret = ip_mc_leave_group(sk, &mreq);
1093 #if IS_ENABLED(CONFIG_IPV6)
1095 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1097 sk = sock6->sock->sk;
1099 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1100 &ip->sin6.sin6_addr);
1108 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1109 struct sk_buff *skb, u32 vxflags)
1111 size_t start, offset;
1113 if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1116 start = vxlan_rco_start(unparsed->vx_vni);
1117 offset = start + vxlan_rco_offset(unparsed->vx_vni);
1119 if (!pskb_may_pull(skb, offset + sizeof(u16)))
1122 skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1123 !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1125 unparsed->vx_flags &= ~VXLAN_HF_RCO;
1126 unparsed->vx_vni &= VXLAN_VNI_MASK;
1130 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1131 struct sk_buff *skb, u32 vxflags,
1132 struct vxlan_metadata *md)
1134 struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1135 struct metadata_dst *tun_dst;
1137 if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1140 md->gbp = ntohs(gbp->policy_id);
1142 tun_dst = (struct metadata_dst *)skb_dst(skb);
1144 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1145 tun_dst->u.tun_info.options_len = sizeof(*md);
1147 if (gbp->dont_learn)
1148 md->gbp |= VXLAN_GBP_DONT_LEARN;
1150 if (gbp->policy_applied)
1151 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1153 /* In flow-based mode, GBP is carried in dst_metadata */
1154 if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1155 skb->mark = md->gbp;
1157 unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1160 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1162 struct sk_buff *skb, u32 vxflags)
1164 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1166 /* Need to have Next Protocol set for interfaces in GPE mode. */
1167 if (!gpe->np_applied)
1169 /* "The initial version is 0. If a receiver does not support the
1170 * version indicated it MUST drop the packet.
1172 if (gpe->version != 0)
1174 /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1175 * processing MUST occur." However, we don't implement OAM
1176 * processing, thus drop the packet.
1181 switch (gpe->next_protocol) {
1182 case VXLAN_GPE_NP_IPV4:
1183 *protocol = htons(ETH_P_IP);
1185 case VXLAN_GPE_NP_IPV6:
1186 *protocol = htons(ETH_P_IPV6);
1188 case VXLAN_GPE_NP_ETHERNET:
1189 *protocol = htons(ETH_P_TEB);
1195 unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1199 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1200 struct vxlan_sock *vs,
1201 struct sk_buff *skb)
1203 union vxlan_addr saddr;
1205 skb_reset_mac_header(skb);
1206 skb->protocol = eth_type_trans(skb, vxlan->dev);
1207 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1209 /* Ignore packet loops (and multicast echo) */
1210 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1213 /* Get address from the outer IP header */
1214 if (vxlan_get_sk_family(vs) == AF_INET) {
1215 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1216 saddr.sa.sa_family = AF_INET;
1217 #if IS_ENABLED(CONFIG_IPV6)
1219 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1220 saddr.sa.sa_family = AF_INET6;
1224 if ((vxlan->flags & VXLAN_F_LEARN) &&
1225 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1231 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1232 struct sk_buff *skb)
1236 if (vxlan_get_sk_family(vs) == AF_INET)
1237 err = IP_ECN_decapsulate(oiph, skb);
1238 #if IS_ENABLED(CONFIG_IPV6)
1240 err = IP6_ECN_decapsulate(oiph, skb);
1243 if (unlikely(err) && log_ecn_error) {
1244 if (vxlan_get_sk_family(vs) == AF_INET)
1245 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1246 &((struct iphdr *)oiph)->saddr,
1247 ((struct iphdr *)oiph)->tos);
1249 net_info_ratelimited("non-ECT from %pI6\n",
1250 &((struct ipv6hdr *)oiph)->saddr);
1255 /* Callback from net/ipv4/udp.c to receive packets */
1256 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1258 struct pcpu_sw_netstats *stats;
1259 struct vxlan_dev *vxlan;
1260 struct vxlan_sock *vs;
1261 struct vxlanhdr unparsed;
1262 struct vxlan_metadata _md;
1263 struct vxlan_metadata *md = &_md;
1264 __be16 protocol = htons(ETH_P_TEB);
1265 bool raw_proto = false;
1268 /* Need UDP and VXLAN header to be present */
1269 if (!pskb_may_pull(skb, VXLAN_HLEN))
1272 unparsed = *vxlan_hdr(skb);
1273 /* VNI flag always required to be set */
1274 if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1275 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1276 ntohl(vxlan_hdr(skb)->vx_flags),
1277 ntohl(vxlan_hdr(skb)->vx_vni));
1278 /* Return non vxlan pkt */
1281 unparsed.vx_flags &= ~VXLAN_HF_VNI;
1282 unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1284 vs = rcu_dereference_sk_user_data(sk);
1288 vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1292 /* For backwards compatibility, only allow reserved fields to be
1293 * used by VXLAN extensions if explicitly requested.
1295 if (vs->flags & VXLAN_F_GPE) {
1296 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1301 if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1302 !net_eq(vxlan->net, dev_net(vxlan->dev))))
1305 if (vxlan_collect_metadata(vs)) {
1306 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1307 struct metadata_dst *tun_dst;
1309 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1310 key32_to_tunnel_id(vni), sizeof(*md));
1315 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1317 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1319 memset(md, 0, sizeof(*md));
1322 if (vs->flags & VXLAN_F_REMCSUM_RX)
1323 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1325 if (vs->flags & VXLAN_F_GBP)
1326 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1327 /* Note that GBP and GPE can never be active together. This is
1328 * ensured in vxlan_dev_configure.
1331 if (unparsed.vx_flags || unparsed.vx_vni) {
1332 /* If there are any unprocessed flags remaining treat
1333 * this as a malformed packet. This behavior diverges from
1334 * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1335 * in reserved fields are to be ignored. The approach here
1336 * maintains compatibility with previous stack code, and also
1337 * is more robust and provides a little more security in
1338 * adding extensions to VXLAN.
1344 if (!vxlan_set_mac(vxlan, vs, skb))
1347 skb_reset_mac_header(skb);
1348 skb->dev = vxlan->dev;
1349 skb->pkt_type = PACKET_HOST;
1352 oiph = skb_network_header(skb);
1353 skb_reset_network_header(skb);
1355 if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1356 ++vxlan->dev->stats.rx_frame_errors;
1357 ++vxlan->dev->stats.rx_errors;
1361 stats = this_cpu_ptr(vxlan->dev->tstats);
1362 u64_stats_update_begin(&stats->syncp);
1363 stats->rx_packets++;
1364 stats->rx_bytes += skb->len;
1365 u64_stats_update_end(&stats->syncp);
1367 gro_cells_receive(&vxlan->gro_cells, skb);
1371 /* Consume bad packet */
1376 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1378 struct vxlan_dev *vxlan = netdev_priv(dev);
1379 struct arphdr *parp;
1382 struct neighbour *n;
1384 if (dev->flags & IFF_NOARP)
1387 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1388 dev->stats.tx_dropped++;
1391 parp = arp_hdr(skb);
1393 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1394 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1395 parp->ar_pro != htons(ETH_P_IP) ||
1396 parp->ar_op != htons(ARPOP_REQUEST) ||
1397 parp->ar_hln != dev->addr_len ||
1400 arpptr = (u8 *)parp + sizeof(struct arphdr);
1402 arpptr += dev->addr_len; /* sha */
1403 memcpy(&sip, arpptr, sizeof(sip));
1404 arpptr += sizeof(sip);
1405 arpptr += dev->addr_len; /* tha */
1406 memcpy(&tip, arpptr, sizeof(tip));
1408 if (ipv4_is_loopback(tip) ||
1409 ipv4_is_multicast(tip))
1412 n = neigh_lookup(&arp_tbl, &tip, dev);
1415 struct vxlan_fdb *f;
1416 struct sk_buff *reply;
1418 if (!(n->nud_state & NUD_CONNECTED)) {
1423 f = vxlan_find_mac(vxlan, n->ha);
1424 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1425 /* bridge-local neighbor */
1430 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1438 skb_reset_mac_header(reply);
1439 __skb_pull(reply, skb_network_offset(reply));
1440 reply->ip_summed = CHECKSUM_UNNECESSARY;
1441 reply->pkt_type = PACKET_HOST;
1443 if (netif_rx_ni(reply) == NET_RX_DROP)
1444 dev->stats.rx_dropped++;
1445 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1446 union vxlan_addr ipa = {
1447 .sin.sin_addr.s_addr = tip,
1448 .sin.sin_family = AF_INET,
1451 vxlan_ip_miss(dev, &ipa);
1455 return NETDEV_TX_OK;
1458 #if IS_ENABLED(CONFIG_IPV6)
1459 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1460 struct neighbour *n, bool isrouter)
1462 struct net_device *dev = request->dev;
1463 struct sk_buff *reply;
1464 struct nd_msg *ns, *na;
1465 struct ipv6hdr *pip6;
1467 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1474 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1475 sizeof(*na) + na_olen + dev->needed_tailroom;
1476 reply = alloc_skb(len, GFP_ATOMIC);
1480 reply->protocol = htons(ETH_P_IPV6);
1482 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1483 skb_push(reply, sizeof(struct ethhdr));
1484 skb_reset_mac_header(reply);
1486 ns = (struct nd_msg *)skb_transport_header(request);
1488 daddr = eth_hdr(request)->h_source;
1489 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1490 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1491 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1492 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1497 /* Ethernet header */
1498 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1499 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1500 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1501 reply->protocol = htons(ETH_P_IPV6);
1503 skb_pull(reply, sizeof(struct ethhdr));
1504 skb_reset_network_header(reply);
1505 skb_put(reply, sizeof(struct ipv6hdr));
1509 pip6 = ipv6_hdr(reply);
1510 memset(pip6, 0, sizeof(struct ipv6hdr));
1512 pip6->priority = ipv6_hdr(request)->priority;
1513 pip6->nexthdr = IPPROTO_ICMPV6;
1514 pip6->hop_limit = 255;
1515 pip6->daddr = ipv6_hdr(request)->saddr;
1516 pip6->saddr = *(struct in6_addr *)n->primary_key;
1518 skb_pull(reply, sizeof(struct ipv6hdr));
1519 skb_reset_transport_header(reply);
1521 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1523 /* Neighbor Advertisement */
1524 memset(na, 0, sizeof(*na)+na_olen);
1525 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1526 na->icmph.icmp6_router = isrouter;
1527 na->icmph.icmp6_override = 1;
1528 na->icmph.icmp6_solicited = 1;
1529 na->target = ns->target;
1530 ether_addr_copy(&na->opt[2], n->ha);
1531 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1532 na->opt[1] = na_olen >> 3;
1534 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1535 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1536 csum_partial(na, sizeof(*na)+na_olen, 0));
1538 pip6->payload_len = htons(sizeof(*na)+na_olen);
1540 skb_push(reply, sizeof(struct ipv6hdr));
1542 reply->ip_summed = CHECKSUM_UNNECESSARY;
1547 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1549 struct vxlan_dev *vxlan = netdev_priv(dev);
1551 const struct ipv6hdr *iphdr;
1552 const struct in6_addr *saddr, *daddr;
1553 struct neighbour *n;
1554 struct inet6_dev *in6_dev;
1556 in6_dev = __in6_dev_get(dev);
1560 iphdr = ipv6_hdr(skb);
1561 saddr = &iphdr->saddr;
1562 daddr = &iphdr->daddr;
1564 msg = (struct nd_msg *)skb_transport_header(skb);
1565 if (msg->icmph.icmp6_code != 0 ||
1566 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1569 if (ipv6_addr_loopback(daddr) ||
1570 ipv6_addr_is_multicast(&msg->target))
1573 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1576 struct vxlan_fdb *f;
1577 struct sk_buff *reply;
1579 if (!(n->nud_state & NUD_CONNECTED)) {
1584 f = vxlan_find_mac(vxlan, n->ha);
1585 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1586 /* bridge-local neighbor */
1591 reply = vxlan_na_create(skb, n,
1592 !!(f ? f->flags & NTF_ROUTER : 0));
1599 if (netif_rx_ni(reply) == NET_RX_DROP)
1600 dev->stats.rx_dropped++;
1602 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1603 union vxlan_addr ipa = {
1604 .sin6.sin6_addr = msg->target,
1605 .sin6.sin6_family = AF_INET6,
1608 vxlan_ip_miss(dev, &ipa);
1613 return NETDEV_TX_OK;
1617 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1619 struct vxlan_dev *vxlan = netdev_priv(dev);
1620 struct neighbour *n;
1622 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1626 switch (ntohs(eth_hdr(skb)->h_proto)) {
1631 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1634 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1635 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1636 union vxlan_addr ipa = {
1637 .sin.sin_addr.s_addr = pip->daddr,
1638 .sin.sin_family = AF_INET,
1641 vxlan_ip_miss(dev, &ipa);
1647 #if IS_ENABLED(CONFIG_IPV6)
1650 struct ipv6hdr *pip6;
1652 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1654 pip6 = ipv6_hdr(skb);
1655 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1656 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1657 union vxlan_addr ipa = {
1658 .sin6.sin6_addr = pip6->daddr,
1659 .sin6.sin6_family = AF_INET6,
1662 vxlan_ip_miss(dev, &ipa);
1676 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1678 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1680 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1689 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1690 struct vxlan_metadata *md)
1692 struct vxlanhdr_gbp *gbp;
1697 gbp = (struct vxlanhdr_gbp *)vxh;
1698 vxh->vx_flags |= VXLAN_HF_GBP;
1700 if (md->gbp & VXLAN_GBP_DONT_LEARN)
1701 gbp->dont_learn = 1;
1703 if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1704 gbp->policy_applied = 1;
1706 gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1709 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1712 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1714 gpe->np_applied = 1;
1717 case htons(ETH_P_IP):
1718 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1720 case htons(ETH_P_IPV6):
1721 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1723 case htons(ETH_P_TEB):
1724 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1727 return -EPFNOSUPPORT;
1730 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1731 int iphdr_len, __be32 vni,
1732 struct vxlan_metadata *md, u32 vxflags,
1735 struct vxlanhdr *vxh;
1738 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1739 __be16 inner_protocol = htons(ETH_P_TEB);
1741 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1742 skb->ip_summed == CHECKSUM_PARTIAL) {
1743 int csum_start = skb_checksum_start_offset(skb);
1745 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1746 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1747 (skb->csum_offset == offsetof(struct udphdr, check) ||
1748 skb->csum_offset == offsetof(struct tcphdr, check)))
1749 type |= SKB_GSO_TUNNEL_REMCSUM;
1752 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1753 + VXLAN_HLEN + iphdr_len
1754 + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1756 /* Need space for new headers (invalidates iph ptr) */
1757 err = skb_cow_head(skb, min_headroom);
1761 skb = vlan_hwaccel_push_inside(skb);
1765 err = iptunnel_handle_offloads(skb, type);
1769 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1770 vxh->vx_flags = VXLAN_HF_VNI;
1771 vxh->vx_vni = vxlan_vni_field(vni);
1773 if (type & SKB_GSO_TUNNEL_REMCSUM) {
1776 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1777 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1778 vxh->vx_flags |= VXLAN_HF_RCO;
1780 if (!skb_is_gso(skb)) {
1781 skb->ip_summed = CHECKSUM_NONE;
1782 skb->encapsulation = 0;
1786 if (vxflags & VXLAN_F_GBP)
1787 vxlan_build_gbp_hdr(vxh, vxflags, md);
1788 if (vxflags & VXLAN_F_GPE) {
1789 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1792 inner_protocol = skb->protocol;
1795 skb_set_inner_protocol(skb, inner_protocol);
1803 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1804 struct sk_buff *skb, int oif, u8 tos,
1805 __be32 daddr, __be32 *saddr,
1806 struct dst_cache *dst_cache,
1807 const struct ip_tunnel_info *info)
1809 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1810 struct rtable *rt = NULL;
1816 rt = dst_cache_get_ip4(dst_cache, saddr);
1821 memset(&fl4, 0, sizeof(fl4));
1822 fl4.flowi4_oif = oif;
1823 fl4.flowi4_tos = RT_TOS(tos);
1824 fl4.flowi4_mark = skb->mark;
1825 fl4.flowi4_proto = IPPROTO_UDP;
1829 rt = ip_route_output_key(vxlan->net, &fl4);
1833 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1838 #if IS_ENABLED(CONFIG_IPV6)
1839 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1840 struct sk_buff *skb, int oif, u8 tos,
1842 const struct in6_addr *daddr,
1843 struct in6_addr *saddr,
1844 struct dst_cache *dst_cache,
1845 const struct ip_tunnel_info *info)
1847 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
1848 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1849 struct dst_entry *ndst;
1854 return ERR_PTR(-EIO);
1859 ndst = dst_cache_get_ip6(dst_cache, saddr);
1864 memset(&fl6, 0, sizeof(fl6));
1865 fl6.flowi6_oif = oif;
1868 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1869 fl6.flowi6_mark = skb->mark;
1870 fl6.flowi6_proto = IPPROTO_UDP;
1872 err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1876 return ERR_PTR(err);
1880 dst_cache_set_ip6(dst_cache, ndst, saddr);
1885 /* Bypass encapsulation if the destination is local */
1886 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1887 struct vxlan_dev *dst_vxlan)
1889 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1890 union vxlan_addr loopback;
1891 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1892 struct net_device *dev = skb->dev;
1895 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1896 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1897 skb->pkt_type = PACKET_HOST;
1898 skb->encapsulation = 0;
1899 skb->dev = dst_vxlan->dev;
1900 __skb_pull(skb, skb_network_offset(skb));
1902 if (remote_ip->sa.sa_family == AF_INET) {
1903 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1904 loopback.sa.sa_family = AF_INET;
1905 #if IS_ENABLED(CONFIG_IPV6)
1907 loopback.sin6.sin6_addr = in6addr_loopback;
1908 loopback.sa.sa_family = AF_INET6;
1912 if (dst_vxlan->flags & VXLAN_F_LEARN)
1913 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1915 u64_stats_update_begin(&tx_stats->syncp);
1916 tx_stats->tx_packets++;
1917 tx_stats->tx_bytes += len;
1918 u64_stats_update_end(&tx_stats->syncp);
1920 if (netif_rx(skb) == NET_RX_SUCCESS) {
1921 u64_stats_update_begin(&rx_stats->syncp);
1922 rx_stats->rx_packets++;
1923 rx_stats->rx_bytes += len;
1924 u64_stats_update_end(&rx_stats->syncp);
1926 dev->stats.rx_dropped++;
1930 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1931 struct vxlan_rdst *rdst, bool did_rsc)
1933 struct dst_cache *dst_cache;
1934 struct ip_tunnel_info *info;
1935 struct vxlan_dev *vxlan = netdev_priv(dev);
1937 struct rtable *rt = NULL;
1938 const struct iphdr *old_iph;
1939 union vxlan_addr *dst;
1940 union vxlan_addr remote_ip, local_ip;
1941 union vxlan_addr *src;
1942 struct vxlan_metadata _md;
1943 struct vxlan_metadata *md = &_md;
1944 __be16 src_port = 0, dst_port;
1949 u32 flags = vxlan->flags;
1950 bool udp_sum = false;
1951 bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1953 info = skb_tunnel_info(skb);
1956 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1957 vni = rdst->remote_vni;
1958 dst = &rdst->remote_ip;
1959 src = &vxlan->cfg.saddr;
1960 dst_cache = &rdst->dst_cache;
1963 WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1967 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1968 vni = tunnel_id_to_key32(info->key.tun_id);
1969 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1970 if (remote_ip.sa.sa_family == AF_INET) {
1971 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1972 local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
1974 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1975 local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
1979 dst_cache = &info->dst_cache;
1982 if (vxlan_addr_any(dst)) {
1984 /* short-circuited back to local bridge */
1985 vxlan_encap_bypass(skb, vxlan, vxlan);
1991 old_iph = ip_hdr(skb);
1993 ttl = vxlan->cfg.ttl;
1994 if (!ttl && vxlan_addr_multicast(dst))
1997 tos = vxlan->cfg.tos;
1999 tos = ip_tunnel_get_dsfield(old_iph, skb);
2001 label = vxlan->cfg.label;
2002 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2003 vxlan->cfg.port_max, true);
2006 ttl = info->key.ttl;
2007 tos = info->key.tos;
2008 label = info->key.label;
2009 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2011 if (info->options_len)
2012 md = ip_tunnel_info_opts(info);
2014 md->gbp = skb->mark;
2017 if (dst->sa.sa_family == AF_INET) {
2018 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2022 sk = sock4->sock->sk;
2024 rt = vxlan_get_route(vxlan, skb,
2025 rdst ? rdst->remote_ifindex : 0, tos,
2026 dst->sin.sin_addr.s_addr,
2027 &src->sin.sin_addr.s_addr,
2030 netdev_dbg(dev, "no route to %pI4\n",
2031 &dst->sin.sin_addr.s_addr);
2032 dev->stats.tx_carrier_errors++;
2036 if (rt->dst.dev == dev) {
2037 netdev_dbg(dev, "circular route to %pI4\n",
2038 &dst->sin.sin_addr.s_addr);
2039 dev->stats.collisions++;
2043 /* Bypass encapsulation if the destination is local */
2044 if (!info && rt->rt_flags & RTCF_LOCAL &&
2045 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2046 struct vxlan_dev *dst_vxlan;
2049 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2050 dst->sa.sa_family, dst_port,
2054 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2059 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2060 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2063 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2064 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2065 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2066 vni, md, flags, udp_sum);
2070 udp_tunnel_xmit_skb(rt, sk, skb, src->sin.sin_addr.s_addr,
2071 dst->sin.sin_addr.s_addr, tos, ttl, df,
2072 src_port, dst_port, xnet, !udp_sum);
2073 #if IS_ENABLED(CONFIG_IPV6)
2075 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2076 struct dst_entry *ndst;
2081 sk = sock6->sock->sk;
2083 ndst = vxlan6_get_route(vxlan, skb,
2084 rdst ? rdst->remote_ifindex : 0, tos,
2085 label, &dst->sin6.sin6_addr,
2086 &src->sin6.sin6_addr,
2089 netdev_dbg(dev, "no route to %pI6\n",
2090 &dst->sin6.sin6_addr);
2091 dev->stats.tx_carrier_errors++;
2095 if (ndst->dev == dev) {
2096 netdev_dbg(dev, "circular route to %pI6\n",
2097 &dst->sin6.sin6_addr);
2099 dev->stats.collisions++;
2103 /* Bypass encapsulation if the destination is local */
2104 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2105 if (!info && rt6i_flags & RTF_LOCAL &&
2106 !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2107 struct vxlan_dev *dst_vxlan;
2110 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2111 dst->sa.sa_family, dst_port,
2115 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2120 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2122 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2123 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2124 skb_scrub_packet(skb, xnet);
2125 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2126 vni, md, flags, udp_sum);
2129 dev->stats.tx_errors++;
2132 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2133 &src->sin6.sin6_addr,
2134 &dst->sin6.sin6_addr, tos, ttl,
2135 label, src_port, dst_port, !udp_sum);
2142 dev->stats.tx_dropped++;
2146 /* skb is already freed. */
2151 dev->stats.tx_errors++;
2156 /* Transmit local packets over Vxlan
2158 * Outer IP header inherits ECN and DF from inner header.
2159 * Outer UDP destination is the VXLAN assigned port.
2160 * source port is based on hash of flow
2162 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2164 struct vxlan_dev *vxlan = netdev_priv(dev);
2165 const struct ip_tunnel_info *info;
2167 bool did_rsc = false;
2168 struct vxlan_rdst *rdst, *fdst = NULL;
2169 struct vxlan_fdb *f;
2171 info = skb_tunnel_info(skb);
2173 skb_reset_mac_header(skb);
2175 if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2176 if (info && info->mode & IP_TUNNEL_INFO_TX)
2177 vxlan_xmit_one(skb, dev, NULL, false);
2180 return NETDEV_TX_OK;
2183 if (vxlan->flags & VXLAN_F_PROXY) {
2185 if (ntohs(eth->h_proto) == ETH_P_ARP)
2186 return arp_reduce(dev, skb);
2187 #if IS_ENABLED(CONFIG_IPV6)
2188 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2189 pskb_may_pull(skb, sizeof(struct ipv6hdr)
2190 + sizeof(struct nd_msg)) &&
2191 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2194 msg = (struct nd_msg *)skb_transport_header(skb);
2195 if (msg->icmph.icmp6_code == 0 &&
2196 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2197 return neigh_reduce(dev, skb);
2203 f = vxlan_find_mac(vxlan, eth->h_dest);
2206 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2207 (ntohs(eth->h_proto) == ETH_P_IP ||
2208 ntohs(eth->h_proto) == ETH_P_IPV6)) {
2209 did_rsc = route_shortcircuit(dev, skb);
2211 f = vxlan_find_mac(vxlan, eth->h_dest);
2215 f = vxlan_find_mac(vxlan, all_zeros_mac);
2217 if ((vxlan->flags & VXLAN_F_L2MISS) &&
2218 !is_multicast_ether_addr(eth->h_dest))
2219 vxlan_fdb_miss(vxlan, eth->h_dest);
2221 dev->stats.tx_dropped++;
2223 return NETDEV_TX_OK;
2227 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2228 struct sk_buff *skb1;
2234 skb1 = skb_clone(skb, GFP_ATOMIC);
2236 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2240 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2243 return NETDEV_TX_OK;
2246 /* Walk the forwarding table and purge stale entries */
2247 static void vxlan_cleanup(unsigned long arg)
2249 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2250 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2253 if (!netif_running(vxlan->dev))
2256 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2257 struct hlist_node *p, *n;
2259 spin_lock_bh(&vxlan->hash_lock);
2260 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2262 = container_of(p, struct vxlan_fdb, hlist);
2263 unsigned long timeout;
2265 if (f->state & NUD_PERMANENT)
2268 timeout = f->used + vxlan->cfg.age_interval * HZ;
2269 if (time_before_eq(timeout, jiffies)) {
2270 netdev_dbg(vxlan->dev,
2271 "garbage collect %pM\n",
2273 f->state = NUD_STALE;
2274 vxlan_fdb_destroy(vxlan, f);
2275 } else if (time_before(timeout, next_timer))
2276 next_timer = timeout;
2278 spin_unlock_bh(&vxlan->hash_lock);
2281 mod_timer(&vxlan->age_timer, next_timer);
2284 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2286 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2287 __be32 vni = vxlan->default_dst.remote_vni;
2289 spin_lock(&vn->sock_lock);
2290 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2291 spin_unlock(&vn->sock_lock);
2294 /* Setup stats when device is created */
2295 static int vxlan_init(struct net_device *dev)
2297 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2304 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2306 struct vxlan_fdb *f;
2308 spin_lock_bh(&vxlan->hash_lock);
2309 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2311 vxlan_fdb_destroy(vxlan, f);
2312 spin_unlock_bh(&vxlan->hash_lock);
2315 static void vxlan_uninit(struct net_device *dev)
2317 struct vxlan_dev *vxlan = netdev_priv(dev);
2319 vxlan_fdb_delete_default(vxlan);
2321 free_percpu(dev->tstats);
2324 /* Start ageing timer and join group when device is brought up */
2325 static int vxlan_open(struct net_device *dev)
2327 struct vxlan_dev *vxlan = netdev_priv(dev);
2330 ret = vxlan_sock_add(vxlan);
2334 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2335 ret = vxlan_igmp_join(vxlan);
2336 if (ret == -EADDRINUSE)
2339 vxlan_sock_release(vxlan);
2344 if (vxlan->cfg.age_interval)
2345 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2350 /* Purge the forwarding table */
2351 static void vxlan_flush(struct vxlan_dev *vxlan)
2355 spin_lock_bh(&vxlan->hash_lock);
2356 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2357 struct hlist_node *p, *n;
2358 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2360 = container_of(p, struct vxlan_fdb, hlist);
2361 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2362 if (!is_zero_ether_addr(f->eth_addr))
2363 vxlan_fdb_destroy(vxlan, f);
2366 spin_unlock_bh(&vxlan->hash_lock);
2369 /* Cleanup timer and forwarding table on shutdown */
2370 static int vxlan_stop(struct net_device *dev)
2372 struct vxlan_dev *vxlan = netdev_priv(dev);
2373 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2376 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2377 !vxlan_group_used(vn, vxlan))
2378 ret = vxlan_igmp_leave(vxlan);
2380 del_timer_sync(&vxlan->age_timer);
2383 vxlan_sock_release(vxlan);
2388 /* Stub, nothing needs to be done. */
2389 static void vxlan_set_multicast_list(struct net_device *dev)
2393 static int __vxlan_change_mtu(struct net_device *dev,
2394 struct net_device *lowerdev,
2395 struct vxlan_rdst *dst, int new_mtu, bool strict)
2397 int max_mtu = IP_MAX_MTU;
2400 max_mtu = lowerdev->mtu;
2402 if (dst->remote_ip.sa.sa_family == AF_INET6)
2403 max_mtu -= VXLAN6_HEADROOM;
2405 max_mtu -= VXLAN_HEADROOM;
2410 if (new_mtu > max_mtu) {
2421 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2423 struct vxlan_dev *vxlan = netdev_priv(dev);
2424 struct vxlan_rdst *dst = &vxlan->default_dst;
2425 struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2426 dst->remote_ifindex);
2427 return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2430 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2432 struct vxlan_dev *vxlan = netdev_priv(dev);
2433 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2434 __be16 sport, dport;
2436 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2437 vxlan->cfg.port_max, true);
2438 dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2440 if (ip_tunnel_info_af(info) == AF_INET) {
2441 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2446 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2447 info->key.u.ipv4.dst,
2448 &info->key.u.ipv4.src, NULL, info);
2453 #if IS_ENABLED(CONFIG_IPV6)
2454 struct dst_entry *ndst;
2456 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2457 info->key.label, &info->key.u.ipv6.dst,
2458 &info->key.u.ipv6.src, NULL, info);
2460 return PTR_ERR(ndst);
2462 #else /* !CONFIG_IPV6 */
2463 return -EPFNOSUPPORT;
2466 info->key.tp_src = sport;
2467 info->key.tp_dst = dport;
2471 static const struct net_device_ops vxlan_netdev_ether_ops = {
2472 .ndo_init = vxlan_init,
2473 .ndo_uninit = vxlan_uninit,
2474 .ndo_open = vxlan_open,
2475 .ndo_stop = vxlan_stop,
2476 .ndo_start_xmit = vxlan_xmit,
2477 .ndo_get_stats64 = ip_tunnel_get_stats64,
2478 .ndo_set_rx_mode = vxlan_set_multicast_list,
2479 .ndo_change_mtu = vxlan_change_mtu,
2480 .ndo_validate_addr = eth_validate_addr,
2481 .ndo_set_mac_address = eth_mac_addr,
2482 .ndo_fdb_add = vxlan_fdb_add,
2483 .ndo_fdb_del = vxlan_fdb_delete,
2484 .ndo_fdb_dump = vxlan_fdb_dump,
2485 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2488 static const struct net_device_ops vxlan_netdev_raw_ops = {
2489 .ndo_init = vxlan_init,
2490 .ndo_uninit = vxlan_uninit,
2491 .ndo_open = vxlan_open,
2492 .ndo_stop = vxlan_stop,
2493 .ndo_start_xmit = vxlan_xmit,
2494 .ndo_get_stats64 = ip_tunnel_get_stats64,
2495 .ndo_change_mtu = vxlan_change_mtu,
2496 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2499 /* Info for udev, that this is a virtual tunnel endpoint */
2500 static struct device_type vxlan_type = {
2504 /* Calls the ndo_udp_tunnel_add of the caller in order to
2505 * supply the listening VXLAN udp ports. Callers are expected
2506 * to implement the ndo_udp_tunnel_add.
2508 static void vxlan_push_rx_ports(struct net_device *dev)
2510 struct vxlan_sock *vs;
2511 struct net *net = dev_net(dev);
2512 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2515 spin_lock(&vn->sock_lock);
2516 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2517 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2518 udp_tunnel_push_rx_port(dev, vs->sock,
2519 (vs->flags & VXLAN_F_GPE) ?
2520 UDP_TUNNEL_TYPE_VXLAN_GPE :
2521 UDP_TUNNEL_TYPE_VXLAN);
2523 spin_unlock(&vn->sock_lock);
2526 /* Initialize the device structure. */
2527 static void vxlan_setup(struct net_device *dev)
2529 struct vxlan_dev *vxlan = netdev_priv(dev);
2532 eth_hw_addr_random(dev);
2535 dev->destructor = free_netdev;
2536 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2538 dev->features |= NETIF_F_LLTX;
2539 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2540 dev->features |= NETIF_F_RXCSUM;
2541 dev->features |= NETIF_F_GSO_SOFTWARE;
2543 dev->vlan_features = dev->features;
2544 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2545 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2546 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2547 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2548 netif_keep_dst(dev);
2549 dev->priv_flags |= IFF_NO_QUEUE;
2551 INIT_LIST_HEAD(&vxlan->next);
2552 spin_lock_init(&vxlan->hash_lock);
2554 init_timer_deferrable(&vxlan->age_timer);
2555 vxlan->age_timer.function = vxlan_cleanup;
2556 vxlan->age_timer.data = (unsigned long) vxlan;
2558 vxlan->cfg.dst_port = htons(vxlan_port);
2562 gro_cells_init(&vxlan->gro_cells, dev);
2564 for (h = 0; h < FDB_HASH_SIZE; ++h)
2565 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2568 static void vxlan_ether_setup(struct net_device *dev)
2570 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2571 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2572 dev->netdev_ops = &vxlan_netdev_ether_ops;
2575 static void vxlan_raw_setup(struct net_device *dev)
2577 dev->header_ops = NULL;
2578 dev->type = ARPHRD_NONE;
2579 dev->hard_header_len = 0;
2581 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2582 dev->netdev_ops = &vxlan_netdev_raw_ops;
2585 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2586 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2587 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2588 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2589 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2590 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2591 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2592 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2593 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2594 [IFLA_VXLAN_LABEL] = { .type = NLA_U32 },
2595 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2596 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2597 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2598 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2599 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2600 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2601 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2602 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2603 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 },
2604 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2605 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 },
2606 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
2607 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
2608 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2609 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2610 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, },
2611 [IFLA_VXLAN_GPE] = { .type = NLA_FLAG, },
2612 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG },
2615 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2617 if (tb[IFLA_ADDRESS]) {
2618 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2619 pr_debug("invalid link address (not ethernet)\n");
2623 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2624 pr_debug("invalid all zero ethernet address\n");
2625 return -EADDRNOTAVAIL;
2632 if (data[IFLA_VXLAN_ID]) {
2633 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2634 if (id >= VXLAN_VID_MASK)
2638 if (data[IFLA_VXLAN_PORT_RANGE]) {
2639 const struct ifla_vxlan_port_range *p
2640 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2642 if (ntohs(p->high) < ntohs(p->low)) {
2643 pr_debug("port range %u .. %u not valid\n",
2644 ntohs(p->low), ntohs(p->high));
2652 static void vxlan_get_drvinfo(struct net_device *netdev,
2653 struct ethtool_drvinfo *drvinfo)
2655 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2656 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2659 static const struct ethtool_ops vxlan_ethtool_ops = {
2660 .get_drvinfo = vxlan_get_drvinfo,
2661 .get_link = ethtool_op_get_link,
2664 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2665 __be16 port, u32 flags)
2667 struct socket *sock;
2668 struct udp_port_cfg udp_conf;
2671 memset(&udp_conf, 0, sizeof(udp_conf));
2674 udp_conf.family = AF_INET6;
2675 udp_conf.use_udp6_rx_checksums =
2676 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2677 udp_conf.ipv6_v6only = 1;
2679 udp_conf.family = AF_INET;
2682 udp_conf.local_udp_port = port;
2684 /* Open UDP socket */
2685 err = udp_sock_create(net, &udp_conf, &sock);
2687 return ERR_PTR(err);
2692 /* Create new listen socket if needed */
2693 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2694 __be16 port, u32 flags)
2696 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2697 struct vxlan_sock *vs;
2698 struct socket *sock;
2700 struct udp_tunnel_sock_cfg tunnel_cfg;
2702 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2704 return ERR_PTR(-ENOMEM);
2706 for (h = 0; h < VNI_HASH_SIZE; ++h)
2707 INIT_HLIST_HEAD(&vs->vni_list[h]);
2709 sock = vxlan_create_sock(net, ipv6, port, flags);
2711 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2714 return ERR_CAST(sock);
2718 atomic_set(&vs->refcnt, 1);
2719 vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2721 spin_lock(&vn->sock_lock);
2722 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2723 udp_tunnel_notify_add_rx_port(sock,
2724 (vs->flags & VXLAN_F_GPE) ?
2725 UDP_TUNNEL_TYPE_VXLAN_GPE :
2726 UDP_TUNNEL_TYPE_VXLAN);
2727 spin_unlock(&vn->sock_lock);
2729 /* Mark socket as an encapsulation socket. */
2730 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2731 tunnel_cfg.sk_user_data = vs;
2732 tunnel_cfg.encap_type = 1;
2733 tunnel_cfg.encap_rcv = vxlan_rcv;
2734 tunnel_cfg.encap_destroy = NULL;
2735 tunnel_cfg.gro_receive = vxlan_gro_receive;
2736 tunnel_cfg.gro_complete = vxlan_gro_complete;
2738 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2743 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2745 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2746 struct vxlan_sock *vs = NULL;
2748 if (!vxlan->cfg.no_share) {
2749 spin_lock(&vn->sock_lock);
2750 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2751 vxlan->cfg.dst_port, vxlan->flags);
2752 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2753 spin_unlock(&vn->sock_lock);
2756 spin_unlock(&vn->sock_lock);
2759 vs = vxlan_socket_create(vxlan->net, ipv6,
2760 vxlan->cfg.dst_port, vxlan->flags);
2763 #if IS_ENABLED(CONFIG_IPV6)
2765 rcu_assign_pointer(vxlan->vn6_sock, vs);
2768 rcu_assign_pointer(vxlan->vn4_sock, vs);
2769 vxlan_vs_add_dev(vs, vxlan);
2773 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2775 bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2776 bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2779 RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2780 #if IS_ENABLED(CONFIG_IPV6)
2781 RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2782 if (ipv6 || metadata)
2783 ret = __vxlan_sock_add(vxlan, true);
2785 if (!ret && (!ipv6 || metadata))
2786 ret = __vxlan_sock_add(vxlan, false);
2788 vxlan_sock_release(vxlan);
2792 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2793 struct vxlan_config *conf)
2795 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2796 struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2797 struct vxlan_rdst *dst = &vxlan->default_dst;
2798 unsigned short needed_headroom = ETH_HLEN;
2800 bool use_ipv6 = false;
2801 __be16 default_port = vxlan->cfg.dst_port;
2802 struct net_device *lowerdev = NULL;
2804 if (conf->flags & VXLAN_F_GPE) {
2805 /* For now, allow GPE only together with COLLECT_METADATA.
2806 * This can be relaxed later; in such case, the other side
2807 * of the PtP link will have to be provided.
2809 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2810 !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2811 pr_info("unsupported combination of extensions\n");
2815 vxlan_raw_setup(dev);
2817 vxlan_ether_setup(dev);
2820 vxlan->net = src_net;
2822 dst->remote_vni = conf->vni;
2824 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2826 /* Unless IPv6 is explicitly requested, assume IPv4 */
2827 if (!dst->remote_ip.sa.sa_family)
2828 dst->remote_ip.sa.sa_family = AF_INET;
2830 if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2831 vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2832 if (!IS_ENABLED(CONFIG_IPV6))
2833 return -EPFNOSUPPORT;
2835 vxlan->flags |= VXLAN_F_IPV6;
2838 if (conf->label && !use_ipv6) {
2839 pr_info("label only supported in use with IPv6\n");
2843 if (conf->remote_ifindex) {
2844 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2845 dst->remote_ifindex = conf->remote_ifindex;
2848 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2852 #if IS_ENABLED(CONFIG_IPV6)
2854 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2855 if (idev && idev->cnf.disable_ipv6) {
2856 pr_info("IPv6 is disabled via sysctl\n");
2863 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2865 needed_headroom = lowerdev->hard_header_len;
2866 } else if (vxlan_addr_multicast(&dst->remote_ip)) {
2867 pr_info("multicast destination requires interface to be specified\n");
2872 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2877 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2878 needed_headroom += VXLAN6_HEADROOM;
2880 needed_headroom += VXLAN_HEADROOM;
2881 dev->needed_headroom = needed_headroom;
2883 memcpy(&vxlan->cfg, conf, sizeof(*conf));
2884 if (!vxlan->cfg.dst_port) {
2885 if (conf->flags & VXLAN_F_GPE)
2886 vxlan->cfg.dst_port = 4790; /* IANA assigned VXLAN-GPE port */
2888 vxlan->cfg.dst_port = default_port;
2890 vxlan->flags |= conf->flags;
2892 if (!vxlan->cfg.age_interval)
2893 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2895 list_for_each_entry(tmp, &vn->vxlan_list, next) {
2896 if (tmp->cfg.vni == conf->vni &&
2897 (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2898 tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2899 tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2900 (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2901 (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2902 pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2907 dev->ethtool_ops = &vxlan_ethtool_ops;
2909 /* create an fdb entry for a valid default destination */
2910 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2911 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2912 &vxlan->default_dst.remote_ip,
2913 NUD_REACHABLE|NUD_PERMANENT,
2914 NLM_F_EXCL|NLM_F_CREATE,
2915 vxlan->cfg.dst_port,
2916 vxlan->default_dst.remote_vni,
2917 vxlan->default_dst.remote_ifindex,
2923 err = register_netdevice(dev);
2925 vxlan_fdb_delete_default(vxlan);
2929 list_add(&vxlan->next, &vn->vxlan_list);
2934 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2935 struct nlattr *tb[], struct nlattr *data[])
2937 struct vxlan_config conf;
2939 memset(&conf, 0, sizeof(conf));
2941 if (data[IFLA_VXLAN_ID])
2942 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2944 if (data[IFLA_VXLAN_GROUP]) {
2945 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2946 } else if (data[IFLA_VXLAN_GROUP6]) {
2947 if (!IS_ENABLED(CONFIG_IPV6))
2948 return -EPFNOSUPPORT;
2950 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2951 conf.remote_ip.sa.sa_family = AF_INET6;
2954 if (data[IFLA_VXLAN_LOCAL]) {
2955 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2956 conf.saddr.sa.sa_family = AF_INET;
2957 } else if (data[IFLA_VXLAN_LOCAL6]) {
2958 if (!IS_ENABLED(CONFIG_IPV6))
2959 return -EPFNOSUPPORT;
2961 /* TODO: respect scope id */
2962 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2963 conf.saddr.sa.sa_family = AF_INET6;
2966 if (data[IFLA_VXLAN_LINK])
2967 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2969 if (data[IFLA_VXLAN_TOS])
2970 conf.tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2972 if (data[IFLA_VXLAN_TTL])
2973 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2975 if (data[IFLA_VXLAN_LABEL])
2976 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
2977 IPV6_FLOWLABEL_MASK;
2979 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2980 conf.flags |= VXLAN_F_LEARN;
2982 if (data[IFLA_VXLAN_AGEING])
2983 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2985 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2986 conf.flags |= VXLAN_F_PROXY;
2988 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2989 conf.flags |= VXLAN_F_RSC;
2991 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2992 conf.flags |= VXLAN_F_L2MISS;
2994 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2995 conf.flags |= VXLAN_F_L3MISS;
2997 if (data[IFLA_VXLAN_LIMIT])
2998 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3000 if (data[IFLA_VXLAN_COLLECT_METADATA] &&
3001 nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3002 conf.flags |= VXLAN_F_COLLECT_METADATA;
3004 if (data[IFLA_VXLAN_PORT_RANGE]) {
3005 const struct ifla_vxlan_port_range *p
3006 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3007 conf.port_min = ntohs(p->low);
3008 conf.port_max = ntohs(p->high);
3011 if (data[IFLA_VXLAN_PORT])
3012 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3014 if (data[IFLA_VXLAN_UDP_CSUM] &&
3015 !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3016 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3018 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3019 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3020 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3022 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3023 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3024 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3026 if (data[IFLA_VXLAN_REMCSUM_TX] &&
3027 nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3028 conf.flags |= VXLAN_F_REMCSUM_TX;
3030 if (data[IFLA_VXLAN_REMCSUM_RX] &&
3031 nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3032 conf.flags |= VXLAN_F_REMCSUM_RX;
3034 if (data[IFLA_VXLAN_GBP])
3035 conf.flags |= VXLAN_F_GBP;
3037 if (data[IFLA_VXLAN_GPE])
3038 conf.flags |= VXLAN_F_GPE;
3040 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3041 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3044 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3046 return vxlan_dev_configure(src_net, dev, &conf);
3049 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3051 struct vxlan_dev *vxlan = netdev_priv(dev);
3052 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3054 spin_lock(&vn->sock_lock);
3055 if (!hlist_unhashed(&vxlan->hlist))
3056 hlist_del_rcu(&vxlan->hlist);
3057 spin_unlock(&vn->sock_lock);
3059 gro_cells_destroy(&vxlan->gro_cells);
3060 list_del(&vxlan->next);
3061 unregister_netdevice_queue(dev, head);
3064 static size_t vxlan_get_size(const struct net_device *dev)
3067 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
3068 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3069 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3070 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3071 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
3072 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
3073 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3074 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
3075 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
3076 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
3077 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
3078 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
3079 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */
3080 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3081 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3082 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3083 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3084 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3085 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3086 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3087 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3088 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3092 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3094 const struct vxlan_dev *vxlan = netdev_priv(dev);
3095 const struct vxlan_rdst *dst = &vxlan->default_dst;
3096 struct ifla_vxlan_port_range ports = {
3097 .low = htons(vxlan->cfg.port_min),
3098 .high = htons(vxlan->cfg.port_max),
3101 if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3102 goto nla_put_failure;
3104 if (!vxlan_addr_any(&dst->remote_ip)) {
3105 if (dst->remote_ip.sa.sa_family == AF_INET) {
3106 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3107 dst->remote_ip.sin.sin_addr.s_addr))
3108 goto nla_put_failure;
3109 #if IS_ENABLED(CONFIG_IPV6)
3111 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3112 &dst->remote_ip.sin6.sin6_addr))
3113 goto nla_put_failure;
3118 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3119 goto nla_put_failure;
3121 if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3122 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3123 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3124 vxlan->cfg.saddr.sin.sin_addr.s_addr))
3125 goto nla_put_failure;
3126 #if IS_ENABLED(CONFIG_IPV6)
3128 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3129 &vxlan->cfg.saddr.sin6.sin6_addr))
3130 goto nla_put_failure;
3135 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3136 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3137 nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3138 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3139 !!(vxlan->flags & VXLAN_F_LEARN)) ||
3140 nla_put_u8(skb, IFLA_VXLAN_PROXY,
3141 !!(vxlan->flags & VXLAN_F_PROXY)) ||
3142 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3143 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3144 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3145 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3146 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3147 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3148 !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3149 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3150 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3151 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3152 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3153 !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3154 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3155 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3156 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3157 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3158 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3159 !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3160 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3161 !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3162 goto nla_put_failure;
3164 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3165 goto nla_put_failure;
3167 if (vxlan->flags & VXLAN_F_GBP &&
3168 nla_put_flag(skb, IFLA_VXLAN_GBP))
3169 goto nla_put_failure;
3171 if (vxlan->flags & VXLAN_F_GPE &&
3172 nla_put_flag(skb, IFLA_VXLAN_GPE))
3173 goto nla_put_failure;
3175 if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3176 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3177 goto nla_put_failure;
3185 static struct net *vxlan_get_link_net(const struct net_device *dev)
3187 struct vxlan_dev *vxlan = netdev_priv(dev);
3192 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3194 .maxtype = IFLA_VXLAN_MAX,
3195 .policy = vxlan_policy,
3196 .priv_size = sizeof(struct vxlan_dev),
3197 .setup = vxlan_setup,
3198 .validate = vxlan_validate,
3199 .newlink = vxlan_newlink,
3200 .dellink = vxlan_dellink,
3201 .get_size = vxlan_get_size,
3202 .fill_info = vxlan_fill_info,
3203 .get_link_net = vxlan_get_link_net,
3206 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3207 u8 name_assign_type,
3208 struct vxlan_config *conf)
3210 struct nlattr *tb[IFLA_MAX + 1];
3211 struct net_device *dev;
3214 memset(&tb, 0, sizeof(tb));
3216 dev = rtnl_create_link(net, name, name_assign_type,
3217 &vxlan_link_ops, tb);
3221 err = vxlan_dev_configure(net, dev, conf);
3224 return ERR_PTR(err);
3227 err = rtnl_configure_link(dev, NULL);
3229 LIST_HEAD(list_kill);
3231 vxlan_dellink(dev, &list_kill);
3232 unregister_netdevice_many(&list_kill);
3233 return ERR_PTR(err);
3238 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3240 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3241 struct net_device *dev)
3243 struct vxlan_dev *vxlan, *next;
3244 LIST_HEAD(list_kill);
3246 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3247 struct vxlan_rdst *dst = &vxlan->default_dst;
3249 /* In case we created vxlan device with carrier
3250 * and we loose the carrier due to module unload
3251 * we also need to remove vxlan device. In other
3252 * cases, it's not necessary and remote_ifindex
3253 * is 0 here, so no matches.
3255 if (dst->remote_ifindex == dev->ifindex)
3256 vxlan_dellink(vxlan->dev, &list_kill);
3259 unregister_netdevice_many(&list_kill);
3262 static int vxlan_netdevice_event(struct notifier_block *unused,
3263 unsigned long event, void *ptr)
3265 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3266 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3268 if (event == NETDEV_UNREGISTER)
3269 vxlan_handle_lowerdev_unregister(vn, dev);
3270 else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3271 vxlan_push_rx_ports(dev);
3276 static struct notifier_block vxlan_notifier_block __read_mostly = {
3277 .notifier_call = vxlan_netdevice_event,
3280 static __net_init int vxlan_init_net(struct net *net)
3282 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3285 INIT_LIST_HEAD(&vn->vxlan_list);
3286 spin_lock_init(&vn->sock_lock);
3288 for (h = 0; h < PORT_HASH_SIZE; ++h)
3289 INIT_HLIST_HEAD(&vn->sock_list[h]);
3294 static void __net_exit vxlan_exit_net(struct net *net)
3296 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3297 struct vxlan_dev *vxlan, *next;
3298 struct net_device *dev, *aux;
3302 for_each_netdev_safe(net, dev, aux)
3303 if (dev->rtnl_link_ops == &vxlan_link_ops)
3304 unregister_netdevice_queue(dev, &list);
3306 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3307 /* If vxlan->dev is in the same netns, it has already been added
3308 * to the list by the previous loop.
3310 if (!net_eq(dev_net(vxlan->dev), net)) {
3311 gro_cells_destroy(&vxlan->gro_cells);
3312 unregister_netdevice_queue(vxlan->dev, &list);
3316 unregister_netdevice_many(&list);
3320 static struct pernet_operations vxlan_net_ops = {
3321 .init = vxlan_init_net,
3322 .exit = vxlan_exit_net,
3323 .id = &vxlan_net_id,
3324 .size = sizeof(struct vxlan_net),
3327 static int __init vxlan_init_module(void)
3331 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3333 rc = register_pernet_subsys(&vxlan_net_ops);
3337 rc = register_netdevice_notifier(&vxlan_notifier_block);
3341 rc = rtnl_link_register(&vxlan_link_ops);
3347 unregister_netdevice_notifier(&vxlan_notifier_block);
3349 unregister_pernet_subsys(&vxlan_net_ops);
3353 late_initcall(vxlan_init_module);
3355 static void __exit vxlan_cleanup_module(void)
3357 rtnl_link_unregister(&vxlan_link_ops);
3358 unregister_netdevice_notifier(&vxlan_notifier_block);
3359 unregister_pernet_subsys(&vxlan_net_ops);
3360 /* rcu_barrier() is called by netns */
3362 module_exit(vxlan_cleanup_module);
3364 MODULE_LICENSE("GPL");
3365 MODULE_VERSION(VXLAN_VERSION);
3366 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3367 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3368 MODULE_ALIAS_RTNL_LINK("vxlan");