Loading...
1/*
2 * VXLAN: Virtual eXtensible Local Area Network
3 *
4 * Copyright (c) 2012-2013 Vyatta Inc.
5 *
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.
9 */
10
11#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12
13#include <linux/kernel.h>
14#include <linux/types.h>
15#include <linux/module.h>
16#include <linux/errno.h>
17#include <linux/slab.h>
18#include <linux/skbuff.h>
19#include <linux/rculist.h>
20#include <linux/netdevice.h>
21#include <linux/in.h>
22#include <linux/ip.h>
23#include <linux/udp.h>
24#include <linux/igmp.h>
25#include <linux/etherdevice.h>
26#include <linux/if_ether.h>
27#include <linux/if_vlan.h>
28#include <linux/hash.h>
29#include <linux/ethtool.h>
30#include <net/arp.h>
31#include <net/ndisc.h>
32#include <net/ip.h>
33#include <net/ip_tunnels.h>
34#include <net/icmp.h>
35#include <net/udp.h>
36#include <net/udp_tunnel.h>
37#include <net/rtnetlink.h>
38#include <net/route.h>
39#include <net/dsfield.h>
40#include <net/inet_ecn.h>
41#include <net/net_namespace.h>
42#include <net/netns/generic.h>
43#include <net/vxlan.h>
44#include <net/protocol.h>
45
46#if IS_ENABLED(CONFIG_IPV6)
47#include <net/ipv6.h>
48#include <net/addrconf.h>
49#include <net/ip6_tunnel.h>
50#include <net/ip6_checksum.h>
51#endif
52#include <net/dst_metadata.h>
53
54#define VXLAN_VERSION "0.1"
55
56#define PORT_HASH_BITS 8
57#define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
58#define FDB_AGE_DEFAULT 300 /* 5 min */
59#define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
60
61/* UDP port for VXLAN traffic.
62 * The IANA assigned port is 4789, but the Linux default is 8472
63 * for compatibility with early adopters.
64 */
65static unsigned short vxlan_port __read_mostly = 8472;
66module_param_named(udp_port, vxlan_port, ushort, 0444);
67MODULE_PARM_DESC(udp_port, "Destination UDP port");
68
69static bool log_ecn_error = true;
70module_param(log_ecn_error, bool, 0644);
71MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
72
73static int vxlan_net_id;
74static struct rtnl_link_ops vxlan_link_ops;
75
76static const u8 all_zeros_mac[ETH_ALEN + 2];
77
78static int vxlan_sock_add(struct vxlan_dev *vxlan);
79
80/* per-network namespace private data for this module */
81struct vxlan_net {
82 struct list_head vxlan_list;
83 struct hlist_head sock_list[PORT_HASH_SIZE];
84 spinlock_t sock_lock;
85};
86
87/* Forwarding table entry */
88struct vxlan_fdb {
89 struct hlist_node hlist; /* linked list of entries */
90 struct rcu_head rcu;
91 unsigned long updated; /* jiffies */
92 unsigned long used;
93 struct list_head remotes;
94 u8 eth_addr[ETH_ALEN];
95 u16 state; /* see ndm_state */
96 u8 flags; /* see ndm_flags */
97};
98
99/* salt for hash table */
100static u32 vxlan_salt __read_mostly;
101static struct workqueue_struct *vxlan_wq;
102
103static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
104{
105 return vs->flags & VXLAN_F_COLLECT_METADATA ||
106 ip_tunnel_collect_metadata();
107}
108
109#if IS_ENABLED(CONFIG_IPV6)
110static inline
111bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
112{
113 if (a->sa.sa_family != b->sa.sa_family)
114 return false;
115 if (a->sa.sa_family == AF_INET6)
116 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
117 else
118 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
119}
120
121static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
122{
123 if (ipa->sa.sa_family == AF_INET6)
124 return ipv6_addr_any(&ipa->sin6.sin6_addr);
125 else
126 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
127}
128
129static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
130{
131 if (ipa->sa.sa_family == AF_INET6)
132 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
133 else
134 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
135}
136
137static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
138{
139 if (nla_len(nla) >= sizeof(struct in6_addr)) {
140 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
141 ip->sa.sa_family = AF_INET6;
142 return 0;
143 } else if (nla_len(nla) >= sizeof(__be32)) {
144 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
145 ip->sa.sa_family = AF_INET;
146 return 0;
147 } else {
148 return -EAFNOSUPPORT;
149 }
150}
151
152static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
153 const union vxlan_addr *ip)
154{
155 if (ip->sa.sa_family == AF_INET6)
156 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
157 else
158 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
159}
160
161#else /* !CONFIG_IPV6 */
162
163static inline
164bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
165{
166 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
167}
168
169static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
170{
171 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
172}
173
174static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
175{
176 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
177}
178
179static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
180{
181 if (nla_len(nla) >= sizeof(struct in6_addr)) {
182 return -EAFNOSUPPORT;
183 } else if (nla_len(nla) >= sizeof(__be32)) {
184 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
185 ip->sa.sa_family = AF_INET;
186 return 0;
187 } else {
188 return -EAFNOSUPPORT;
189 }
190}
191
192static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
193 const union vxlan_addr *ip)
194{
195 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
196}
197#endif
198
199/* Virtual Network hash table head */
200static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
201{
202 return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
203}
204
205/* Socket hash table head */
206static inline struct hlist_head *vs_head(struct net *net, __be16 port)
207{
208 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
209
210 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
211}
212
213/* First remote destination for a forwarding entry.
214 * Guaranteed to be non-NULL because remotes are never deleted.
215 */
216static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
217{
218 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
219}
220
221static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
222{
223 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
224}
225
226/* Find VXLAN socket based on network namespace, address family and UDP port
227 * and enabled unshareable flags.
228 */
229static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
230 __be16 port, u32 flags)
231{
232 struct vxlan_sock *vs;
233
234 flags &= VXLAN_F_RCV_FLAGS;
235
236 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
237 if (inet_sk(vs->sock->sk)->inet_sport == port &&
238 vxlan_get_sk_family(vs) == family &&
239 vs->flags == flags)
240 return vs;
241 }
242 return NULL;
243}
244
245static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
246{
247 struct vxlan_dev *vxlan;
248
249 /* For flow based devices, map all packets to VNI 0 */
250 if (vs->flags & VXLAN_F_COLLECT_METADATA)
251 vni = 0;
252
253 hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
254 if (vxlan->default_dst.remote_vni == vni)
255 return vxlan;
256 }
257
258 return NULL;
259}
260
261/* Look up VNI in a per net namespace table */
262static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
263 sa_family_t family, __be16 port,
264 u32 flags)
265{
266 struct vxlan_sock *vs;
267
268 vs = vxlan_find_sock(net, family, port, flags);
269 if (!vs)
270 return NULL;
271
272 return vxlan_vs_find_vni(vs, vni);
273}
274
275/* Fill in neighbour message in skbuff. */
276static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
277 const struct vxlan_fdb *fdb,
278 u32 portid, u32 seq, int type, unsigned int flags,
279 const struct vxlan_rdst *rdst)
280{
281 unsigned long now = jiffies;
282 struct nda_cacheinfo ci;
283 struct nlmsghdr *nlh;
284 struct ndmsg *ndm;
285 bool send_ip, send_eth;
286
287 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
288 if (nlh == NULL)
289 return -EMSGSIZE;
290
291 ndm = nlmsg_data(nlh);
292 memset(ndm, 0, sizeof(*ndm));
293
294 send_eth = send_ip = true;
295
296 if (type == RTM_GETNEIGH) {
297 ndm->ndm_family = AF_INET;
298 send_ip = !vxlan_addr_any(&rdst->remote_ip);
299 send_eth = !is_zero_ether_addr(fdb->eth_addr);
300 } else
301 ndm->ndm_family = AF_BRIDGE;
302 ndm->ndm_state = fdb->state;
303 ndm->ndm_ifindex = vxlan->dev->ifindex;
304 ndm->ndm_flags = fdb->flags;
305 ndm->ndm_type = RTN_UNICAST;
306
307 if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
308 nla_put_s32(skb, NDA_LINK_NETNSID,
309 peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
310 goto nla_put_failure;
311
312 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
313 goto nla_put_failure;
314
315 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
316 goto nla_put_failure;
317
318 if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
319 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
320 goto nla_put_failure;
321 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
322 nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
323 goto nla_put_failure;
324 if (rdst->remote_ifindex &&
325 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
326 goto nla_put_failure;
327
328 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
329 ci.ndm_confirmed = 0;
330 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
331 ci.ndm_refcnt = 0;
332
333 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
334 goto nla_put_failure;
335
336 nlmsg_end(skb, nlh);
337 return 0;
338
339nla_put_failure:
340 nlmsg_cancel(skb, nlh);
341 return -EMSGSIZE;
342}
343
344static inline size_t vxlan_nlmsg_size(void)
345{
346 return NLMSG_ALIGN(sizeof(struct ndmsg))
347 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
348 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
349 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
350 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
351 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
352 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
353 + nla_total_size(sizeof(struct nda_cacheinfo));
354}
355
356static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
357 struct vxlan_rdst *rd, int type)
358{
359 struct net *net = dev_net(vxlan->dev);
360 struct sk_buff *skb;
361 int err = -ENOBUFS;
362
363 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
364 if (skb == NULL)
365 goto errout;
366
367 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
368 if (err < 0) {
369 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
370 WARN_ON(err == -EMSGSIZE);
371 kfree_skb(skb);
372 goto errout;
373 }
374
375 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
376 return;
377errout:
378 if (err < 0)
379 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
380}
381
382static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
383{
384 struct vxlan_dev *vxlan = netdev_priv(dev);
385 struct vxlan_fdb f = {
386 .state = NUD_STALE,
387 };
388 struct vxlan_rdst remote = {
389 .remote_ip = *ipa, /* goes to NDA_DST */
390 .remote_vni = cpu_to_be32(VXLAN_N_VID),
391 };
392
393 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
394}
395
396static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
397{
398 struct vxlan_fdb f = {
399 .state = NUD_STALE,
400 };
401 struct vxlan_rdst remote = { };
402
403 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
404
405 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
406}
407
408/* Hash Ethernet address */
409static u32 eth_hash(const unsigned char *addr)
410{
411 u64 value = get_unaligned((u64 *)addr);
412
413 /* only want 6 bytes */
414#ifdef __BIG_ENDIAN
415 value >>= 16;
416#else
417 value <<= 16;
418#endif
419 return hash_64(value, FDB_HASH_BITS);
420}
421
422/* Hash chain to use given mac address */
423static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
424 const u8 *mac)
425{
426 return &vxlan->fdb_head[eth_hash(mac)];
427}
428
429/* Look up Ethernet address in forwarding table */
430static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
431 const u8 *mac)
432{
433 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
434 struct vxlan_fdb *f;
435
436 hlist_for_each_entry_rcu(f, head, hlist) {
437 if (ether_addr_equal(mac, f->eth_addr))
438 return f;
439 }
440
441 return NULL;
442}
443
444static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
445 const u8 *mac)
446{
447 struct vxlan_fdb *f;
448
449 f = __vxlan_find_mac(vxlan, mac);
450 if (f)
451 f->used = jiffies;
452
453 return f;
454}
455
456/* caller should hold vxlan->hash_lock */
457static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
458 union vxlan_addr *ip, __be16 port,
459 __be32 vni, __u32 ifindex)
460{
461 struct vxlan_rdst *rd;
462
463 list_for_each_entry(rd, &f->remotes, list) {
464 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
465 rd->remote_port == port &&
466 rd->remote_vni == vni &&
467 rd->remote_ifindex == ifindex)
468 return rd;
469 }
470
471 return NULL;
472}
473
474/* Replace destination of unicast mac */
475static int vxlan_fdb_replace(struct vxlan_fdb *f,
476 union vxlan_addr *ip, __be16 port, __be32 vni,
477 __u32 ifindex)
478{
479 struct vxlan_rdst *rd;
480
481 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
482 if (rd)
483 return 0;
484
485 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
486 if (!rd)
487 return 0;
488
489 dst_cache_reset(&rd->dst_cache);
490 rd->remote_ip = *ip;
491 rd->remote_port = port;
492 rd->remote_vni = vni;
493 rd->remote_ifindex = ifindex;
494 return 1;
495}
496
497/* Add/update destinations for multicast */
498static int vxlan_fdb_append(struct vxlan_fdb *f,
499 union vxlan_addr *ip, __be16 port, __be32 vni,
500 __u32 ifindex, struct vxlan_rdst **rdp)
501{
502 struct vxlan_rdst *rd;
503
504 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
505 if (rd)
506 return 0;
507
508 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
509 if (rd == NULL)
510 return -ENOBUFS;
511
512 if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
513 kfree(rd);
514 return -ENOBUFS;
515 }
516
517 rd->remote_ip = *ip;
518 rd->remote_port = port;
519 rd->remote_vni = vni;
520 rd->remote_ifindex = ifindex;
521
522 list_add_tail_rcu(&rd->list, &f->remotes);
523
524 *rdp = rd;
525 return 1;
526}
527
528static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
529 unsigned int off,
530 struct vxlanhdr *vh, size_t hdrlen,
531 __be32 vni_field,
532 struct gro_remcsum *grc,
533 bool nopartial)
534{
535 size_t start, offset;
536
537 if (skb->remcsum_offload)
538 return vh;
539
540 if (!NAPI_GRO_CB(skb)->csum_valid)
541 return NULL;
542
543 start = vxlan_rco_start(vni_field);
544 offset = start + vxlan_rco_offset(vni_field);
545
546 vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
547 start, offset, grc, nopartial);
548
549 skb->remcsum_offload = 1;
550
551 return vh;
552}
553
554static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
555 struct sk_buff *skb,
556 struct udp_offload *uoff)
557{
558 struct sk_buff *p, **pp = NULL;
559 struct vxlanhdr *vh, *vh2;
560 unsigned int hlen, off_vx;
561 int flush = 1;
562 struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
563 udp_offloads);
564 __be32 flags;
565 struct gro_remcsum grc;
566
567 skb_gro_remcsum_init(&grc);
568
569 off_vx = skb_gro_offset(skb);
570 hlen = off_vx + sizeof(*vh);
571 vh = skb_gro_header_fast(skb, off_vx);
572 if (skb_gro_header_hard(skb, hlen)) {
573 vh = skb_gro_header_slow(skb, hlen, off_vx);
574 if (unlikely(!vh))
575 goto out;
576 }
577
578 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
579
580 flags = vh->vx_flags;
581
582 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
583 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
584 vh->vx_vni, &grc,
585 !!(vs->flags &
586 VXLAN_F_REMCSUM_NOPARTIAL));
587
588 if (!vh)
589 goto out;
590 }
591
592 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
593
594 for (p = *head; p; p = p->next) {
595 if (!NAPI_GRO_CB(p)->same_flow)
596 continue;
597
598 vh2 = (struct vxlanhdr *)(p->data + off_vx);
599 if (vh->vx_flags != vh2->vx_flags ||
600 vh->vx_vni != vh2->vx_vni) {
601 NAPI_GRO_CB(p)->same_flow = 0;
602 continue;
603 }
604 }
605
606 pp = eth_gro_receive(head, skb);
607 flush = 0;
608
609out:
610 skb_gro_remcsum_cleanup(skb, &grc);
611 NAPI_GRO_CB(skb)->flush |= flush;
612
613 return pp;
614}
615
616static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
617 struct udp_offload *uoff)
618{
619 /* Sets 'skb->inner_mac_header' since we are always called with
620 * 'skb->encapsulation' set.
621 */
622 return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
623}
624
625/* Notify netdevs that UDP port started listening */
626static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
627{
628 struct net_device *dev;
629 struct sock *sk = vs->sock->sk;
630 struct net *net = sock_net(sk);
631 sa_family_t sa_family = vxlan_get_sk_family(vs);
632 __be16 port = inet_sk(sk)->inet_sport;
633 int err;
634
635 if (sa_family == AF_INET) {
636 err = udp_add_offload(net, &vs->udp_offloads);
637 if (err)
638 pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
639 }
640
641 rcu_read_lock();
642 for_each_netdev_rcu(net, dev) {
643 if (dev->netdev_ops->ndo_add_vxlan_port)
644 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
645 port);
646 }
647 rcu_read_unlock();
648}
649
650/* Notify netdevs that UDP port is no more listening */
651static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
652{
653 struct net_device *dev;
654 struct sock *sk = vs->sock->sk;
655 struct net *net = sock_net(sk);
656 sa_family_t sa_family = vxlan_get_sk_family(vs);
657 __be16 port = inet_sk(sk)->inet_sport;
658
659 rcu_read_lock();
660 for_each_netdev_rcu(net, dev) {
661 if (dev->netdev_ops->ndo_del_vxlan_port)
662 dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
663 port);
664 }
665 rcu_read_unlock();
666
667 if (sa_family == AF_INET)
668 udp_del_offload(&vs->udp_offloads);
669}
670
671/* Add new entry to forwarding table -- assumes lock held */
672static int vxlan_fdb_create(struct vxlan_dev *vxlan,
673 const u8 *mac, union vxlan_addr *ip,
674 __u16 state, __u16 flags,
675 __be16 port, __be32 vni, __u32 ifindex,
676 __u8 ndm_flags)
677{
678 struct vxlan_rdst *rd = NULL;
679 struct vxlan_fdb *f;
680 int notify = 0;
681
682 f = __vxlan_find_mac(vxlan, mac);
683 if (f) {
684 if (flags & NLM_F_EXCL) {
685 netdev_dbg(vxlan->dev,
686 "lost race to create %pM\n", mac);
687 return -EEXIST;
688 }
689 if (f->state != state) {
690 f->state = state;
691 f->updated = jiffies;
692 notify = 1;
693 }
694 if (f->flags != ndm_flags) {
695 f->flags = ndm_flags;
696 f->updated = jiffies;
697 notify = 1;
698 }
699 if ((flags & NLM_F_REPLACE)) {
700 /* Only change unicasts */
701 if (!(is_multicast_ether_addr(f->eth_addr) ||
702 is_zero_ether_addr(f->eth_addr))) {
703 notify |= vxlan_fdb_replace(f, ip, port, vni,
704 ifindex);
705 } else
706 return -EOPNOTSUPP;
707 }
708 if ((flags & NLM_F_APPEND) &&
709 (is_multicast_ether_addr(f->eth_addr) ||
710 is_zero_ether_addr(f->eth_addr))) {
711 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
712 &rd);
713
714 if (rc < 0)
715 return rc;
716 notify |= rc;
717 }
718 } else {
719 if (!(flags & NLM_F_CREATE))
720 return -ENOENT;
721
722 if (vxlan->cfg.addrmax &&
723 vxlan->addrcnt >= vxlan->cfg.addrmax)
724 return -ENOSPC;
725
726 /* Disallow replace to add a multicast entry */
727 if ((flags & NLM_F_REPLACE) &&
728 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
729 return -EOPNOTSUPP;
730
731 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
732 f = kmalloc(sizeof(*f), GFP_ATOMIC);
733 if (!f)
734 return -ENOMEM;
735
736 notify = 1;
737 f->state = state;
738 f->flags = ndm_flags;
739 f->updated = f->used = jiffies;
740 INIT_LIST_HEAD(&f->remotes);
741 memcpy(f->eth_addr, mac, ETH_ALEN);
742
743 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
744
745 ++vxlan->addrcnt;
746 hlist_add_head_rcu(&f->hlist,
747 vxlan_fdb_head(vxlan, mac));
748 }
749
750 if (notify) {
751 if (rd == NULL)
752 rd = first_remote_rtnl(f);
753 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
754 }
755
756 return 0;
757}
758
759static void vxlan_fdb_free(struct rcu_head *head)
760{
761 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
762 struct vxlan_rdst *rd, *nd;
763
764 list_for_each_entry_safe(rd, nd, &f->remotes, list) {
765 dst_cache_destroy(&rd->dst_cache);
766 kfree(rd);
767 }
768 kfree(f);
769}
770
771static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
772{
773 netdev_dbg(vxlan->dev,
774 "delete %pM\n", f->eth_addr);
775
776 --vxlan->addrcnt;
777 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
778
779 hlist_del_rcu(&f->hlist);
780 call_rcu(&f->rcu, vxlan_fdb_free);
781}
782
783static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
784 union vxlan_addr *ip, __be16 *port, __be32 *vni,
785 u32 *ifindex)
786{
787 struct net *net = dev_net(vxlan->dev);
788 int err;
789
790 if (tb[NDA_DST]) {
791 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
792 if (err)
793 return err;
794 } else {
795 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
796 if (remote->sa.sa_family == AF_INET) {
797 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
798 ip->sa.sa_family = AF_INET;
799#if IS_ENABLED(CONFIG_IPV6)
800 } else {
801 ip->sin6.sin6_addr = in6addr_any;
802 ip->sa.sa_family = AF_INET6;
803#endif
804 }
805 }
806
807 if (tb[NDA_PORT]) {
808 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
809 return -EINVAL;
810 *port = nla_get_be16(tb[NDA_PORT]);
811 } else {
812 *port = vxlan->cfg.dst_port;
813 }
814
815 if (tb[NDA_VNI]) {
816 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
817 return -EINVAL;
818 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
819 } else {
820 *vni = vxlan->default_dst.remote_vni;
821 }
822
823 if (tb[NDA_IFINDEX]) {
824 struct net_device *tdev;
825
826 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
827 return -EINVAL;
828 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
829 tdev = __dev_get_by_index(net, *ifindex);
830 if (!tdev)
831 return -EADDRNOTAVAIL;
832 } else {
833 *ifindex = 0;
834 }
835
836 return 0;
837}
838
839/* Add static entry (via netlink) */
840static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
841 struct net_device *dev,
842 const unsigned char *addr, u16 vid, u16 flags)
843{
844 struct vxlan_dev *vxlan = netdev_priv(dev);
845 /* struct net *net = dev_net(vxlan->dev); */
846 union vxlan_addr ip;
847 __be16 port;
848 __be32 vni;
849 u32 ifindex;
850 int err;
851
852 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
853 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
854 ndm->ndm_state);
855 return -EINVAL;
856 }
857
858 if (tb[NDA_DST] == NULL)
859 return -EINVAL;
860
861 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
862 if (err)
863 return err;
864
865 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
866 return -EAFNOSUPPORT;
867
868 spin_lock_bh(&vxlan->hash_lock);
869 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
870 port, vni, ifindex, ndm->ndm_flags);
871 spin_unlock_bh(&vxlan->hash_lock);
872
873 return err;
874}
875
876/* Delete entry (via netlink) */
877static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
878 struct net_device *dev,
879 const unsigned char *addr, u16 vid)
880{
881 struct vxlan_dev *vxlan = netdev_priv(dev);
882 struct vxlan_fdb *f;
883 struct vxlan_rdst *rd = NULL;
884 union vxlan_addr ip;
885 __be16 port;
886 __be32 vni;
887 u32 ifindex;
888 int err;
889
890 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
891 if (err)
892 return err;
893
894 err = -ENOENT;
895
896 spin_lock_bh(&vxlan->hash_lock);
897 f = vxlan_find_mac(vxlan, addr);
898 if (!f)
899 goto out;
900
901 if (!vxlan_addr_any(&ip)) {
902 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
903 if (!rd)
904 goto out;
905 }
906
907 err = 0;
908
909 /* remove a destination if it's not the only one on the list,
910 * otherwise destroy the fdb entry
911 */
912 if (rd && !list_is_singular(&f->remotes)) {
913 list_del_rcu(&rd->list);
914 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
915 kfree_rcu(rd, rcu);
916 goto out;
917 }
918
919 vxlan_fdb_destroy(vxlan, f);
920
921out:
922 spin_unlock_bh(&vxlan->hash_lock);
923
924 return err;
925}
926
927/* Dump forwarding table */
928static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
929 struct net_device *dev,
930 struct net_device *filter_dev, int idx)
931{
932 struct vxlan_dev *vxlan = netdev_priv(dev);
933 unsigned int h;
934
935 for (h = 0; h < FDB_HASH_SIZE; ++h) {
936 struct vxlan_fdb *f;
937 int err;
938
939 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
940 struct vxlan_rdst *rd;
941
942 list_for_each_entry_rcu(rd, &f->remotes, list) {
943 if (idx < cb->args[0])
944 goto skip;
945
946 err = vxlan_fdb_info(skb, vxlan, f,
947 NETLINK_CB(cb->skb).portid,
948 cb->nlh->nlmsg_seq,
949 RTM_NEWNEIGH,
950 NLM_F_MULTI, rd);
951 if (err < 0) {
952 cb->args[1] = err;
953 goto out;
954 }
955skip:
956 ++idx;
957 }
958 }
959 }
960out:
961 return idx;
962}
963
964/* Watch incoming packets to learn mapping between Ethernet address
965 * and Tunnel endpoint.
966 * Return true if packet is bogus and should be dropped.
967 */
968static bool vxlan_snoop(struct net_device *dev,
969 union vxlan_addr *src_ip, const u8 *src_mac)
970{
971 struct vxlan_dev *vxlan = netdev_priv(dev);
972 struct vxlan_fdb *f;
973
974 f = vxlan_find_mac(vxlan, src_mac);
975 if (likely(f)) {
976 struct vxlan_rdst *rdst = first_remote_rcu(f);
977
978 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
979 return false;
980
981 /* Don't migrate static entries, drop packets */
982 if (f->state & NUD_NOARP)
983 return true;
984
985 if (net_ratelimit())
986 netdev_info(dev,
987 "%pM migrated from %pIS to %pIS\n",
988 src_mac, &rdst->remote_ip.sa, &src_ip->sa);
989
990 rdst->remote_ip = *src_ip;
991 f->updated = jiffies;
992 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
993 } else {
994 /* learned new entry */
995 spin_lock(&vxlan->hash_lock);
996
997 /* close off race between vxlan_flush and incoming packets */
998 if (netif_running(dev))
999 vxlan_fdb_create(vxlan, src_mac, src_ip,
1000 NUD_REACHABLE,
1001 NLM_F_EXCL|NLM_F_CREATE,
1002 vxlan->cfg.dst_port,
1003 vxlan->default_dst.remote_vni,
1004 0, NTF_SELF);
1005 spin_unlock(&vxlan->hash_lock);
1006 }
1007
1008 return false;
1009}
1010
1011/* See if multicast group is already in use by other ID */
1012static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1013{
1014 struct vxlan_dev *vxlan;
1015 unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1016
1017 /* The vxlan_sock is only used by dev, leaving group has
1018 * no effect on other vxlan devices.
1019 */
1020 if (family == AF_INET && dev->vn4_sock &&
1021 atomic_read(&dev->vn4_sock->refcnt) == 1)
1022 return false;
1023#if IS_ENABLED(CONFIG_IPV6)
1024 if (family == AF_INET6 && dev->vn6_sock &&
1025 atomic_read(&dev->vn6_sock->refcnt) == 1)
1026 return false;
1027#endif
1028
1029 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1030 if (!netif_running(vxlan->dev) || vxlan == dev)
1031 continue;
1032
1033 if (family == AF_INET && vxlan->vn4_sock != dev->vn4_sock)
1034 continue;
1035#if IS_ENABLED(CONFIG_IPV6)
1036 if (family == AF_INET6 && vxlan->vn6_sock != dev->vn6_sock)
1037 continue;
1038#endif
1039
1040 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1041 &dev->default_dst.remote_ip))
1042 continue;
1043
1044 if (vxlan->default_dst.remote_ifindex !=
1045 dev->default_dst.remote_ifindex)
1046 continue;
1047
1048 return true;
1049 }
1050
1051 return false;
1052}
1053
1054static void __vxlan_sock_release(struct vxlan_sock *vs)
1055{
1056 struct vxlan_net *vn;
1057
1058 if (!vs)
1059 return;
1060 if (!atomic_dec_and_test(&vs->refcnt))
1061 return;
1062
1063 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1064 spin_lock(&vn->sock_lock);
1065 hlist_del_rcu(&vs->hlist);
1066 vxlan_notify_del_rx_port(vs);
1067 spin_unlock(&vn->sock_lock);
1068
1069 queue_work(vxlan_wq, &vs->del_work);
1070}
1071
1072static void vxlan_sock_release(struct vxlan_dev *vxlan)
1073{
1074 __vxlan_sock_release(vxlan->vn4_sock);
1075#if IS_ENABLED(CONFIG_IPV6)
1076 __vxlan_sock_release(vxlan->vn6_sock);
1077#endif
1078}
1079
1080/* Update multicast group membership when first VNI on
1081 * multicast address is brought up
1082 */
1083static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1084{
1085 struct sock *sk;
1086 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1087 int ifindex = vxlan->default_dst.remote_ifindex;
1088 int ret = -EINVAL;
1089
1090 if (ip->sa.sa_family == AF_INET) {
1091 struct ip_mreqn mreq = {
1092 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1093 .imr_ifindex = ifindex,
1094 };
1095
1096 sk = vxlan->vn4_sock->sock->sk;
1097 lock_sock(sk);
1098 ret = ip_mc_join_group(sk, &mreq);
1099 release_sock(sk);
1100#if IS_ENABLED(CONFIG_IPV6)
1101 } else {
1102 sk = vxlan->vn6_sock->sock->sk;
1103 lock_sock(sk);
1104 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1105 &ip->sin6.sin6_addr);
1106 release_sock(sk);
1107#endif
1108 }
1109
1110 return ret;
1111}
1112
1113/* Inverse of vxlan_igmp_join when last VNI is brought down */
1114static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1115{
1116 struct sock *sk;
1117 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1118 int ifindex = vxlan->default_dst.remote_ifindex;
1119 int ret = -EINVAL;
1120
1121 if (ip->sa.sa_family == AF_INET) {
1122 struct ip_mreqn mreq = {
1123 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1124 .imr_ifindex = ifindex,
1125 };
1126
1127 sk = vxlan->vn4_sock->sock->sk;
1128 lock_sock(sk);
1129 ret = ip_mc_leave_group(sk, &mreq);
1130 release_sock(sk);
1131#if IS_ENABLED(CONFIG_IPV6)
1132 } else {
1133 sk = vxlan->vn6_sock->sock->sk;
1134 lock_sock(sk);
1135 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1136 &ip->sin6.sin6_addr);
1137 release_sock(sk);
1138#endif
1139 }
1140
1141 return ret;
1142}
1143
1144static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1145 struct sk_buff *skb, u32 vxflags)
1146{
1147 size_t start, offset;
1148
1149 if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1150 goto out;
1151
1152 start = vxlan_rco_start(unparsed->vx_vni);
1153 offset = start + vxlan_rco_offset(unparsed->vx_vni);
1154
1155 if (!pskb_may_pull(skb, offset + sizeof(u16)))
1156 return false;
1157
1158 skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1159 !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1160out:
1161 unparsed->vx_flags &= ~VXLAN_HF_RCO;
1162 unparsed->vx_vni &= VXLAN_VNI_MASK;
1163 return true;
1164}
1165
1166static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1167 struct sk_buff *skb, u32 vxflags,
1168 struct vxlan_metadata *md)
1169{
1170 struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1171 struct metadata_dst *tun_dst;
1172
1173 if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1174 goto out;
1175
1176 md->gbp = ntohs(gbp->policy_id);
1177
1178 tun_dst = (struct metadata_dst *)skb_dst(skb);
1179 if (tun_dst) {
1180 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1181 tun_dst->u.tun_info.options_len = sizeof(*md);
1182 }
1183 if (gbp->dont_learn)
1184 md->gbp |= VXLAN_GBP_DONT_LEARN;
1185
1186 if (gbp->policy_applied)
1187 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1188
1189 /* In flow-based mode, GBP is carried in dst_metadata */
1190 if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1191 skb->mark = md->gbp;
1192out:
1193 unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1194}
1195
1196static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1197 struct vxlan_sock *vs,
1198 struct sk_buff *skb)
1199{
1200 union vxlan_addr saddr;
1201
1202 skb_reset_mac_header(skb);
1203 skb->protocol = eth_type_trans(skb, vxlan->dev);
1204 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1205
1206 /* Ignore packet loops (and multicast echo) */
1207 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1208 return false;
1209
1210 /* Get address from the outer IP header */
1211 if (vxlan_get_sk_family(vs) == AF_INET) {
1212 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1213 saddr.sa.sa_family = AF_INET;
1214#if IS_ENABLED(CONFIG_IPV6)
1215 } else {
1216 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1217 saddr.sa.sa_family = AF_INET6;
1218#endif
1219 }
1220
1221 if ((vxlan->flags & VXLAN_F_LEARN) &&
1222 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1223 return false;
1224
1225 return true;
1226}
1227
1228static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1229 struct sk_buff *skb)
1230{
1231 int err = 0;
1232
1233 if (vxlan_get_sk_family(vs) == AF_INET)
1234 err = IP_ECN_decapsulate(oiph, skb);
1235#if IS_ENABLED(CONFIG_IPV6)
1236 else
1237 err = IP6_ECN_decapsulate(oiph, skb);
1238#endif
1239
1240 if (unlikely(err) && log_ecn_error) {
1241 if (vxlan_get_sk_family(vs) == AF_INET)
1242 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1243 &((struct iphdr *)oiph)->saddr,
1244 ((struct iphdr *)oiph)->tos);
1245 else
1246 net_info_ratelimited("non-ECT from %pI6\n",
1247 &((struct ipv6hdr *)oiph)->saddr);
1248 }
1249 return err <= 1;
1250}
1251
1252/* Callback from net/ipv4/udp.c to receive packets */
1253static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1254{
1255 struct pcpu_sw_netstats *stats;
1256 struct vxlan_dev *vxlan;
1257 struct vxlan_sock *vs;
1258 struct vxlanhdr unparsed;
1259 struct vxlan_metadata _md;
1260 struct vxlan_metadata *md = &_md;
1261 void *oiph;
1262
1263 /* Need Vxlan and inner Ethernet header to be present */
1264 if (!pskb_may_pull(skb, VXLAN_HLEN))
1265 return 1;
1266
1267 unparsed = *vxlan_hdr(skb);
1268 /* VNI flag always required to be set */
1269 if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1270 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1271 ntohl(vxlan_hdr(skb)->vx_flags),
1272 ntohl(vxlan_hdr(skb)->vx_vni));
1273 /* Return non vxlan pkt */
1274 return 1;
1275 }
1276 unparsed.vx_flags &= ~VXLAN_HF_VNI;
1277 unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1278
1279 vs = rcu_dereference_sk_user_data(sk);
1280 if (!vs)
1281 goto drop;
1282
1283 vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1284 if (!vxlan)
1285 goto drop;
1286
1287 if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB),
1288 !net_eq(vxlan->net, dev_net(vxlan->dev))))
1289 goto drop;
1290
1291 if (vxlan_collect_metadata(vs)) {
1292 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1293 struct metadata_dst *tun_dst;
1294
1295 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1296 vxlan_vni_to_tun_id(vni), sizeof(*md));
1297
1298 if (!tun_dst)
1299 goto drop;
1300
1301 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1302
1303 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1304 } else {
1305 memset(md, 0, sizeof(*md));
1306 }
1307
1308 /* For backwards compatibility, only allow reserved fields to be
1309 * used by VXLAN extensions if explicitly requested.
1310 */
1311 if (vs->flags & VXLAN_F_REMCSUM_RX)
1312 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1313 goto drop;
1314 if (vs->flags & VXLAN_F_GBP)
1315 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1316
1317 if (unparsed.vx_flags || unparsed.vx_vni) {
1318 /* If there are any unprocessed flags remaining treat
1319 * this as a malformed packet. This behavior diverges from
1320 * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1321 * in reserved fields are to be ignored. The approach here
1322 * maintains compatibility with previous stack code, and also
1323 * is more robust and provides a little more security in
1324 * adding extensions to VXLAN.
1325 */
1326 goto drop;
1327 }
1328
1329 if (!vxlan_set_mac(vxlan, vs, skb))
1330 goto drop;
1331
1332 oiph = skb_network_header(skb);
1333 skb_reset_network_header(skb);
1334
1335 if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1336 ++vxlan->dev->stats.rx_frame_errors;
1337 ++vxlan->dev->stats.rx_errors;
1338 goto drop;
1339 }
1340
1341 stats = this_cpu_ptr(vxlan->dev->tstats);
1342 u64_stats_update_begin(&stats->syncp);
1343 stats->rx_packets++;
1344 stats->rx_bytes += skb->len;
1345 u64_stats_update_end(&stats->syncp);
1346
1347 gro_cells_receive(&vxlan->gro_cells, skb);
1348 return 0;
1349
1350drop:
1351 /* Consume bad packet */
1352 kfree_skb(skb);
1353 return 0;
1354}
1355
1356static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1357{
1358 struct vxlan_dev *vxlan = netdev_priv(dev);
1359 struct arphdr *parp;
1360 u8 *arpptr, *sha;
1361 __be32 sip, tip;
1362 struct neighbour *n;
1363
1364 if (dev->flags & IFF_NOARP)
1365 goto out;
1366
1367 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1368 dev->stats.tx_dropped++;
1369 goto out;
1370 }
1371 parp = arp_hdr(skb);
1372
1373 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1374 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1375 parp->ar_pro != htons(ETH_P_IP) ||
1376 parp->ar_op != htons(ARPOP_REQUEST) ||
1377 parp->ar_hln != dev->addr_len ||
1378 parp->ar_pln != 4)
1379 goto out;
1380 arpptr = (u8 *)parp + sizeof(struct arphdr);
1381 sha = arpptr;
1382 arpptr += dev->addr_len; /* sha */
1383 memcpy(&sip, arpptr, sizeof(sip));
1384 arpptr += sizeof(sip);
1385 arpptr += dev->addr_len; /* tha */
1386 memcpy(&tip, arpptr, sizeof(tip));
1387
1388 if (ipv4_is_loopback(tip) ||
1389 ipv4_is_multicast(tip))
1390 goto out;
1391
1392 n = neigh_lookup(&arp_tbl, &tip, dev);
1393
1394 if (n) {
1395 struct vxlan_fdb *f;
1396 struct sk_buff *reply;
1397
1398 if (!(n->nud_state & NUD_CONNECTED)) {
1399 neigh_release(n);
1400 goto out;
1401 }
1402
1403 f = vxlan_find_mac(vxlan, n->ha);
1404 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1405 /* bridge-local neighbor */
1406 neigh_release(n);
1407 goto out;
1408 }
1409
1410 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1411 n->ha, sha);
1412
1413 neigh_release(n);
1414
1415 if (reply == NULL)
1416 goto out;
1417
1418 skb_reset_mac_header(reply);
1419 __skb_pull(reply, skb_network_offset(reply));
1420 reply->ip_summed = CHECKSUM_UNNECESSARY;
1421 reply->pkt_type = PACKET_HOST;
1422
1423 if (netif_rx_ni(reply) == NET_RX_DROP)
1424 dev->stats.rx_dropped++;
1425 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1426 union vxlan_addr ipa = {
1427 .sin.sin_addr.s_addr = tip,
1428 .sin.sin_family = AF_INET,
1429 };
1430
1431 vxlan_ip_miss(dev, &ipa);
1432 }
1433out:
1434 consume_skb(skb);
1435 return NETDEV_TX_OK;
1436}
1437
1438#if IS_ENABLED(CONFIG_IPV6)
1439static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1440 struct neighbour *n, bool isrouter)
1441{
1442 struct net_device *dev = request->dev;
1443 struct sk_buff *reply;
1444 struct nd_msg *ns, *na;
1445 struct ipv6hdr *pip6;
1446 u8 *daddr;
1447 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1448 int ns_olen;
1449 int i, len;
1450
1451 if (dev == NULL)
1452 return NULL;
1453
1454 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1455 sizeof(*na) + na_olen + dev->needed_tailroom;
1456 reply = alloc_skb(len, GFP_ATOMIC);
1457 if (reply == NULL)
1458 return NULL;
1459
1460 reply->protocol = htons(ETH_P_IPV6);
1461 reply->dev = dev;
1462 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1463 skb_push(reply, sizeof(struct ethhdr));
1464 skb_reset_mac_header(reply);
1465
1466 ns = (struct nd_msg *)skb_transport_header(request);
1467
1468 daddr = eth_hdr(request)->h_source;
1469 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1470 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1471 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1472 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1473 break;
1474 }
1475 }
1476
1477 /* Ethernet header */
1478 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1479 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1480 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1481 reply->protocol = htons(ETH_P_IPV6);
1482
1483 skb_pull(reply, sizeof(struct ethhdr));
1484 skb_reset_network_header(reply);
1485 skb_put(reply, sizeof(struct ipv6hdr));
1486
1487 /* IPv6 header */
1488
1489 pip6 = ipv6_hdr(reply);
1490 memset(pip6, 0, sizeof(struct ipv6hdr));
1491 pip6->version = 6;
1492 pip6->priority = ipv6_hdr(request)->priority;
1493 pip6->nexthdr = IPPROTO_ICMPV6;
1494 pip6->hop_limit = 255;
1495 pip6->daddr = ipv6_hdr(request)->saddr;
1496 pip6->saddr = *(struct in6_addr *)n->primary_key;
1497
1498 skb_pull(reply, sizeof(struct ipv6hdr));
1499 skb_reset_transport_header(reply);
1500
1501 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1502
1503 /* Neighbor Advertisement */
1504 memset(na, 0, sizeof(*na)+na_olen);
1505 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1506 na->icmph.icmp6_router = isrouter;
1507 na->icmph.icmp6_override = 1;
1508 na->icmph.icmp6_solicited = 1;
1509 na->target = ns->target;
1510 ether_addr_copy(&na->opt[2], n->ha);
1511 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1512 na->opt[1] = na_olen >> 3;
1513
1514 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1515 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1516 csum_partial(na, sizeof(*na)+na_olen, 0));
1517
1518 pip6->payload_len = htons(sizeof(*na)+na_olen);
1519
1520 skb_push(reply, sizeof(struct ipv6hdr));
1521
1522 reply->ip_summed = CHECKSUM_UNNECESSARY;
1523
1524 return reply;
1525}
1526
1527static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1528{
1529 struct vxlan_dev *vxlan = netdev_priv(dev);
1530 struct nd_msg *msg;
1531 const struct ipv6hdr *iphdr;
1532 const struct in6_addr *saddr, *daddr;
1533 struct neighbour *n;
1534 struct inet6_dev *in6_dev;
1535
1536 in6_dev = __in6_dev_get(dev);
1537 if (!in6_dev)
1538 goto out;
1539
1540 iphdr = ipv6_hdr(skb);
1541 saddr = &iphdr->saddr;
1542 daddr = &iphdr->daddr;
1543
1544 msg = (struct nd_msg *)skb_transport_header(skb);
1545 if (msg->icmph.icmp6_code != 0 ||
1546 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1547 goto out;
1548
1549 if (ipv6_addr_loopback(daddr) ||
1550 ipv6_addr_is_multicast(&msg->target))
1551 goto out;
1552
1553 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1554
1555 if (n) {
1556 struct vxlan_fdb *f;
1557 struct sk_buff *reply;
1558
1559 if (!(n->nud_state & NUD_CONNECTED)) {
1560 neigh_release(n);
1561 goto out;
1562 }
1563
1564 f = vxlan_find_mac(vxlan, n->ha);
1565 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1566 /* bridge-local neighbor */
1567 neigh_release(n);
1568 goto out;
1569 }
1570
1571 reply = vxlan_na_create(skb, n,
1572 !!(f ? f->flags & NTF_ROUTER : 0));
1573
1574 neigh_release(n);
1575
1576 if (reply == NULL)
1577 goto out;
1578
1579 if (netif_rx_ni(reply) == NET_RX_DROP)
1580 dev->stats.rx_dropped++;
1581
1582 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1583 union vxlan_addr ipa = {
1584 .sin6.sin6_addr = msg->target,
1585 .sin6.sin6_family = AF_INET6,
1586 };
1587
1588 vxlan_ip_miss(dev, &ipa);
1589 }
1590
1591out:
1592 consume_skb(skb);
1593 return NETDEV_TX_OK;
1594}
1595#endif
1596
1597static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1598{
1599 struct vxlan_dev *vxlan = netdev_priv(dev);
1600 struct neighbour *n;
1601
1602 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1603 return false;
1604
1605 n = NULL;
1606 switch (ntohs(eth_hdr(skb)->h_proto)) {
1607 case ETH_P_IP:
1608 {
1609 struct iphdr *pip;
1610
1611 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1612 return false;
1613 pip = ip_hdr(skb);
1614 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1615 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1616 union vxlan_addr ipa = {
1617 .sin.sin_addr.s_addr = pip->daddr,
1618 .sin.sin_family = AF_INET,
1619 };
1620
1621 vxlan_ip_miss(dev, &ipa);
1622 return false;
1623 }
1624
1625 break;
1626 }
1627#if IS_ENABLED(CONFIG_IPV6)
1628 case ETH_P_IPV6:
1629 {
1630 struct ipv6hdr *pip6;
1631
1632 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1633 return false;
1634 pip6 = ipv6_hdr(skb);
1635 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1636 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1637 union vxlan_addr ipa = {
1638 .sin6.sin6_addr = pip6->daddr,
1639 .sin6.sin6_family = AF_INET6,
1640 };
1641
1642 vxlan_ip_miss(dev, &ipa);
1643 return false;
1644 }
1645
1646 break;
1647 }
1648#endif
1649 default:
1650 return false;
1651 }
1652
1653 if (n) {
1654 bool diff;
1655
1656 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1657 if (diff) {
1658 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1659 dev->addr_len);
1660 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1661 }
1662 neigh_release(n);
1663 return diff;
1664 }
1665
1666 return false;
1667}
1668
1669static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1670 struct vxlan_metadata *md)
1671{
1672 struct vxlanhdr_gbp *gbp;
1673
1674 if (!md->gbp)
1675 return;
1676
1677 gbp = (struct vxlanhdr_gbp *)vxh;
1678 vxh->vx_flags |= VXLAN_HF_GBP;
1679
1680 if (md->gbp & VXLAN_GBP_DONT_LEARN)
1681 gbp->dont_learn = 1;
1682
1683 if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1684 gbp->policy_applied = 1;
1685
1686 gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1687}
1688
1689static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1690 int iphdr_len, __be32 vni,
1691 struct vxlan_metadata *md, u32 vxflags,
1692 bool udp_sum)
1693{
1694 struct vxlanhdr *vxh;
1695 int min_headroom;
1696 int err;
1697 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1698
1699 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1700 skb->ip_summed == CHECKSUM_PARTIAL) {
1701 int csum_start = skb_checksum_start_offset(skb);
1702
1703 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1704 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1705 (skb->csum_offset == offsetof(struct udphdr, check) ||
1706 skb->csum_offset == offsetof(struct tcphdr, check)))
1707 type |= SKB_GSO_TUNNEL_REMCSUM;
1708 }
1709
1710 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1711 + VXLAN_HLEN + iphdr_len
1712 + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1713
1714 /* Need space for new headers (invalidates iph ptr) */
1715 err = skb_cow_head(skb, min_headroom);
1716 if (unlikely(err)) {
1717 kfree_skb(skb);
1718 return err;
1719 }
1720
1721 skb = vlan_hwaccel_push_inside(skb);
1722 if (WARN_ON(!skb))
1723 return -ENOMEM;
1724
1725 skb = iptunnel_handle_offloads(skb, type);
1726 if (IS_ERR(skb))
1727 return PTR_ERR(skb);
1728
1729 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1730 vxh->vx_flags = VXLAN_HF_VNI;
1731 vxh->vx_vni = vxlan_vni_field(vni);
1732
1733 if (type & SKB_GSO_TUNNEL_REMCSUM) {
1734 unsigned int start;
1735
1736 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1737 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1738 vxh->vx_flags |= VXLAN_HF_RCO;
1739
1740 if (!skb_is_gso(skb)) {
1741 skb->ip_summed = CHECKSUM_NONE;
1742 skb->encapsulation = 0;
1743 }
1744 }
1745
1746 if (vxflags & VXLAN_F_GBP)
1747 vxlan_build_gbp_hdr(vxh, vxflags, md);
1748
1749 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1750 return 0;
1751}
1752
1753static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1754 struct sk_buff *skb, int oif, u8 tos,
1755 __be32 daddr, __be32 *saddr,
1756 struct dst_cache *dst_cache,
1757 const struct ip_tunnel_info *info)
1758{
1759 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1760 struct rtable *rt = NULL;
1761 struct flowi4 fl4;
1762
1763 if (tos && !info)
1764 use_cache = false;
1765 if (use_cache) {
1766 rt = dst_cache_get_ip4(dst_cache, saddr);
1767 if (rt)
1768 return rt;
1769 }
1770
1771 memset(&fl4, 0, sizeof(fl4));
1772 fl4.flowi4_oif = oif;
1773 fl4.flowi4_tos = RT_TOS(tos);
1774 fl4.flowi4_mark = skb->mark;
1775 fl4.flowi4_proto = IPPROTO_UDP;
1776 fl4.daddr = daddr;
1777 fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
1778
1779 rt = ip_route_output_key(vxlan->net, &fl4);
1780 if (!IS_ERR(rt)) {
1781 *saddr = fl4.saddr;
1782 if (use_cache)
1783 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1784 }
1785 return rt;
1786}
1787
1788#if IS_ENABLED(CONFIG_IPV6)
1789static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1790 struct sk_buff *skb, int oif, u8 tos,
1791 __be32 label,
1792 const struct in6_addr *daddr,
1793 struct in6_addr *saddr,
1794 struct dst_cache *dst_cache,
1795 const struct ip_tunnel_info *info)
1796{
1797 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1798 struct dst_entry *ndst;
1799 struct flowi6 fl6;
1800 int err;
1801
1802 if (tos && !info)
1803 use_cache = false;
1804 if (use_cache) {
1805 ndst = dst_cache_get_ip6(dst_cache, saddr);
1806 if (ndst)
1807 return ndst;
1808 }
1809
1810 memset(&fl6, 0, sizeof(fl6));
1811 fl6.flowi6_oif = oif;
1812 fl6.daddr = *daddr;
1813 fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
1814 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1815 fl6.flowi6_mark = skb->mark;
1816 fl6.flowi6_proto = IPPROTO_UDP;
1817
1818 err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1819 vxlan->vn6_sock->sock->sk,
1820 &ndst, &fl6);
1821 if (err < 0)
1822 return ERR_PTR(err);
1823
1824 *saddr = fl6.saddr;
1825 if (use_cache)
1826 dst_cache_set_ip6(dst_cache, ndst, saddr);
1827 return ndst;
1828}
1829#endif
1830
1831/* Bypass encapsulation if the destination is local */
1832static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1833 struct vxlan_dev *dst_vxlan)
1834{
1835 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1836 union vxlan_addr loopback;
1837 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1838 struct net_device *dev = skb->dev;
1839 int len = skb->len;
1840
1841 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1842 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1843 skb->pkt_type = PACKET_HOST;
1844 skb->encapsulation = 0;
1845 skb->dev = dst_vxlan->dev;
1846 __skb_pull(skb, skb_network_offset(skb));
1847
1848 if (remote_ip->sa.sa_family == AF_INET) {
1849 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1850 loopback.sa.sa_family = AF_INET;
1851#if IS_ENABLED(CONFIG_IPV6)
1852 } else {
1853 loopback.sin6.sin6_addr = in6addr_loopback;
1854 loopback.sa.sa_family = AF_INET6;
1855#endif
1856 }
1857
1858 if (dst_vxlan->flags & VXLAN_F_LEARN)
1859 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1860
1861 u64_stats_update_begin(&tx_stats->syncp);
1862 tx_stats->tx_packets++;
1863 tx_stats->tx_bytes += len;
1864 u64_stats_update_end(&tx_stats->syncp);
1865
1866 if (netif_rx(skb) == NET_RX_SUCCESS) {
1867 u64_stats_update_begin(&rx_stats->syncp);
1868 rx_stats->rx_packets++;
1869 rx_stats->rx_bytes += len;
1870 u64_stats_update_end(&rx_stats->syncp);
1871 } else {
1872 dev->stats.rx_dropped++;
1873 }
1874}
1875
1876static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1877 struct vxlan_rdst *rdst, bool did_rsc)
1878{
1879 struct dst_cache *dst_cache;
1880 struct ip_tunnel_info *info;
1881 struct vxlan_dev *vxlan = netdev_priv(dev);
1882 struct sock *sk;
1883 struct rtable *rt = NULL;
1884 const struct iphdr *old_iph;
1885 union vxlan_addr *dst;
1886 union vxlan_addr remote_ip;
1887 struct vxlan_metadata _md;
1888 struct vxlan_metadata *md = &_md;
1889 __be16 src_port = 0, dst_port;
1890 __be32 vni, label;
1891 __be16 df = 0;
1892 __u8 tos, ttl;
1893 int err;
1894 u32 flags = vxlan->flags;
1895 bool udp_sum = false;
1896 bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1897
1898 info = skb_tunnel_info(skb);
1899
1900 if (rdst) {
1901 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1902 vni = rdst->remote_vni;
1903 dst = &rdst->remote_ip;
1904 dst_cache = &rdst->dst_cache;
1905 } else {
1906 if (!info) {
1907 WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1908 dev->name);
1909 goto drop;
1910 }
1911 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1912 vni = vxlan_tun_id_to_vni(info->key.tun_id);
1913 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1914 if (remote_ip.sa.sa_family == AF_INET)
1915 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1916 else
1917 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1918 dst = &remote_ip;
1919 dst_cache = &info->dst_cache;
1920 }
1921
1922 if (vxlan_addr_any(dst)) {
1923 if (did_rsc) {
1924 /* short-circuited back to local bridge */
1925 vxlan_encap_bypass(skb, vxlan, vxlan);
1926 return;
1927 }
1928 goto drop;
1929 }
1930
1931 old_iph = ip_hdr(skb);
1932
1933 ttl = vxlan->cfg.ttl;
1934 if (!ttl && vxlan_addr_multicast(dst))
1935 ttl = 1;
1936
1937 tos = vxlan->cfg.tos;
1938 if (tos == 1)
1939 tos = ip_tunnel_get_dsfield(old_iph, skb);
1940
1941 label = vxlan->cfg.label;
1942 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1943 vxlan->cfg.port_max, true);
1944
1945 if (info) {
1946 ttl = info->key.ttl;
1947 tos = info->key.tos;
1948 label = info->key.label;
1949 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
1950
1951 if (info->options_len)
1952 md = ip_tunnel_info_opts(info);
1953 } else {
1954 md->gbp = skb->mark;
1955 }
1956
1957 if (dst->sa.sa_family == AF_INET) {
1958 __be32 saddr;
1959
1960 if (!vxlan->vn4_sock)
1961 goto drop;
1962 sk = vxlan->vn4_sock->sock->sk;
1963
1964 rt = vxlan_get_route(vxlan, skb,
1965 rdst ? rdst->remote_ifindex : 0, tos,
1966 dst->sin.sin_addr.s_addr, &saddr,
1967 dst_cache, info);
1968 if (IS_ERR(rt)) {
1969 netdev_dbg(dev, "no route to %pI4\n",
1970 &dst->sin.sin_addr.s_addr);
1971 dev->stats.tx_carrier_errors++;
1972 goto tx_error;
1973 }
1974
1975 if (rt->dst.dev == dev) {
1976 netdev_dbg(dev, "circular route to %pI4\n",
1977 &dst->sin.sin_addr.s_addr);
1978 dev->stats.collisions++;
1979 goto rt_tx_error;
1980 }
1981
1982 /* Bypass encapsulation if the destination is local */
1983 if (rt->rt_flags & RTCF_LOCAL &&
1984 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1985 struct vxlan_dev *dst_vxlan;
1986
1987 ip_rt_put(rt);
1988 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1989 dst->sa.sa_family, dst_port,
1990 vxlan->flags);
1991 if (!dst_vxlan)
1992 goto tx_error;
1993 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1994 return;
1995 }
1996
1997 if (!info)
1998 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
1999 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2000 df = htons(IP_DF);
2001
2002 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2003 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2004 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2005 vni, md, flags, udp_sum);
2006 if (err < 0)
2007 goto xmit_tx_error;
2008
2009 udp_tunnel_xmit_skb(rt, sk, skb, saddr,
2010 dst->sin.sin_addr.s_addr, tos, ttl, df,
2011 src_port, dst_port, xnet, !udp_sum);
2012#if IS_ENABLED(CONFIG_IPV6)
2013 } else {
2014 struct dst_entry *ndst;
2015 struct in6_addr saddr;
2016 u32 rt6i_flags;
2017
2018 if (!vxlan->vn6_sock)
2019 goto drop;
2020 sk = vxlan->vn6_sock->sock->sk;
2021
2022 ndst = vxlan6_get_route(vxlan, skb,
2023 rdst ? rdst->remote_ifindex : 0, tos,
2024 label, &dst->sin6.sin6_addr, &saddr,
2025 dst_cache, info);
2026 if (IS_ERR(ndst)) {
2027 netdev_dbg(dev, "no route to %pI6\n",
2028 &dst->sin6.sin6_addr);
2029 dev->stats.tx_carrier_errors++;
2030 goto tx_error;
2031 }
2032
2033 if (ndst->dev == dev) {
2034 netdev_dbg(dev, "circular route to %pI6\n",
2035 &dst->sin6.sin6_addr);
2036 dst_release(ndst);
2037 dev->stats.collisions++;
2038 goto tx_error;
2039 }
2040
2041 /* Bypass encapsulation if the destination is local */
2042 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2043 if (rt6i_flags & RTF_LOCAL &&
2044 !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2045 struct vxlan_dev *dst_vxlan;
2046
2047 dst_release(ndst);
2048 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2049 dst->sa.sa_family, dst_port,
2050 vxlan->flags);
2051 if (!dst_vxlan)
2052 goto tx_error;
2053 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2054 return;
2055 }
2056
2057 if (!info)
2058 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2059
2060 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2061 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2062 skb_scrub_packet(skb, xnet);
2063 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2064 vni, md, flags, udp_sum);
2065 if (err < 0) {
2066 dst_release(ndst);
2067 return;
2068 }
2069 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2070 &saddr, &dst->sin6.sin6_addr, tos, ttl,
2071 label, src_port, dst_port, !udp_sum);
2072#endif
2073 }
2074
2075 return;
2076
2077drop:
2078 dev->stats.tx_dropped++;
2079 goto tx_free;
2080
2081xmit_tx_error:
2082 /* skb is already freed. */
2083 skb = NULL;
2084rt_tx_error:
2085 ip_rt_put(rt);
2086tx_error:
2087 dev->stats.tx_errors++;
2088tx_free:
2089 dev_kfree_skb(skb);
2090}
2091
2092/* Transmit local packets over Vxlan
2093 *
2094 * Outer IP header inherits ECN and DF from inner header.
2095 * Outer UDP destination is the VXLAN assigned port.
2096 * source port is based on hash of flow
2097 */
2098static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2099{
2100 struct vxlan_dev *vxlan = netdev_priv(dev);
2101 const struct ip_tunnel_info *info;
2102 struct ethhdr *eth;
2103 bool did_rsc = false;
2104 struct vxlan_rdst *rdst, *fdst = NULL;
2105 struct vxlan_fdb *f;
2106
2107 info = skb_tunnel_info(skb);
2108
2109 skb_reset_mac_header(skb);
2110 eth = eth_hdr(skb);
2111
2112 if ((vxlan->flags & VXLAN_F_PROXY)) {
2113 if (ntohs(eth->h_proto) == ETH_P_ARP)
2114 return arp_reduce(dev, skb);
2115#if IS_ENABLED(CONFIG_IPV6)
2116 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2117 pskb_may_pull(skb, sizeof(struct ipv6hdr)
2118 + sizeof(struct nd_msg)) &&
2119 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2120 struct nd_msg *msg;
2121
2122 msg = (struct nd_msg *)skb_transport_header(skb);
2123 if (msg->icmph.icmp6_code == 0 &&
2124 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2125 return neigh_reduce(dev, skb);
2126 }
2127 eth = eth_hdr(skb);
2128#endif
2129 }
2130
2131 if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2132 if (info && info->mode & IP_TUNNEL_INFO_TX)
2133 vxlan_xmit_one(skb, dev, NULL, false);
2134 else
2135 kfree_skb(skb);
2136 return NETDEV_TX_OK;
2137 }
2138
2139 f = vxlan_find_mac(vxlan, eth->h_dest);
2140 did_rsc = false;
2141
2142 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2143 (ntohs(eth->h_proto) == ETH_P_IP ||
2144 ntohs(eth->h_proto) == ETH_P_IPV6)) {
2145 did_rsc = route_shortcircuit(dev, skb);
2146 if (did_rsc)
2147 f = vxlan_find_mac(vxlan, eth->h_dest);
2148 }
2149
2150 if (f == NULL) {
2151 f = vxlan_find_mac(vxlan, all_zeros_mac);
2152 if (f == NULL) {
2153 if ((vxlan->flags & VXLAN_F_L2MISS) &&
2154 !is_multicast_ether_addr(eth->h_dest))
2155 vxlan_fdb_miss(vxlan, eth->h_dest);
2156
2157 dev->stats.tx_dropped++;
2158 kfree_skb(skb);
2159 return NETDEV_TX_OK;
2160 }
2161 }
2162
2163 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2164 struct sk_buff *skb1;
2165
2166 if (!fdst) {
2167 fdst = rdst;
2168 continue;
2169 }
2170 skb1 = skb_clone(skb, GFP_ATOMIC);
2171 if (skb1)
2172 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2173 }
2174
2175 if (fdst)
2176 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2177 else
2178 kfree_skb(skb);
2179 return NETDEV_TX_OK;
2180}
2181
2182/* Walk the forwarding table and purge stale entries */
2183static void vxlan_cleanup(unsigned long arg)
2184{
2185 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2186 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2187 unsigned int h;
2188
2189 if (!netif_running(vxlan->dev))
2190 return;
2191
2192 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2193 struct hlist_node *p, *n;
2194
2195 spin_lock_bh(&vxlan->hash_lock);
2196 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2197 struct vxlan_fdb *f
2198 = container_of(p, struct vxlan_fdb, hlist);
2199 unsigned long timeout;
2200
2201 if (f->state & NUD_PERMANENT)
2202 continue;
2203
2204 timeout = f->used + vxlan->cfg.age_interval * HZ;
2205 if (time_before_eq(timeout, jiffies)) {
2206 netdev_dbg(vxlan->dev,
2207 "garbage collect %pM\n",
2208 f->eth_addr);
2209 f->state = NUD_STALE;
2210 vxlan_fdb_destroy(vxlan, f);
2211 } else if (time_before(timeout, next_timer))
2212 next_timer = timeout;
2213 }
2214 spin_unlock_bh(&vxlan->hash_lock);
2215 }
2216
2217 mod_timer(&vxlan->age_timer, next_timer);
2218}
2219
2220static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2221{
2222 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2223 __be32 vni = vxlan->default_dst.remote_vni;
2224
2225 spin_lock(&vn->sock_lock);
2226 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2227 spin_unlock(&vn->sock_lock);
2228}
2229
2230/* Setup stats when device is created */
2231static int vxlan_init(struct net_device *dev)
2232{
2233 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2234 if (!dev->tstats)
2235 return -ENOMEM;
2236
2237 return 0;
2238}
2239
2240static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2241{
2242 struct vxlan_fdb *f;
2243
2244 spin_lock_bh(&vxlan->hash_lock);
2245 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2246 if (f)
2247 vxlan_fdb_destroy(vxlan, f);
2248 spin_unlock_bh(&vxlan->hash_lock);
2249}
2250
2251static void vxlan_uninit(struct net_device *dev)
2252{
2253 struct vxlan_dev *vxlan = netdev_priv(dev);
2254
2255 vxlan_fdb_delete_default(vxlan);
2256
2257 free_percpu(dev->tstats);
2258}
2259
2260/* Start ageing timer and join group when device is brought up */
2261static int vxlan_open(struct net_device *dev)
2262{
2263 struct vxlan_dev *vxlan = netdev_priv(dev);
2264 int ret;
2265
2266 ret = vxlan_sock_add(vxlan);
2267 if (ret < 0)
2268 return ret;
2269
2270 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2271 ret = vxlan_igmp_join(vxlan);
2272 if (ret == -EADDRINUSE)
2273 ret = 0;
2274 if (ret) {
2275 vxlan_sock_release(vxlan);
2276 return ret;
2277 }
2278 }
2279
2280 if (vxlan->cfg.age_interval)
2281 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2282
2283 return ret;
2284}
2285
2286/* Purge the forwarding table */
2287static void vxlan_flush(struct vxlan_dev *vxlan)
2288{
2289 unsigned int h;
2290
2291 spin_lock_bh(&vxlan->hash_lock);
2292 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2293 struct hlist_node *p, *n;
2294 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2295 struct vxlan_fdb *f
2296 = container_of(p, struct vxlan_fdb, hlist);
2297 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2298 if (!is_zero_ether_addr(f->eth_addr))
2299 vxlan_fdb_destroy(vxlan, f);
2300 }
2301 }
2302 spin_unlock_bh(&vxlan->hash_lock);
2303}
2304
2305/* Cleanup timer and forwarding table on shutdown */
2306static int vxlan_stop(struct net_device *dev)
2307{
2308 struct vxlan_dev *vxlan = netdev_priv(dev);
2309 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2310 int ret = 0;
2311
2312 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2313 !vxlan_group_used(vn, vxlan))
2314 ret = vxlan_igmp_leave(vxlan);
2315
2316 del_timer_sync(&vxlan->age_timer);
2317
2318 vxlan_flush(vxlan);
2319 vxlan_sock_release(vxlan);
2320
2321 return ret;
2322}
2323
2324/* Stub, nothing needs to be done. */
2325static void vxlan_set_multicast_list(struct net_device *dev)
2326{
2327}
2328
2329static int __vxlan_change_mtu(struct net_device *dev,
2330 struct net_device *lowerdev,
2331 struct vxlan_rdst *dst, int new_mtu, bool strict)
2332{
2333 int max_mtu = IP_MAX_MTU;
2334
2335 if (lowerdev)
2336 max_mtu = lowerdev->mtu;
2337
2338 if (dst->remote_ip.sa.sa_family == AF_INET6)
2339 max_mtu -= VXLAN6_HEADROOM;
2340 else
2341 max_mtu -= VXLAN_HEADROOM;
2342
2343 if (new_mtu < 68)
2344 return -EINVAL;
2345
2346 if (new_mtu > max_mtu) {
2347 if (strict)
2348 return -EINVAL;
2349
2350 new_mtu = max_mtu;
2351 }
2352
2353 dev->mtu = new_mtu;
2354 return 0;
2355}
2356
2357static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2358{
2359 struct vxlan_dev *vxlan = netdev_priv(dev);
2360 struct vxlan_rdst *dst = &vxlan->default_dst;
2361 struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2362 dst->remote_ifindex);
2363 return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2364}
2365
2366static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2367{
2368 struct vxlan_dev *vxlan = netdev_priv(dev);
2369 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2370 __be16 sport, dport;
2371
2372 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2373 vxlan->cfg.port_max, true);
2374 dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2375
2376 if (ip_tunnel_info_af(info) == AF_INET) {
2377 struct rtable *rt;
2378
2379 if (!vxlan->vn4_sock)
2380 return -EINVAL;
2381 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2382 info->key.u.ipv4.dst,
2383 &info->key.u.ipv4.src, NULL, info);
2384 if (IS_ERR(rt))
2385 return PTR_ERR(rt);
2386 ip_rt_put(rt);
2387 } else {
2388#if IS_ENABLED(CONFIG_IPV6)
2389 struct dst_entry *ndst;
2390
2391 if (!vxlan->vn6_sock)
2392 return -EINVAL;
2393 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2394 info->key.label, &info->key.u.ipv6.dst,
2395 &info->key.u.ipv6.src, NULL, info);
2396 if (IS_ERR(ndst))
2397 return PTR_ERR(ndst);
2398 dst_release(ndst);
2399#else /* !CONFIG_IPV6 */
2400 return -EPFNOSUPPORT;
2401#endif
2402 }
2403 info->key.tp_src = sport;
2404 info->key.tp_dst = dport;
2405 return 0;
2406}
2407
2408static const struct net_device_ops vxlan_netdev_ops = {
2409 .ndo_init = vxlan_init,
2410 .ndo_uninit = vxlan_uninit,
2411 .ndo_open = vxlan_open,
2412 .ndo_stop = vxlan_stop,
2413 .ndo_start_xmit = vxlan_xmit,
2414 .ndo_get_stats64 = ip_tunnel_get_stats64,
2415 .ndo_set_rx_mode = vxlan_set_multicast_list,
2416 .ndo_change_mtu = vxlan_change_mtu,
2417 .ndo_validate_addr = eth_validate_addr,
2418 .ndo_set_mac_address = eth_mac_addr,
2419 .ndo_fdb_add = vxlan_fdb_add,
2420 .ndo_fdb_del = vxlan_fdb_delete,
2421 .ndo_fdb_dump = vxlan_fdb_dump,
2422 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2423};
2424
2425/* Info for udev, that this is a virtual tunnel endpoint */
2426static struct device_type vxlan_type = {
2427 .name = "vxlan",
2428};
2429
2430/* Calls the ndo_add_vxlan_port of the caller in order to
2431 * supply the listening VXLAN udp ports. Callers are expected
2432 * to implement the ndo_add_vxlan_port.
2433 */
2434void vxlan_get_rx_port(struct net_device *dev)
2435{
2436 struct vxlan_sock *vs;
2437 struct net *net = dev_net(dev);
2438 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2439 sa_family_t sa_family;
2440 __be16 port;
2441 unsigned int i;
2442
2443 spin_lock(&vn->sock_lock);
2444 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2445 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2446 port = inet_sk(vs->sock->sk)->inet_sport;
2447 sa_family = vxlan_get_sk_family(vs);
2448 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2449 port);
2450 }
2451 }
2452 spin_unlock(&vn->sock_lock);
2453}
2454EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2455
2456/* Initialize the device structure. */
2457static void vxlan_setup(struct net_device *dev)
2458{
2459 struct vxlan_dev *vxlan = netdev_priv(dev);
2460 unsigned int h;
2461
2462 eth_hw_addr_random(dev);
2463 ether_setup(dev);
2464
2465 dev->netdev_ops = &vxlan_netdev_ops;
2466 dev->destructor = free_netdev;
2467 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2468
2469 dev->features |= NETIF_F_LLTX;
2470 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2471 dev->features |= NETIF_F_RXCSUM;
2472 dev->features |= NETIF_F_GSO_SOFTWARE;
2473
2474 dev->vlan_features = dev->features;
2475 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2476 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2477 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2478 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2479 netif_keep_dst(dev);
2480 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2481 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
2482
2483 INIT_LIST_HEAD(&vxlan->next);
2484 spin_lock_init(&vxlan->hash_lock);
2485
2486 init_timer_deferrable(&vxlan->age_timer);
2487 vxlan->age_timer.function = vxlan_cleanup;
2488 vxlan->age_timer.data = (unsigned long) vxlan;
2489
2490 vxlan->cfg.dst_port = htons(vxlan_port);
2491
2492 vxlan->dev = dev;
2493
2494 gro_cells_init(&vxlan->gro_cells, dev);
2495
2496 for (h = 0; h < FDB_HASH_SIZE; ++h)
2497 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2498}
2499
2500static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2501 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2502 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2503 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2504 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2505 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2506 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2507 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2508 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2509 [IFLA_VXLAN_LABEL] = { .type = NLA_U32 },
2510 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2511 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2512 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2513 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2514 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2515 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2516 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2517 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2518 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 },
2519 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2520 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 },
2521 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
2522 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
2523 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2524 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2525 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, },
2526 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG },
2527};
2528
2529static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2530{
2531 if (tb[IFLA_ADDRESS]) {
2532 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2533 pr_debug("invalid link address (not ethernet)\n");
2534 return -EINVAL;
2535 }
2536
2537 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2538 pr_debug("invalid all zero ethernet address\n");
2539 return -EADDRNOTAVAIL;
2540 }
2541 }
2542
2543 if (!data)
2544 return -EINVAL;
2545
2546 if (data[IFLA_VXLAN_ID]) {
2547 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2548 if (id >= VXLAN_VID_MASK)
2549 return -ERANGE;
2550 }
2551
2552 if (data[IFLA_VXLAN_PORT_RANGE]) {
2553 const struct ifla_vxlan_port_range *p
2554 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2555
2556 if (ntohs(p->high) < ntohs(p->low)) {
2557 pr_debug("port range %u .. %u not valid\n",
2558 ntohs(p->low), ntohs(p->high));
2559 return -EINVAL;
2560 }
2561 }
2562
2563 return 0;
2564}
2565
2566static void vxlan_get_drvinfo(struct net_device *netdev,
2567 struct ethtool_drvinfo *drvinfo)
2568{
2569 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2570 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2571}
2572
2573static const struct ethtool_ops vxlan_ethtool_ops = {
2574 .get_drvinfo = vxlan_get_drvinfo,
2575 .get_link = ethtool_op_get_link,
2576};
2577
2578static void vxlan_del_work(struct work_struct *work)
2579{
2580 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2581 udp_tunnel_sock_release(vs->sock);
2582 kfree_rcu(vs, rcu);
2583}
2584
2585static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2586 __be16 port, u32 flags)
2587{
2588 struct socket *sock;
2589 struct udp_port_cfg udp_conf;
2590 int err;
2591
2592 memset(&udp_conf, 0, sizeof(udp_conf));
2593
2594 if (ipv6) {
2595 udp_conf.family = AF_INET6;
2596 udp_conf.use_udp6_rx_checksums =
2597 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2598 udp_conf.ipv6_v6only = 1;
2599 } else {
2600 udp_conf.family = AF_INET;
2601 }
2602
2603 udp_conf.local_udp_port = port;
2604
2605 /* Open UDP socket */
2606 err = udp_sock_create(net, &udp_conf, &sock);
2607 if (err < 0)
2608 return ERR_PTR(err);
2609
2610 return sock;
2611}
2612
2613/* Create new listen socket if needed */
2614static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2615 __be16 port, u32 flags)
2616{
2617 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2618 struct vxlan_sock *vs;
2619 struct socket *sock;
2620 unsigned int h;
2621 struct udp_tunnel_sock_cfg tunnel_cfg;
2622
2623 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2624 if (!vs)
2625 return ERR_PTR(-ENOMEM);
2626
2627 for (h = 0; h < VNI_HASH_SIZE; ++h)
2628 INIT_HLIST_HEAD(&vs->vni_list[h]);
2629
2630 INIT_WORK(&vs->del_work, vxlan_del_work);
2631
2632 sock = vxlan_create_sock(net, ipv6, port, flags);
2633 if (IS_ERR(sock)) {
2634 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2635 PTR_ERR(sock));
2636 kfree(vs);
2637 return ERR_CAST(sock);
2638 }
2639
2640 vs->sock = sock;
2641 atomic_set(&vs->refcnt, 1);
2642 vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2643
2644 /* Initialize the vxlan udp offloads structure */
2645 vs->udp_offloads.port = port;
2646 vs->udp_offloads.callbacks.gro_receive = vxlan_gro_receive;
2647 vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2648
2649 spin_lock(&vn->sock_lock);
2650 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2651 vxlan_notify_add_rx_port(vs);
2652 spin_unlock(&vn->sock_lock);
2653
2654 /* Mark socket as an encapsulation socket. */
2655 tunnel_cfg.sk_user_data = vs;
2656 tunnel_cfg.encap_type = 1;
2657 tunnel_cfg.encap_rcv = vxlan_rcv;
2658 tunnel_cfg.encap_destroy = NULL;
2659
2660 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2661
2662 return vs;
2663}
2664
2665static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2666{
2667 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2668 struct vxlan_sock *vs = NULL;
2669
2670 if (!vxlan->cfg.no_share) {
2671 spin_lock(&vn->sock_lock);
2672 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2673 vxlan->cfg.dst_port, vxlan->flags);
2674 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2675 spin_unlock(&vn->sock_lock);
2676 return -EBUSY;
2677 }
2678 spin_unlock(&vn->sock_lock);
2679 }
2680 if (!vs)
2681 vs = vxlan_socket_create(vxlan->net, ipv6,
2682 vxlan->cfg.dst_port, vxlan->flags);
2683 if (IS_ERR(vs))
2684 return PTR_ERR(vs);
2685#if IS_ENABLED(CONFIG_IPV6)
2686 if (ipv6)
2687 vxlan->vn6_sock = vs;
2688 else
2689#endif
2690 vxlan->vn4_sock = vs;
2691 vxlan_vs_add_dev(vs, vxlan);
2692 return 0;
2693}
2694
2695static int vxlan_sock_add(struct vxlan_dev *vxlan)
2696{
2697 bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2698 bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2699 int ret = 0;
2700
2701 vxlan->vn4_sock = NULL;
2702#if IS_ENABLED(CONFIG_IPV6)
2703 vxlan->vn6_sock = NULL;
2704 if (ipv6 || metadata)
2705 ret = __vxlan_sock_add(vxlan, true);
2706#endif
2707 if (!ret && (!ipv6 || metadata))
2708 ret = __vxlan_sock_add(vxlan, false);
2709 if (ret < 0)
2710 vxlan_sock_release(vxlan);
2711 return ret;
2712}
2713
2714static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2715 struct vxlan_config *conf)
2716{
2717 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2718 struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2719 struct vxlan_rdst *dst = &vxlan->default_dst;
2720 unsigned short needed_headroom = ETH_HLEN;
2721 int err;
2722 bool use_ipv6 = false;
2723 __be16 default_port = vxlan->cfg.dst_port;
2724 struct net_device *lowerdev = NULL;
2725
2726 vxlan->net = src_net;
2727
2728 dst->remote_vni = conf->vni;
2729
2730 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2731
2732 /* Unless IPv6 is explicitly requested, assume IPv4 */
2733 if (!dst->remote_ip.sa.sa_family)
2734 dst->remote_ip.sa.sa_family = AF_INET;
2735
2736 if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2737 vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2738 if (!IS_ENABLED(CONFIG_IPV6))
2739 return -EPFNOSUPPORT;
2740 use_ipv6 = true;
2741 vxlan->flags |= VXLAN_F_IPV6;
2742 }
2743
2744 if (conf->label && !use_ipv6) {
2745 pr_info("label only supported in use with IPv6\n");
2746 return -EINVAL;
2747 }
2748
2749 if (conf->remote_ifindex) {
2750 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2751 dst->remote_ifindex = conf->remote_ifindex;
2752
2753 if (!lowerdev) {
2754 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2755 return -ENODEV;
2756 }
2757
2758#if IS_ENABLED(CONFIG_IPV6)
2759 if (use_ipv6) {
2760 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2761 if (idev && idev->cnf.disable_ipv6) {
2762 pr_info("IPv6 is disabled via sysctl\n");
2763 return -EPERM;
2764 }
2765 }
2766#endif
2767
2768 if (!conf->mtu)
2769 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2770
2771 needed_headroom = lowerdev->hard_header_len;
2772 }
2773
2774 if (conf->mtu) {
2775 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2776 if (err)
2777 return err;
2778 }
2779
2780 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2781 needed_headroom += VXLAN6_HEADROOM;
2782 else
2783 needed_headroom += VXLAN_HEADROOM;
2784 dev->needed_headroom = needed_headroom;
2785
2786 memcpy(&vxlan->cfg, conf, sizeof(*conf));
2787 if (!vxlan->cfg.dst_port)
2788 vxlan->cfg.dst_port = default_port;
2789 vxlan->flags |= conf->flags;
2790
2791 if (!vxlan->cfg.age_interval)
2792 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2793
2794 list_for_each_entry(tmp, &vn->vxlan_list, next) {
2795 if (tmp->cfg.vni == conf->vni &&
2796 (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2797 tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2798 tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2799 (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2800 (vxlan->flags & VXLAN_F_RCV_FLAGS))
2801 return -EEXIST;
2802 }
2803
2804 dev->ethtool_ops = &vxlan_ethtool_ops;
2805
2806 /* create an fdb entry for a valid default destination */
2807 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2808 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2809 &vxlan->default_dst.remote_ip,
2810 NUD_REACHABLE|NUD_PERMANENT,
2811 NLM_F_EXCL|NLM_F_CREATE,
2812 vxlan->cfg.dst_port,
2813 vxlan->default_dst.remote_vni,
2814 vxlan->default_dst.remote_ifindex,
2815 NTF_SELF);
2816 if (err)
2817 return err;
2818 }
2819
2820 err = register_netdevice(dev);
2821 if (err) {
2822 vxlan_fdb_delete_default(vxlan);
2823 return err;
2824 }
2825
2826 list_add(&vxlan->next, &vn->vxlan_list);
2827
2828 return 0;
2829}
2830
2831struct net_device *vxlan_dev_create(struct net *net, const char *name,
2832 u8 name_assign_type, struct vxlan_config *conf)
2833{
2834 struct nlattr *tb[IFLA_MAX+1];
2835 struct net_device *dev;
2836 int err;
2837
2838 memset(&tb, 0, sizeof(tb));
2839
2840 dev = rtnl_create_link(net, name, name_assign_type,
2841 &vxlan_link_ops, tb);
2842 if (IS_ERR(dev))
2843 return dev;
2844
2845 err = vxlan_dev_configure(net, dev, conf);
2846 if (err < 0) {
2847 free_netdev(dev);
2848 return ERR_PTR(err);
2849 }
2850
2851 return dev;
2852}
2853EXPORT_SYMBOL_GPL(vxlan_dev_create);
2854
2855static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2856 struct nlattr *tb[], struct nlattr *data[])
2857{
2858 struct vxlan_config conf;
2859 int err;
2860
2861 memset(&conf, 0, sizeof(conf));
2862
2863 if (data[IFLA_VXLAN_ID])
2864 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2865
2866 if (data[IFLA_VXLAN_GROUP]) {
2867 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2868 } else if (data[IFLA_VXLAN_GROUP6]) {
2869 if (!IS_ENABLED(CONFIG_IPV6))
2870 return -EPFNOSUPPORT;
2871
2872 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2873 conf.remote_ip.sa.sa_family = AF_INET6;
2874 }
2875
2876 if (data[IFLA_VXLAN_LOCAL]) {
2877 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2878 conf.saddr.sa.sa_family = AF_INET;
2879 } else if (data[IFLA_VXLAN_LOCAL6]) {
2880 if (!IS_ENABLED(CONFIG_IPV6))
2881 return -EPFNOSUPPORT;
2882
2883 /* TODO: respect scope id */
2884 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2885 conf.saddr.sa.sa_family = AF_INET6;
2886 }
2887
2888 if (data[IFLA_VXLAN_LINK])
2889 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2890
2891 if (data[IFLA_VXLAN_TOS])
2892 conf.tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2893
2894 if (data[IFLA_VXLAN_TTL])
2895 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2896
2897 if (data[IFLA_VXLAN_LABEL])
2898 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
2899 IPV6_FLOWLABEL_MASK;
2900
2901 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2902 conf.flags |= VXLAN_F_LEARN;
2903
2904 if (data[IFLA_VXLAN_AGEING])
2905 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2906
2907 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2908 conf.flags |= VXLAN_F_PROXY;
2909
2910 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2911 conf.flags |= VXLAN_F_RSC;
2912
2913 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2914 conf.flags |= VXLAN_F_L2MISS;
2915
2916 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2917 conf.flags |= VXLAN_F_L3MISS;
2918
2919 if (data[IFLA_VXLAN_LIMIT])
2920 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2921
2922 if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2923 nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2924 conf.flags |= VXLAN_F_COLLECT_METADATA;
2925
2926 if (data[IFLA_VXLAN_PORT_RANGE]) {
2927 const struct ifla_vxlan_port_range *p
2928 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2929 conf.port_min = ntohs(p->low);
2930 conf.port_max = ntohs(p->high);
2931 }
2932
2933 if (data[IFLA_VXLAN_PORT])
2934 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2935
2936 if (data[IFLA_VXLAN_UDP_CSUM] &&
2937 !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2938 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
2939
2940 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2941 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2942 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2943
2944 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2945 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2946 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2947
2948 if (data[IFLA_VXLAN_REMCSUM_TX] &&
2949 nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2950 conf.flags |= VXLAN_F_REMCSUM_TX;
2951
2952 if (data[IFLA_VXLAN_REMCSUM_RX] &&
2953 nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2954 conf.flags |= VXLAN_F_REMCSUM_RX;
2955
2956 if (data[IFLA_VXLAN_GBP])
2957 conf.flags |= VXLAN_F_GBP;
2958
2959 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2960 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2961
2962 err = vxlan_dev_configure(src_net, dev, &conf);
2963 switch (err) {
2964 case -ENODEV:
2965 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
2966 break;
2967
2968 case -EPERM:
2969 pr_info("IPv6 is disabled via sysctl\n");
2970 break;
2971
2972 case -EEXIST:
2973 pr_info("duplicate VNI %u\n", be32_to_cpu(conf.vni));
2974 break;
2975 }
2976
2977 return err;
2978}
2979
2980static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2981{
2982 struct vxlan_dev *vxlan = netdev_priv(dev);
2983 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2984
2985 spin_lock(&vn->sock_lock);
2986 if (!hlist_unhashed(&vxlan->hlist))
2987 hlist_del_rcu(&vxlan->hlist);
2988 spin_unlock(&vn->sock_lock);
2989
2990 gro_cells_destroy(&vxlan->gro_cells);
2991 list_del(&vxlan->next);
2992 unregister_netdevice_queue(dev, head);
2993}
2994
2995static size_t vxlan_get_size(const struct net_device *dev)
2996{
2997
2998 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
2999 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3000 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3001 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3002 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
3003 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
3004 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3005 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
3006 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
3007 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
3008 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
3009 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
3010 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */
3011 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3012 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3013 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3014 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3015 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3016 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3017 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3018 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3019 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3020 0;
3021}
3022
3023static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3024{
3025 const struct vxlan_dev *vxlan = netdev_priv(dev);
3026 const struct vxlan_rdst *dst = &vxlan->default_dst;
3027 struct ifla_vxlan_port_range ports = {
3028 .low = htons(vxlan->cfg.port_min),
3029 .high = htons(vxlan->cfg.port_max),
3030 };
3031
3032 if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3033 goto nla_put_failure;
3034
3035 if (!vxlan_addr_any(&dst->remote_ip)) {
3036 if (dst->remote_ip.sa.sa_family == AF_INET) {
3037 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3038 dst->remote_ip.sin.sin_addr.s_addr))
3039 goto nla_put_failure;
3040#if IS_ENABLED(CONFIG_IPV6)
3041 } else {
3042 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3043 &dst->remote_ip.sin6.sin6_addr))
3044 goto nla_put_failure;
3045#endif
3046 }
3047 }
3048
3049 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3050 goto nla_put_failure;
3051
3052 if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3053 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3054 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3055 vxlan->cfg.saddr.sin.sin_addr.s_addr))
3056 goto nla_put_failure;
3057#if IS_ENABLED(CONFIG_IPV6)
3058 } else {
3059 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3060 &vxlan->cfg.saddr.sin6.sin6_addr))
3061 goto nla_put_failure;
3062#endif
3063 }
3064 }
3065
3066 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3067 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3068 nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3069 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3070 !!(vxlan->flags & VXLAN_F_LEARN)) ||
3071 nla_put_u8(skb, IFLA_VXLAN_PROXY,
3072 !!(vxlan->flags & VXLAN_F_PROXY)) ||
3073 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3074 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3075 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3076 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3077 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3078 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3079 !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3080 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3081 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3082 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3083 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3084 !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3085 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3086 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3087 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3088 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3089 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3090 !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3091 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3092 !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3093 goto nla_put_failure;
3094
3095 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3096 goto nla_put_failure;
3097
3098 if (vxlan->flags & VXLAN_F_GBP &&
3099 nla_put_flag(skb, IFLA_VXLAN_GBP))
3100 goto nla_put_failure;
3101
3102 if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3103 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3104 goto nla_put_failure;
3105
3106 return 0;
3107
3108nla_put_failure:
3109 return -EMSGSIZE;
3110}
3111
3112static struct net *vxlan_get_link_net(const struct net_device *dev)
3113{
3114 struct vxlan_dev *vxlan = netdev_priv(dev);
3115
3116 return vxlan->net;
3117}
3118
3119static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3120 .kind = "vxlan",
3121 .maxtype = IFLA_VXLAN_MAX,
3122 .policy = vxlan_policy,
3123 .priv_size = sizeof(struct vxlan_dev),
3124 .setup = vxlan_setup,
3125 .validate = vxlan_validate,
3126 .newlink = vxlan_newlink,
3127 .dellink = vxlan_dellink,
3128 .get_size = vxlan_get_size,
3129 .fill_info = vxlan_fill_info,
3130 .get_link_net = vxlan_get_link_net,
3131};
3132
3133static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3134 struct net_device *dev)
3135{
3136 struct vxlan_dev *vxlan, *next;
3137 LIST_HEAD(list_kill);
3138
3139 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3140 struct vxlan_rdst *dst = &vxlan->default_dst;
3141
3142 /* In case we created vxlan device with carrier
3143 * and we loose the carrier due to module unload
3144 * we also need to remove vxlan device. In other
3145 * cases, it's not necessary and remote_ifindex
3146 * is 0 here, so no matches.
3147 */
3148 if (dst->remote_ifindex == dev->ifindex)
3149 vxlan_dellink(vxlan->dev, &list_kill);
3150 }
3151
3152 unregister_netdevice_many(&list_kill);
3153}
3154
3155static int vxlan_lowerdev_event(struct notifier_block *unused,
3156 unsigned long event, void *ptr)
3157{
3158 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3159 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3160
3161 if (event == NETDEV_UNREGISTER)
3162 vxlan_handle_lowerdev_unregister(vn, dev);
3163
3164 return NOTIFY_DONE;
3165}
3166
3167static struct notifier_block vxlan_notifier_block __read_mostly = {
3168 .notifier_call = vxlan_lowerdev_event,
3169};
3170
3171static __net_init int vxlan_init_net(struct net *net)
3172{
3173 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3174 unsigned int h;
3175
3176 INIT_LIST_HEAD(&vn->vxlan_list);
3177 spin_lock_init(&vn->sock_lock);
3178
3179 for (h = 0; h < PORT_HASH_SIZE; ++h)
3180 INIT_HLIST_HEAD(&vn->sock_list[h]);
3181
3182 return 0;
3183}
3184
3185static void __net_exit vxlan_exit_net(struct net *net)
3186{
3187 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3188 struct vxlan_dev *vxlan, *next;
3189 struct net_device *dev, *aux;
3190 LIST_HEAD(list);
3191
3192 rtnl_lock();
3193 for_each_netdev_safe(net, dev, aux)
3194 if (dev->rtnl_link_ops == &vxlan_link_ops)
3195 unregister_netdevice_queue(dev, &list);
3196
3197 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3198 /* If vxlan->dev is in the same netns, it has already been added
3199 * to the list by the previous loop.
3200 */
3201 if (!net_eq(dev_net(vxlan->dev), net)) {
3202 gro_cells_destroy(&vxlan->gro_cells);
3203 unregister_netdevice_queue(vxlan->dev, &list);
3204 }
3205 }
3206
3207 unregister_netdevice_many(&list);
3208 rtnl_unlock();
3209}
3210
3211static struct pernet_operations vxlan_net_ops = {
3212 .init = vxlan_init_net,
3213 .exit = vxlan_exit_net,
3214 .id = &vxlan_net_id,
3215 .size = sizeof(struct vxlan_net),
3216};
3217
3218static int __init vxlan_init_module(void)
3219{
3220 int rc;
3221
3222 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3223 if (!vxlan_wq)
3224 return -ENOMEM;
3225
3226 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3227
3228 rc = register_pernet_subsys(&vxlan_net_ops);
3229 if (rc)
3230 goto out1;
3231
3232 rc = register_netdevice_notifier(&vxlan_notifier_block);
3233 if (rc)
3234 goto out2;
3235
3236 rc = rtnl_link_register(&vxlan_link_ops);
3237 if (rc)
3238 goto out3;
3239
3240 return 0;
3241out3:
3242 unregister_netdevice_notifier(&vxlan_notifier_block);
3243out2:
3244 unregister_pernet_subsys(&vxlan_net_ops);
3245out1:
3246 destroy_workqueue(vxlan_wq);
3247 return rc;
3248}
3249late_initcall(vxlan_init_module);
3250
3251static void __exit vxlan_cleanup_module(void)
3252{
3253 rtnl_link_unregister(&vxlan_link_ops);
3254 unregister_netdevice_notifier(&vxlan_notifier_block);
3255 destroy_workqueue(vxlan_wq);
3256 unregister_pernet_subsys(&vxlan_net_ops);
3257 /* rcu_barrier() is called by netns */
3258}
3259module_exit(vxlan_cleanup_module);
3260
3261MODULE_LICENSE("GPL");
3262MODULE_VERSION(VXLAN_VERSION);
3263MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3264MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3265MODULE_ALIAS_RTNL_LINK("vxlan");
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * VXLAN: Virtual eXtensible Local Area Network
4 *
5 * Copyright (c) 2012-2013 Vyatta Inc.
6 */
7
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10#include <linux/kernel.h>
11#include <linux/module.h>
12#include <linux/errno.h>
13#include <linux/slab.h>
14#include <linux/udp.h>
15#include <linux/igmp.h>
16#include <linux/if_ether.h>
17#include <linux/ethtool.h>
18#include <net/arp.h>
19#include <net/ndisc.h>
20#include <net/ipv6_stubs.h>
21#include <net/ip.h>
22#include <net/icmp.h>
23#include <net/rtnetlink.h>
24#include <net/inet_ecn.h>
25#include <net/net_namespace.h>
26#include <net/netns/generic.h>
27#include <net/tun_proto.h>
28#include <net/vxlan.h>
29#include <net/nexthop.h>
30
31#if IS_ENABLED(CONFIG_IPV6)
32#include <net/ip6_tunnel.h>
33#include <net/ip6_checksum.h>
34#endif
35
36#define VXLAN_VERSION "0.1"
37
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 */
42
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.
46 */
47static unsigned short vxlan_port __read_mostly = 8472;
48module_param_named(udp_port, vxlan_port, ushort, 0444);
49MODULE_PARM_DESC(udp_port, "Destination UDP port");
50
51static bool log_ecn_error = true;
52module_param(log_ecn_error, bool, 0644);
53MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
54
55static unsigned int vxlan_net_id;
56static struct rtnl_link_ops vxlan_link_ops;
57
58static const u8 all_zeros_mac[ETH_ALEN + 2];
59
60static int vxlan_sock_add(struct vxlan_dev *vxlan);
61
62static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
63
64/* per-network namespace private data for this module */
65struct vxlan_net {
66 struct list_head vxlan_list;
67 struct hlist_head sock_list[PORT_HASH_SIZE];
68 spinlock_t sock_lock;
69};
70
71/* Forwarding table entry */
72struct vxlan_fdb {
73 struct hlist_node hlist; /* linked list of entries */
74 struct rcu_head rcu;
75 unsigned long updated; /* jiffies */
76 unsigned long used;
77 struct list_head remotes;
78 u8 eth_addr[ETH_ALEN];
79 u16 state; /* see ndm_state */
80 __be32 vni;
81 u16 flags; /* see ndm_flags and below */
82 struct list_head nh_list;
83 struct nexthop __rcu *nh;
84 struct vxlan_dev __rcu *vdev;
85};
86
87#define NTF_VXLAN_ADDED_BY_USER 0x100
88
89/* salt for hash table */
90static u32 vxlan_salt __read_mostly;
91
92static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
93{
94 return vs->flags & VXLAN_F_COLLECT_METADATA ||
95 ip_tunnel_collect_metadata();
96}
97
98#if IS_ENABLED(CONFIG_IPV6)
99static inline
100bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
101{
102 if (a->sa.sa_family != b->sa.sa_family)
103 return false;
104 if (a->sa.sa_family == AF_INET6)
105 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
106 else
107 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
108}
109
110static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
111{
112 if (nla_len(nla) >= sizeof(struct in6_addr)) {
113 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
114 ip->sa.sa_family = AF_INET6;
115 return 0;
116 } else if (nla_len(nla) >= sizeof(__be32)) {
117 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
118 ip->sa.sa_family = AF_INET;
119 return 0;
120 } else {
121 return -EAFNOSUPPORT;
122 }
123}
124
125static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
126 const union vxlan_addr *ip)
127{
128 if (ip->sa.sa_family == AF_INET6)
129 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
130 else
131 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
132}
133
134#else /* !CONFIG_IPV6 */
135
136static inline
137bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
138{
139 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
140}
141
142static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
143{
144 if (nla_len(nla) >= sizeof(struct in6_addr)) {
145 return -EAFNOSUPPORT;
146 } else if (nla_len(nla) >= sizeof(__be32)) {
147 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
148 ip->sa.sa_family = AF_INET;
149 return 0;
150 } else {
151 return -EAFNOSUPPORT;
152 }
153}
154
155static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
156 const union vxlan_addr *ip)
157{
158 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
159}
160#endif
161
162/* Virtual Network hash table head */
163static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
164{
165 return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
166}
167
168/* Socket hash table head */
169static inline struct hlist_head *vs_head(struct net *net, __be16 port)
170{
171 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
172
173 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
174}
175
176/* First remote destination for a forwarding entry.
177 * Guaranteed to be non-NULL because remotes are never deleted.
178 */
179static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
180{
181 if (rcu_access_pointer(fdb->nh))
182 return NULL;
183 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
184}
185
186static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
187{
188 if (rcu_access_pointer(fdb->nh))
189 return NULL;
190 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
191}
192
193/* Find VXLAN socket based on network namespace, address family and UDP port
194 * and enabled unshareable flags.
195 */
196static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
197 __be16 port, u32 flags, int ifindex)
198{
199 struct vxlan_sock *vs;
200
201 flags &= VXLAN_F_RCV_FLAGS;
202
203 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
204 if (inet_sk(vs->sock->sk)->inet_sport == port &&
205 vxlan_get_sk_family(vs) == family &&
206 vs->flags == flags &&
207 vs->sock->sk->sk_bound_dev_if == ifindex)
208 return vs;
209 }
210 return NULL;
211}
212
213static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex,
214 __be32 vni)
215{
216 struct vxlan_dev_node *node;
217
218 /* For flow based devices, map all packets to VNI 0 */
219 if (vs->flags & VXLAN_F_COLLECT_METADATA)
220 vni = 0;
221
222 hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
223 if (node->vxlan->default_dst.remote_vni != vni)
224 continue;
225
226 if (IS_ENABLED(CONFIG_IPV6)) {
227 const struct vxlan_config *cfg = &node->vxlan->cfg;
228
229 if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
230 cfg->remote_ifindex != ifindex)
231 continue;
232 }
233
234 return node->vxlan;
235 }
236
237 return NULL;
238}
239
240/* Look up VNI in a per net namespace table */
241static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
242 __be32 vni, sa_family_t family,
243 __be16 port, u32 flags)
244{
245 struct vxlan_sock *vs;
246
247 vs = vxlan_find_sock(net, family, port, flags, ifindex);
248 if (!vs)
249 return NULL;
250
251 return vxlan_vs_find_vni(vs, ifindex, vni);
252}
253
254/* Fill in neighbour message in skbuff. */
255static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
256 const struct vxlan_fdb *fdb,
257 u32 portid, u32 seq, int type, unsigned int flags,
258 const struct vxlan_rdst *rdst)
259{
260 unsigned long now = jiffies;
261 struct nda_cacheinfo ci;
262 bool send_ip, send_eth;
263 struct nlmsghdr *nlh;
264 struct nexthop *nh;
265 struct ndmsg *ndm;
266 int nh_family;
267 u32 nh_id;
268
269 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
270 if (nlh == NULL)
271 return -EMSGSIZE;
272
273 ndm = nlmsg_data(nlh);
274 memset(ndm, 0, sizeof(*ndm));
275
276 send_eth = send_ip = true;
277
278 rcu_read_lock();
279 nh = rcu_dereference(fdb->nh);
280 if (nh) {
281 nh_family = nexthop_get_family(nh);
282 nh_id = nh->id;
283 }
284 rcu_read_unlock();
285
286 if (type == RTM_GETNEIGH) {
287 if (rdst) {
288 send_ip = !vxlan_addr_any(&rdst->remote_ip);
289 ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
290 } else if (nh) {
291 ndm->ndm_family = nh_family;
292 }
293 send_eth = !is_zero_ether_addr(fdb->eth_addr);
294 } else
295 ndm->ndm_family = AF_BRIDGE;
296 ndm->ndm_state = fdb->state;
297 ndm->ndm_ifindex = vxlan->dev->ifindex;
298 ndm->ndm_flags = fdb->flags;
299 if (rdst && rdst->offloaded)
300 ndm->ndm_flags |= NTF_OFFLOADED;
301 ndm->ndm_type = RTN_UNICAST;
302
303 if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
304 nla_put_s32(skb, NDA_LINK_NETNSID,
305 peernet2id(dev_net(vxlan->dev), vxlan->net)))
306 goto nla_put_failure;
307
308 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
309 goto nla_put_failure;
310 if (nh) {
311 if (nla_put_u32(skb, NDA_NH_ID, nh_id))
312 goto nla_put_failure;
313 } else if (rdst) {
314 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST,
315 &rdst->remote_ip))
316 goto nla_put_failure;
317
318 if (rdst->remote_port &&
319 rdst->remote_port != vxlan->cfg.dst_port &&
320 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
321 goto nla_put_failure;
322 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
323 nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
324 goto nla_put_failure;
325 if (rdst->remote_ifindex &&
326 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
327 goto nla_put_failure;
328 }
329
330 if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
331 nla_put_u32(skb, NDA_SRC_VNI,
332 be32_to_cpu(fdb->vni)))
333 goto nla_put_failure;
334
335 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
336 ci.ndm_confirmed = 0;
337 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
338 ci.ndm_refcnt = 0;
339
340 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
341 goto nla_put_failure;
342
343 nlmsg_end(skb, nlh);
344 return 0;
345
346nla_put_failure:
347 nlmsg_cancel(skb, nlh);
348 return -EMSGSIZE;
349}
350
351static inline size_t vxlan_nlmsg_size(void)
352{
353 return NLMSG_ALIGN(sizeof(struct ndmsg))
354 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
355 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
356 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
357 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
358 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
359 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
360 + nla_total_size(sizeof(struct nda_cacheinfo));
361}
362
363static void __vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
364 struct vxlan_rdst *rd, int type)
365{
366 struct net *net = dev_net(vxlan->dev);
367 struct sk_buff *skb;
368 int err = -ENOBUFS;
369
370 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
371 if (skb == NULL)
372 goto errout;
373
374 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
375 if (err < 0) {
376 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
377 WARN_ON(err == -EMSGSIZE);
378 kfree_skb(skb);
379 goto errout;
380 }
381
382 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
383 return;
384errout:
385 if (err < 0)
386 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
387}
388
389static void vxlan_fdb_switchdev_notifier_info(const struct vxlan_dev *vxlan,
390 const struct vxlan_fdb *fdb,
391 const struct vxlan_rdst *rd,
392 struct netlink_ext_ack *extack,
393 struct switchdev_notifier_vxlan_fdb_info *fdb_info)
394{
395 fdb_info->info.dev = vxlan->dev;
396 fdb_info->info.extack = extack;
397 fdb_info->remote_ip = rd->remote_ip;
398 fdb_info->remote_port = rd->remote_port;
399 fdb_info->remote_vni = rd->remote_vni;
400 fdb_info->remote_ifindex = rd->remote_ifindex;
401 memcpy(fdb_info->eth_addr, fdb->eth_addr, ETH_ALEN);
402 fdb_info->vni = fdb->vni;
403 fdb_info->offloaded = rd->offloaded;
404 fdb_info->added_by_user = fdb->flags & NTF_VXLAN_ADDED_BY_USER;
405}
406
407static int vxlan_fdb_switchdev_call_notifiers(struct vxlan_dev *vxlan,
408 struct vxlan_fdb *fdb,
409 struct vxlan_rdst *rd,
410 bool adding,
411 struct netlink_ext_ack *extack)
412{
413 struct switchdev_notifier_vxlan_fdb_info info;
414 enum switchdev_notifier_type notifier_type;
415 int ret;
416
417 if (WARN_ON(!rd))
418 return 0;
419
420 notifier_type = adding ? SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE
421 : SWITCHDEV_VXLAN_FDB_DEL_TO_DEVICE;
422 vxlan_fdb_switchdev_notifier_info(vxlan, fdb, rd, NULL, &info);
423 ret = call_switchdev_notifiers(notifier_type, vxlan->dev,
424 &info.info, extack);
425 return notifier_to_errno(ret);
426}
427
428static int vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
429 struct vxlan_rdst *rd, int type, bool swdev_notify,
430 struct netlink_ext_ack *extack)
431{
432 int err;
433
434 if (swdev_notify && rd) {
435 switch (type) {
436 case RTM_NEWNEIGH:
437 err = vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
438 true, extack);
439 if (err)
440 return err;
441 break;
442 case RTM_DELNEIGH:
443 vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
444 false, extack);
445 break;
446 }
447 }
448
449 __vxlan_fdb_notify(vxlan, fdb, rd, type);
450 return 0;
451}
452
453static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
454{
455 struct vxlan_dev *vxlan = netdev_priv(dev);
456 struct vxlan_fdb f = {
457 .state = NUD_STALE,
458 };
459 struct vxlan_rdst remote = {
460 .remote_ip = *ipa, /* goes to NDA_DST */
461 .remote_vni = cpu_to_be32(VXLAN_N_VID),
462 };
463
464 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
465}
466
467static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
468{
469 struct vxlan_fdb f = {
470 .state = NUD_STALE,
471 };
472 struct vxlan_rdst remote = { };
473
474 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
475
476 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
477}
478
479/* Hash Ethernet address */
480static u32 eth_hash(const unsigned char *addr)
481{
482 u64 value = get_unaligned((u64 *)addr);
483
484 /* only want 6 bytes */
485#ifdef __BIG_ENDIAN
486 value >>= 16;
487#else
488 value <<= 16;
489#endif
490 return hash_64(value, FDB_HASH_BITS);
491}
492
493static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
494{
495 /* use 1 byte of OUI and 3 bytes of NIC */
496 u32 key = get_unaligned((u32 *)(addr + 2));
497
498 return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
499}
500
501static u32 fdb_head_index(struct vxlan_dev *vxlan, const u8 *mac, __be32 vni)
502{
503 if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
504 return eth_vni_hash(mac, vni);
505 else
506 return eth_hash(mac);
507}
508
509/* Hash chain to use given mac address */
510static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
511 const u8 *mac, __be32 vni)
512{
513 return &vxlan->fdb_head[fdb_head_index(vxlan, mac, vni)];
514}
515
516/* Look up Ethernet address in forwarding table */
517static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
518 const u8 *mac, __be32 vni)
519{
520 struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
521 struct vxlan_fdb *f;
522
523 hlist_for_each_entry_rcu(f, head, hlist) {
524 if (ether_addr_equal(mac, f->eth_addr)) {
525 if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
526 if (vni == f->vni)
527 return f;
528 } else {
529 return f;
530 }
531 }
532 }
533
534 return NULL;
535}
536
537static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
538 const u8 *mac, __be32 vni)
539{
540 struct vxlan_fdb *f;
541
542 f = __vxlan_find_mac(vxlan, mac, vni);
543 if (f && f->used != jiffies)
544 f->used = jiffies;
545
546 return f;
547}
548
549/* caller should hold vxlan->hash_lock */
550static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
551 union vxlan_addr *ip, __be16 port,
552 __be32 vni, __u32 ifindex)
553{
554 struct vxlan_rdst *rd;
555
556 list_for_each_entry(rd, &f->remotes, list) {
557 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
558 rd->remote_port == port &&
559 rd->remote_vni == vni &&
560 rd->remote_ifindex == ifindex)
561 return rd;
562 }
563
564 return NULL;
565}
566
567int vxlan_fdb_find_uc(struct net_device *dev, const u8 *mac, __be32 vni,
568 struct switchdev_notifier_vxlan_fdb_info *fdb_info)
569{
570 struct vxlan_dev *vxlan = netdev_priv(dev);
571 u8 eth_addr[ETH_ALEN + 2] = { 0 };
572 struct vxlan_rdst *rdst;
573 struct vxlan_fdb *f;
574 int rc = 0;
575
576 if (is_multicast_ether_addr(mac) ||
577 is_zero_ether_addr(mac))
578 return -EINVAL;
579
580 ether_addr_copy(eth_addr, mac);
581
582 rcu_read_lock();
583
584 f = __vxlan_find_mac(vxlan, eth_addr, vni);
585 if (!f) {
586 rc = -ENOENT;
587 goto out;
588 }
589
590 rdst = first_remote_rcu(f);
591 vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, NULL, fdb_info);
592
593out:
594 rcu_read_unlock();
595 return rc;
596}
597EXPORT_SYMBOL_GPL(vxlan_fdb_find_uc);
598
599static int vxlan_fdb_notify_one(struct notifier_block *nb,
600 const struct vxlan_dev *vxlan,
601 const struct vxlan_fdb *f,
602 const struct vxlan_rdst *rdst,
603 struct netlink_ext_ack *extack)
604{
605 struct switchdev_notifier_vxlan_fdb_info fdb_info;
606 int rc;
607
608 vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, extack, &fdb_info);
609 rc = nb->notifier_call(nb, SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE,
610 &fdb_info);
611 return notifier_to_errno(rc);
612}
613
614int vxlan_fdb_replay(const struct net_device *dev, __be32 vni,
615 struct notifier_block *nb,
616 struct netlink_ext_ack *extack)
617{
618 struct vxlan_dev *vxlan;
619 struct vxlan_rdst *rdst;
620 struct vxlan_fdb *f;
621 unsigned int h;
622 int rc = 0;
623
624 if (!netif_is_vxlan(dev))
625 return -EINVAL;
626 vxlan = netdev_priv(dev);
627
628 for (h = 0; h < FDB_HASH_SIZE; ++h) {
629 spin_lock_bh(&vxlan->hash_lock[h]);
630 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) {
631 if (f->vni == vni) {
632 list_for_each_entry(rdst, &f->remotes, list) {
633 rc = vxlan_fdb_notify_one(nb, vxlan,
634 f, rdst,
635 extack);
636 if (rc)
637 goto unlock;
638 }
639 }
640 }
641 spin_unlock_bh(&vxlan->hash_lock[h]);
642 }
643 return 0;
644
645unlock:
646 spin_unlock_bh(&vxlan->hash_lock[h]);
647 return rc;
648}
649EXPORT_SYMBOL_GPL(vxlan_fdb_replay);
650
651void vxlan_fdb_clear_offload(const struct net_device *dev, __be32 vni)
652{
653 struct vxlan_dev *vxlan;
654 struct vxlan_rdst *rdst;
655 struct vxlan_fdb *f;
656 unsigned int h;
657
658 if (!netif_is_vxlan(dev))
659 return;
660 vxlan = netdev_priv(dev);
661
662 for (h = 0; h < FDB_HASH_SIZE; ++h) {
663 spin_lock_bh(&vxlan->hash_lock[h]);
664 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist)
665 if (f->vni == vni)
666 list_for_each_entry(rdst, &f->remotes, list)
667 rdst->offloaded = false;
668 spin_unlock_bh(&vxlan->hash_lock[h]);
669 }
670
671}
672EXPORT_SYMBOL_GPL(vxlan_fdb_clear_offload);
673
674/* Replace destination of unicast mac */
675static int vxlan_fdb_replace(struct vxlan_fdb *f,
676 union vxlan_addr *ip, __be16 port, __be32 vni,
677 __u32 ifindex, struct vxlan_rdst *oldrd)
678{
679 struct vxlan_rdst *rd;
680
681 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
682 if (rd)
683 return 0;
684
685 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
686 if (!rd)
687 return 0;
688
689 *oldrd = *rd;
690 dst_cache_reset(&rd->dst_cache);
691 rd->remote_ip = *ip;
692 rd->remote_port = port;
693 rd->remote_vni = vni;
694 rd->remote_ifindex = ifindex;
695 rd->offloaded = false;
696 return 1;
697}
698
699/* Add/update destinations for multicast */
700static int vxlan_fdb_append(struct vxlan_fdb *f,
701 union vxlan_addr *ip, __be16 port, __be32 vni,
702 __u32 ifindex, struct vxlan_rdst **rdp)
703{
704 struct vxlan_rdst *rd;
705
706 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
707 if (rd)
708 return 0;
709
710 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
711 if (rd == NULL)
712 return -ENOBUFS;
713
714 if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
715 kfree(rd);
716 return -ENOBUFS;
717 }
718
719 rd->remote_ip = *ip;
720 rd->remote_port = port;
721 rd->offloaded = false;
722 rd->remote_vni = vni;
723 rd->remote_ifindex = ifindex;
724
725 list_add_tail_rcu(&rd->list, &f->remotes);
726
727 *rdp = rd;
728 return 1;
729}
730
731static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
732 unsigned int off,
733 struct vxlanhdr *vh, size_t hdrlen,
734 __be32 vni_field,
735 struct gro_remcsum *grc,
736 bool nopartial)
737{
738 size_t start, offset;
739
740 if (skb->remcsum_offload)
741 return vh;
742
743 if (!NAPI_GRO_CB(skb)->csum_valid)
744 return NULL;
745
746 start = vxlan_rco_start(vni_field);
747 offset = start + vxlan_rco_offset(vni_field);
748
749 vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
750 start, offset, grc, nopartial);
751
752 skb->remcsum_offload = 1;
753
754 return vh;
755}
756
757static struct sk_buff *vxlan_gro_receive(struct sock *sk,
758 struct list_head *head,
759 struct sk_buff *skb)
760{
761 struct sk_buff *pp = NULL;
762 struct sk_buff *p;
763 struct vxlanhdr *vh, *vh2;
764 unsigned int hlen, off_vx;
765 int flush = 1;
766 struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
767 __be32 flags;
768 struct gro_remcsum grc;
769
770 skb_gro_remcsum_init(&grc);
771
772 off_vx = skb_gro_offset(skb);
773 hlen = off_vx + sizeof(*vh);
774 vh = skb_gro_header_fast(skb, off_vx);
775 if (skb_gro_header_hard(skb, hlen)) {
776 vh = skb_gro_header_slow(skb, hlen, off_vx);
777 if (unlikely(!vh))
778 goto out;
779 }
780
781 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
782
783 flags = vh->vx_flags;
784
785 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
786 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
787 vh->vx_vni, &grc,
788 !!(vs->flags &
789 VXLAN_F_REMCSUM_NOPARTIAL));
790
791 if (!vh)
792 goto out;
793 }
794
795 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
796
797 list_for_each_entry(p, head, list) {
798 if (!NAPI_GRO_CB(p)->same_flow)
799 continue;
800
801 vh2 = (struct vxlanhdr *)(p->data + off_vx);
802 if (vh->vx_flags != vh2->vx_flags ||
803 vh->vx_vni != vh2->vx_vni) {
804 NAPI_GRO_CB(p)->same_flow = 0;
805 continue;
806 }
807 }
808
809 pp = call_gro_receive(eth_gro_receive, head, skb);
810 flush = 0;
811
812out:
813 skb_gro_flush_final_remcsum(skb, pp, flush, &grc);
814
815 return pp;
816}
817
818static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
819{
820 /* Sets 'skb->inner_mac_header' since we are always called with
821 * 'skb->encapsulation' set.
822 */
823 return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
824}
825
826static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan, const u8 *mac,
827 __u16 state, __be32 src_vni,
828 __u16 ndm_flags)
829{
830 struct vxlan_fdb *f;
831
832 f = kmalloc(sizeof(*f), GFP_ATOMIC);
833 if (!f)
834 return NULL;
835 f->state = state;
836 f->flags = ndm_flags;
837 f->updated = f->used = jiffies;
838 f->vni = src_vni;
839 f->nh = NULL;
840 RCU_INIT_POINTER(f->vdev, vxlan);
841 INIT_LIST_HEAD(&f->nh_list);
842 INIT_LIST_HEAD(&f->remotes);
843 memcpy(f->eth_addr, mac, ETH_ALEN);
844
845 return f;
846}
847
848static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac,
849 __be32 src_vni, struct vxlan_fdb *f)
850{
851 ++vxlan->addrcnt;
852 hlist_add_head_rcu(&f->hlist,
853 vxlan_fdb_head(vxlan, mac, src_vni));
854}
855
856static int vxlan_fdb_nh_update(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
857 u32 nhid, struct netlink_ext_ack *extack)
858{
859 struct nexthop *old_nh = rtnl_dereference(fdb->nh);
860 struct nexthop *nh;
861 int err = -EINVAL;
862
863 if (old_nh && old_nh->id == nhid)
864 return 0;
865
866 nh = nexthop_find_by_id(vxlan->net, nhid);
867 if (!nh) {
868 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
869 goto err_inval;
870 }
871
872 if (nh) {
873 if (!nexthop_get(nh)) {
874 NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
875 nh = NULL;
876 goto err_inval;
877 }
878 if (!nexthop_is_fdb(nh)) {
879 NL_SET_ERR_MSG(extack, "Nexthop is not a fdb nexthop");
880 goto err_inval;
881 }
882
883 if (!nexthop_is_multipath(nh)) {
884 NL_SET_ERR_MSG(extack, "Nexthop is not a multipath group");
885 goto err_inval;
886 }
887
888 /* check nexthop group family */
889 switch (vxlan->default_dst.remote_ip.sa.sa_family) {
890 case AF_INET:
891 if (!nexthop_has_v4(nh)) {
892 err = -EAFNOSUPPORT;
893 NL_SET_ERR_MSG(extack, "Nexthop group family not supported");
894 goto err_inval;
895 }
896 break;
897 case AF_INET6:
898 if (nexthop_has_v4(nh)) {
899 err = -EAFNOSUPPORT;
900 NL_SET_ERR_MSG(extack, "Nexthop group family not supported");
901 goto err_inval;
902 }
903 }
904 }
905
906 if (old_nh) {
907 list_del_rcu(&fdb->nh_list);
908 nexthop_put(old_nh);
909 }
910 rcu_assign_pointer(fdb->nh, nh);
911 list_add_tail_rcu(&fdb->nh_list, &nh->fdb_list);
912 return 1;
913
914err_inval:
915 if (nh)
916 nexthop_put(nh);
917 return err;
918}
919
920static int vxlan_fdb_create(struct vxlan_dev *vxlan,
921 const u8 *mac, union vxlan_addr *ip,
922 __u16 state, __be16 port, __be32 src_vni,
923 __be32 vni, __u32 ifindex, __u16 ndm_flags,
924 u32 nhid, struct vxlan_fdb **fdb,
925 struct netlink_ext_ack *extack)
926{
927 struct vxlan_rdst *rd = NULL;
928 struct vxlan_fdb *f;
929 int rc;
930
931 if (vxlan->cfg.addrmax &&
932 vxlan->addrcnt >= vxlan->cfg.addrmax)
933 return -ENOSPC;
934
935 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
936 f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags);
937 if (!f)
938 return -ENOMEM;
939
940 if (nhid)
941 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack);
942 else
943 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
944 if (rc < 0)
945 goto errout;
946
947 *fdb = f;
948
949 return 0;
950
951errout:
952 kfree(f);
953 return rc;
954}
955
956static void __vxlan_fdb_free(struct vxlan_fdb *f)
957{
958 struct vxlan_rdst *rd, *nd;
959 struct nexthop *nh;
960
961 nh = rcu_dereference_raw(f->nh);
962 if (nh) {
963 rcu_assign_pointer(f->nh, NULL);
964 rcu_assign_pointer(f->vdev, NULL);
965 nexthop_put(nh);
966 }
967
968 list_for_each_entry_safe(rd, nd, &f->remotes, list) {
969 dst_cache_destroy(&rd->dst_cache);
970 kfree(rd);
971 }
972 kfree(f);
973}
974
975static void vxlan_fdb_free(struct rcu_head *head)
976{
977 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
978
979 __vxlan_fdb_free(f);
980}
981
982static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
983 bool do_notify, bool swdev_notify)
984{
985 struct vxlan_rdst *rd;
986
987 netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr);
988
989 --vxlan->addrcnt;
990 if (do_notify) {
991 if (rcu_access_pointer(f->nh))
992 vxlan_fdb_notify(vxlan, f, NULL, RTM_DELNEIGH,
993 swdev_notify, NULL);
994 else
995 list_for_each_entry(rd, &f->remotes, list)
996 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH,
997 swdev_notify, NULL);
998 }
999
1000 hlist_del_rcu(&f->hlist);
1001 list_del_rcu(&f->nh_list);
1002 call_rcu(&f->rcu, vxlan_fdb_free);
1003}
1004
1005static void vxlan_dst_free(struct rcu_head *head)
1006{
1007 struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
1008
1009 dst_cache_destroy(&rd->dst_cache);
1010 kfree(rd);
1011}
1012
1013static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan,
1014 union vxlan_addr *ip,
1015 __u16 state, __u16 flags,
1016 __be16 port, __be32 vni,
1017 __u32 ifindex, __u16 ndm_flags,
1018 struct vxlan_fdb *f, u32 nhid,
1019 bool swdev_notify,
1020 struct netlink_ext_ack *extack)
1021{
1022 __u16 fdb_flags = (ndm_flags & ~NTF_USE);
1023 struct vxlan_rdst *rd = NULL;
1024 struct vxlan_rdst oldrd;
1025 int notify = 0;
1026 int rc = 0;
1027 int err;
1028
1029 if (nhid && !rcu_access_pointer(f->nh)) {
1030 NL_SET_ERR_MSG(extack,
1031 "Cannot replace an existing non nexthop fdb with a nexthop");
1032 return -EOPNOTSUPP;
1033 }
1034
1035 if (nhid && (flags & NLM_F_APPEND)) {
1036 NL_SET_ERR_MSG(extack,
1037 "Cannot append to a nexthop fdb");
1038 return -EOPNOTSUPP;
1039 }
1040
1041 /* Do not allow an externally learned entry to take over an entry added
1042 * by the user.
1043 */
1044 if (!(fdb_flags & NTF_EXT_LEARNED) ||
1045 !(f->flags & NTF_VXLAN_ADDED_BY_USER)) {
1046 if (f->state != state) {
1047 f->state = state;
1048 f->updated = jiffies;
1049 notify = 1;
1050 }
1051 if (f->flags != fdb_flags) {
1052 f->flags = fdb_flags;
1053 f->updated = jiffies;
1054 notify = 1;
1055 }
1056 }
1057
1058 if ((flags & NLM_F_REPLACE)) {
1059 /* Only change unicasts */
1060 if (!(is_multicast_ether_addr(f->eth_addr) ||
1061 is_zero_ether_addr(f->eth_addr))) {
1062 if (nhid) {
1063 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack);
1064 if (rc < 0)
1065 return rc;
1066 } else {
1067 rc = vxlan_fdb_replace(f, ip, port, vni,
1068 ifindex, &oldrd);
1069 }
1070 notify |= rc;
1071 } else {
1072 NL_SET_ERR_MSG(extack, "Cannot replace non-unicast fdb entries");
1073 return -EOPNOTSUPP;
1074 }
1075 }
1076 if ((flags & NLM_F_APPEND) &&
1077 (is_multicast_ether_addr(f->eth_addr) ||
1078 is_zero_ether_addr(f->eth_addr))) {
1079 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
1080
1081 if (rc < 0)
1082 return rc;
1083 notify |= rc;
1084 }
1085
1086 if (ndm_flags & NTF_USE)
1087 f->used = jiffies;
1088
1089 if (notify) {
1090 if (rd == NULL)
1091 rd = first_remote_rtnl(f);
1092
1093 err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH,
1094 swdev_notify, extack);
1095 if (err)
1096 goto err_notify;
1097 }
1098
1099 return 0;
1100
1101err_notify:
1102 if (nhid)
1103 return err;
1104 if ((flags & NLM_F_REPLACE) && rc)
1105 *rd = oldrd;
1106 else if ((flags & NLM_F_APPEND) && rc) {
1107 list_del_rcu(&rd->list);
1108 call_rcu(&rd->rcu, vxlan_dst_free);
1109 }
1110 return err;
1111}
1112
1113static int vxlan_fdb_update_create(struct vxlan_dev *vxlan,
1114 const u8 *mac, union vxlan_addr *ip,
1115 __u16 state, __u16 flags,
1116 __be16 port, __be32 src_vni, __be32 vni,
1117 __u32 ifindex, __u16 ndm_flags, u32 nhid,
1118 bool swdev_notify,
1119 struct netlink_ext_ack *extack)
1120{
1121 __u16 fdb_flags = (ndm_flags & ~NTF_USE);
1122 struct vxlan_fdb *f;
1123 int rc;
1124
1125 /* Disallow replace to add a multicast entry */
1126 if ((flags & NLM_F_REPLACE) &&
1127 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
1128 return -EOPNOTSUPP;
1129
1130 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
1131 rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
1132 vni, ifindex, fdb_flags, nhid, &f, extack);
1133 if (rc < 0)
1134 return rc;
1135
1136 vxlan_fdb_insert(vxlan, mac, src_vni, f);
1137 rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH,
1138 swdev_notify, extack);
1139 if (rc)
1140 goto err_notify;
1141
1142 return 0;
1143
1144err_notify:
1145 vxlan_fdb_destroy(vxlan, f, false, false);
1146 return rc;
1147}
1148
1149/* Add new entry to forwarding table -- assumes lock held */
1150static int vxlan_fdb_update(struct vxlan_dev *vxlan,
1151 const u8 *mac, union vxlan_addr *ip,
1152 __u16 state, __u16 flags,
1153 __be16 port, __be32 src_vni, __be32 vni,
1154 __u32 ifindex, __u16 ndm_flags, u32 nhid,
1155 bool swdev_notify,
1156 struct netlink_ext_ack *extack)
1157{
1158 struct vxlan_fdb *f;
1159
1160 f = __vxlan_find_mac(vxlan, mac, src_vni);
1161 if (f) {
1162 if (flags & NLM_F_EXCL) {
1163 netdev_dbg(vxlan->dev,
1164 "lost race to create %pM\n", mac);
1165 return -EEXIST;
1166 }
1167
1168 return vxlan_fdb_update_existing(vxlan, ip, state, flags, port,
1169 vni, ifindex, ndm_flags, f,
1170 nhid, swdev_notify, extack);
1171 } else {
1172 if (!(flags & NLM_F_CREATE))
1173 return -ENOENT;
1174
1175 return vxlan_fdb_update_create(vxlan, mac, ip, state, flags,
1176 port, src_vni, vni, ifindex,
1177 ndm_flags, nhid, swdev_notify,
1178 extack);
1179 }
1180}
1181
1182static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
1183 struct vxlan_rdst *rd, bool swdev_notify)
1184{
1185 list_del_rcu(&rd->list);
1186 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL);
1187 call_rcu(&rd->rcu, vxlan_dst_free);
1188}
1189
1190static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
1191 union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
1192 __be32 *vni, u32 *ifindex, u32 *nhid)
1193{
1194 struct net *net = dev_net(vxlan->dev);
1195 int err;
1196
1197 if (tb[NDA_NH_ID] && (tb[NDA_DST] || tb[NDA_VNI] || tb[NDA_IFINDEX] ||
1198 tb[NDA_PORT]))
1199 return -EINVAL;
1200
1201 if (tb[NDA_DST]) {
1202 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
1203 if (err)
1204 return err;
1205 } else {
1206 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
1207
1208 if (remote->sa.sa_family == AF_INET) {
1209 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
1210 ip->sa.sa_family = AF_INET;
1211#if IS_ENABLED(CONFIG_IPV6)
1212 } else {
1213 ip->sin6.sin6_addr = in6addr_any;
1214 ip->sa.sa_family = AF_INET6;
1215#endif
1216 }
1217 }
1218
1219 if (tb[NDA_PORT]) {
1220 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
1221 return -EINVAL;
1222 *port = nla_get_be16(tb[NDA_PORT]);
1223 } else {
1224 *port = vxlan->cfg.dst_port;
1225 }
1226
1227 if (tb[NDA_VNI]) {
1228 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
1229 return -EINVAL;
1230 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1231 } else {
1232 *vni = vxlan->default_dst.remote_vni;
1233 }
1234
1235 if (tb[NDA_SRC_VNI]) {
1236 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
1237 return -EINVAL;
1238 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
1239 } else {
1240 *src_vni = vxlan->default_dst.remote_vni;
1241 }
1242
1243 if (tb[NDA_IFINDEX]) {
1244 struct net_device *tdev;
1245
1246 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
1247 return -EINVAL;
1248 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
1249 tdev = __dev_get_by_index(net, *ifindex);
1250 if (!tdev)
1251 return -EADDRNOTAVAIL;
1252 } else {
1253 *ifindex = 0;
1254 }
1255
1256 if (tb[NDA_NH_ID])
1257 *nhid = nla_get_u32(tb[NDA_NH_ID]);
1258 else
1259 *nhid = 0;
1260
1261 return 0;
1262}
1263
1264/* Add static entry (via netlink) */
1265static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1266 struct net_device *dev,
1267 const unsigned char *addr, u16 vid, u16 flags,
1268 struct netlink_ext_ack *extack)
1269{
1270 struct vxlan_dev *vxlan = netdev_priv(dev);
1271 /* struct net *net = dev_net(vxlan->dev); */
1272 union vxlan_addr ip;
1273 __be16 port;
1274 __be32 src_vni, vni;
1275 u32 ifindex, nhid;
1276 u32 hash_index;
1277 int err;
1278
1279 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
1280 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
1281 ndm->ndm_state);
1282 return -EINVAL;
1283 }
1284
1285 if (!tb || (!tb[NDA_DST] && !tb[NDA_NH_ID]))
1286 return -EINVAL;
1287
1288 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex,
1289 &nhid);
1290 if (err)
1291 return err;
1292
1293 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
1294 return -EAFNOSUPPORT;
1295
1296 hash_index = fdb_head_index(vxlan, addr, src_vni);
1297 spin_lock_bh(&vxlan->hash_lock[hash_index]);
1298 err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
1299 port, src_vni, vni, ifindex,
1300 ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER,
1301 nhid, true, extack);
1302 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1303
1304 return err;
1305}
1306
1307static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
1308 const unsigned char *addr, union vxlan_addr ip,
1309 __be16 port, __be32 src_vni, __be32 vni,
1310 u32 ifindex, bool swdev_notify)
1311{
1312 struct vxlan_rdst *rd = NULL;
1313 struct vxlan_fdb *f;
1314 int err = -ENOENT;
1315
1316 f = vxlan_find_mac(vxlan, addr, src_vni);
1317 if (!f)
1318 return err;
1319
1320 if (!vxlan_addr_any(&ip)) {
1321 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
1322 if (!rd)
1323 goto out;
1324 }
1325
1326 /* remove a destination if it's not the only one on the list,
1327 * otherwise destroy the fdb entry
1328 */
1329 if (rd && !list_is_singular(&f->remotes)) {
1330 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify);
1331 goto out;
1332 }
1333
1334 vxlan_fdb_destroy(vxlan, f, true, swdev_notify);
1335
1336out:
1337 return 0;
1338}
1339
1340/* Delete entry (via netlink) */
1341static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
1342 struct net_device *dev,
1343 const unsigned char *addr, u16 vid)
1344{
1345 struct vxlan_dev *vxlan = netdev_priv(dev);
1346 union vxlan_addr ip;
1347 __be32 src_vni, vni;
1348 u32 ifindex, nhid;
1349 u32 hash_index;
1350 __be16 port;
1351 int err;
1352
1353 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex,
1354 &nhid);
1355 if (err)
1356 return err;
1357
1358 hash_index = fdb_head_index(vxlan, addr, src_vni);
1359 spin_lock_bh(&vxlan->hash_lock[hash_index]);
1360 err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
1361 true);
1362 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1363
1364 return err;
1365}
1366
1367/* Dump forwarding table */
1368static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1369 struct net_device *dev,
1370 struct net_device *filter_dev, int *idx)
1371{
1372 struct vxlan_dev *vxlan = netdev_priv(dev);
1373 unsigned int h;
1374 int err = 0;
1375
1376 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1377 struct vxlan_fdb *f;
1378
1379 rcu_read_lock();
1380 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
1381 struct vxlan_rdst *rd;
1382
1383 if (rcu_access_pointer(f->nh)) {
1384 if (*idx < cb->args[2])
1385 goto skip_nh;
1386 err = vxlan_fdb_info(skb, vxlan, f,
1387 NETLINK_CB(cb->skb).portid,
1388 cb->nlh->nlmsg_seq,
1389 RTM_NEWNEIGH,
1390 NLM_F_MULTI, NULL);
1391 if (err < 0) {
1392 rcu_read_unlock();
1393 goto out;
1394 }
1395skip_nh:
1396 *idx += 1;
1397 continue;
1398 }
1399
1400 list_for_each_entry_rcu(rd, &f->remotes, list) {
1401 if (*idx < cb->args[2])
1402 goto skip;
1403
1404 err = vxlan_fdb_info(skb, vxlan, f,
1405 NETLINK_CB(cb->skb).portid,
1406 cb->nlh->nlmsg_seq,
1407 RTM_NEWNEIGH,
1408 NLM_F_MULTI, rd);
1409 if (err < 0) {
1410 rcu_read_unlock();
1411 goto out;
1412 }
1413skip:
1414 *idx += 1;
1415 }
1416 }
1417 rcu_read_unlock();
1418 }
1419out:
1420 return err;
1421}
1422
1423static int vxlan_fdb_get(struct sk_buff *skb,
1424 struct nlattr *tb[],
1425 struct net_device *dev,
1426 const unsigned char *addr,
1427 u16 vid, u32 portid, u32 seq,
1428 struct netlink_ext_ack *extack)
1429{
1430 struct vxlan_dev *vxlan = netdev_priv(dev);
1431 struct vxlan_fdb *f;
1432 __be32 vni;
1433 int err;
1434
1435 if (tb[NDA_VNI])
1436 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1437 else
1438 vni = vxlan->default_dst.remote_vni;
1439
1440 rcu_read_lock();
1441
1442 f = __vxlan_find_mac(vxlan, addr, vni);
1443 if (!f) {
1444 NL_SET_ERR_MSG(extack, "Fdb entry not found");
1445 err = -ENOENT;
1446 goto errout;
1447 }
1448
1449 err = vxlan_fdb_info(skb, vxlan, f, portid, seq,
1450 RTM_NEWNEIGH, 0, first_remote_rcu(f));
1451errout:
1452 rcu_read_unlock();
1453 return err;
1454}
1455
1456/* Watch incoming packets to learn mapping between Ethernet address
1457 * and Tunnel endpoint.
1458 * Return true if packet is bogus and should be dropped.
1459 */
1460static bool vxlan_snoop(struct net_device *dev,
1461 union vxlan_addr *src_ip, const u8 *src_mac,
1462 u32 src_ifindex, __be32 vni)
1463{
1464 struct vxlan_dev *vxlan = netdev_priv(dev);
1465 struct vxlan_fdb *f;
1466 u32 ifindex = 0;
1467
1468#if IS_ENABLED(CONFIG_IPV6)
1469 if (src_ip->sa.sa_family == AF_INET6 &&
1470 (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1471 ifindex = src_ifindex;
1472#endif
1473
1474 f = vxlan_find_mac(vxlan, src_mac, vni);
1475 if (likely(f)) {
1476 struct vxlan_rdst *rdst = first_remote_rcu(f);
1477
1478 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1479 rdst->remote_ifindex == ifindex))
1480 return false;
1481
1482 /* Don't migrate static entries, drop packets */
1483 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1484 return true;
1485
1486 /* Don't override an fdb with nexthop with a learnt entry */
1487 if (rcu_access_pointer(f->nh))
1488 return true;
1489
1490 if (net_ratelimit())
1491 netdev_info(dev,
1492 "%pM migrated from %pIS to %pIS\n",
1493 src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1494
1495 rdst->remote_ip = *src_ip;
1496 f->updated = jiffies;
1497 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL);
1498 } else {
1499 u32 hash_index = fdb_head_index(vxlan, src_mac, vni);
1500
1501 /* learned new entry */
1502 spin_lock(&vxlan->hash_lock[hash_index]);
1503
1504 /* close off race between vxlan_flush and incoming packets */
1505 if (netif_running(dev))
1506 vxlan_fdb_update(vxlan, src_mac, src_ip,
1507 NUD_REACHABLE,
1508 NLM_F_EXCL|NLM_F_CREATE,
1509 vxlan->cfg.dst_port,
1510 vni,
1511 vxlan->default_dst.remote_vni,
1512 ifindex, NTF_SELF, 0, true, NULL);
1513 spin_unlock(&vxlan->hash_lock[hash_index]);
1514 }
1515
1516 return false;
1517}
1518
1519/* See if multicast group is already in use by other ID */
1520static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1521{
1522 struct vxlan_dev *vxlan;
1523 struct vxlan_sock *sock4;
1524#if IS_ENABLED(CONFIG_IPV6)
1525 struct vxlan_sock *sock6;
1526#endif
1527 unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1528
1529 sock4 = rtnl_dereference(dev->vn4_sock);
1530
1531 /* The vxlan_sock is only used by dev, leaving group has
1532 * no effect on other vxlan devices.
1533 */
1534 if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1535 return false;
1536#if IS_ENABLED(CONFIG_IPV6)
1537 sock6 = rtnl_dereference(dev->vn6_sock);
1538 if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1539 return false;
1540#endif
1541
1542 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1543 if (!netif_running(vxlan->dev) || vxlan == dev)
1544 continue;
1545
1546 if (family == AF_INET &&
1547 rtnl_dereference(vxlan->vn4_sock) != sock4)
1548 continue;
1549#if IS_ENABLED(CONFIG_IPV6)
1550 if (family == AF_INET6 &&
1551 rtnl_dereference(vxlan->vn6_sock) != sock6)
1552 continue;
1553#endif
1554
1555 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1556 &dev->default_dst.remote_ip))
1557 continue;
1558
1559 if (vxlan->default_dst.remote_ifindex !=
1560 dev->default_dst.remote_ifindex)
1561 continue;
1562
1563 return true;
1564 }
1565
1566 return false;
1567}
1568
1569static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1570{
1571 struct vxlan_net *vn;
1572
1573 if (!vs)
1574 return false;
1575 if (!refcount_dec_and_test(&vs->refcnt))
1576 return false;
1577
1578 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1579 spin_lock(&vn->sock_lock);
1580 hlist_del_rcu(&vs->hlist);
1581 udp_tunnel_notify_del_rx_port(vs->sock,
1582 (vs->flags & VXLAN_F_GPE) ?
1583 UDP_TUNNEL_TYPE_VXLAN_GPE :
1584 UDP_TUNNEL_TYPE_VXLAN);
1585 spin_unlock(&vn->sock_lock);
1586
1587 return true;
1588}
1589
1590static void vxlan_sock_release(struct vxlan_dev *vxlan)
1591{
1592 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1593#if IS_ENABLED(CONFIG_IPV6)
1594 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1595
1596 RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1597#endif
1598
1599 RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1600 synchronize_net();
1601
1602 vxlan_vs_del_dev(vxlan);
1603
1604 if (__vxlan_sock_release_prep(sock4)) {
1605 udp_tunnel_sock_release(sock4->sock);
1606 kfree(sock4);
1607 }
1608
1609#if IS_ENABLED(CONFIG_IPV6)
1610 if (__vxlan_sock_release_prep(sock6)) {
1611 udp_tunnel_sock_release(sock6->sock);
1612 kfree(sock6);
1613 }
1614#endif
1615}
1616
1617/* Update multicast group membership when first VNI on
1618 * multicast address is brought up
1619 */
1620static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1621{
1622 struct sock *sk;
1623 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1624 int ifindex = vxlan->default_dst.remote_ifindex;
1625 int ret = -EINVAL;
1626
1627 if (ip->sa.sa_family == AF_INET) {
1628 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1629 struct ip_mreqn mreq = {
1630 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1631 .imr_ifindex = ifindex,
1632 };
1633
1634 sk = sock4->sock->sk;
1635 lock_sock(sk);
1636 ret = ip_mc_join_group(sk, &mreq);
1637 release_sock(sk);
1638#if IS_ENABLED(CONFIG_IPV6)
1639 } else {
1640 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1641
1642 sk = sock6->sock->sk;
1643 lock_sock(sk);
1644 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1645 &ip->sin6.sin6_addr);
1646 release_sock(sk);
1647#endif
1648 }
1649
1650 return ret;
1651}
1652
1653/* Inverse of vxlan_igmp_join when last VNI is brought down */
1654static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1655{
1656 struct sock *sk;
1657 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1658 int ifindex = vxlan->default_dst.remote_ifindex;
1659 int ret = -EINVAL;
1660
1661 if (ip->sa.sa_family == AF_INET) {
1662 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1663 struct ip_mreqn mreq = {
1664 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1665 .imr_ifindex = ifindex,
1666 };
1667
1668 sk = sock4->sock->sk;
1669 lock_sock(sk);
1670 ret = ip_mc_leave_group(sk, &mreq);
1671 release_sock(sk);
1672#if IS_ENABLED(CONFIG_IPV6)
1673 } else {
1674 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1675
1676 sk = sock6->sock->sk;
1677 lock_sock(sk);
1678 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1679 &ip->sin6.sin6_addr);
1680 release_sock(sk);
1681#endif
1682 }
1683
1684 return ret;
1685}
1686
1687static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1688 struct sk_buff *skb, u32 vxflags)
1689{
1690 size_t start, offset;
1691
1692 if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1693 goto out;
1694
1695 start = vxlan_rco_start(unparsed->vx_vni);
1696 offset = start + vxlan_rco_offset(unparsed->vx_vni);
1697
1698 if (!pskb_may_pull(skb, offset + sizeof(u16)))
1699 return false;
1700
1701 skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1702 !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1703out:
1704 unparsed->vx_flags &= ~VXLAN_HF_RCO;
1705 unparsed->vx_vni &= VXLAN_VNI_MASK;
1706 return true;
1707}
1708
1709static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1710 struct sk_buff *skb, u32 vxflags,
1711 struct vxlan_metadata *md)
1712{
1713 struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1714 struct metadata_dst *tun_dst;
1715
1716 if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1717 goto out;
1718
1719 md->gbp = ntohs(gbp->policy_id);
1720
1721 tun_dst = (struct metadata_dst *)skb_dst(skb);
1722 if (tun_dst) {
1723 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1724 tun_dst->u.tun_info.options_len = sizeof(*md);
1725 }
1726 if (gbp->dont_learn)
1727 md->gbp |= VXLAN_GBP_DONT_LEARN;
1728
1729 if (gbp->policy_applied)
1730 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1731
1732 /* In flow-based mode, GBP is carried in dst_metadata */
1733 if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1734 skb->mark = md->gbp;
1735out:
1736 unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1737}
1738
1739static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1740 __be16 *protocol,
1741 struct sk_buff *skb, u32 vxflags)
1742{
1743 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1744
1745 /* Need to have Next Protocol set for interfaces in GPE mode. */
1746 if (!gpe->np_applied)
1747 return false;
1748 /* "The initial version is 0. If a receiver does not support the
1749 * version indicated it MUST drop the packet.
1750 */
1751 if (gpe->version != 0)
1752 return false;
1753 /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1754 * processing MUST occur." However, we don't implement OAM
1755 * processing, thus drop the packet.
1756 */
1757 if (gpe->oam_flag)
1758 return false;
1759
1760 *protocol = tun_p_to_eth_p(gpe->next_protocol);
1761 if (!*protocol)
1762 return false;
1763
1764 unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1765 return true;
1766}
1767
1768static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1769 struct vxlan_sock *vs,
1770 struct sk_buff *skb, __be32 vni)
1771{
1772 union vxlan_addr saddr;
1773 u32 ifindex = skb->dev->ifindex;
1774
1775 skb_reset_mac_header(skb);
1776 skb->protocol = eth_type_trans(skb, vxlan->dev);
1777 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1778
1779 /* Ignore packet loops (and multicast echo) */
1780 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1781 return false;
1782
1783 /* Get address from the outer IP header */
1784 if (vxlan_get_sk_family(vs) == AF_INET) {
1785 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1786 saddr.sa.sa_family = AF_INET;
1787#if IS_ENABLED(CONFIG_IPV6)
1788 } else {
1789 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1790 saddr.sa.sa_family = AF_INET6;
1791#endif
1792 }
1793
1794 if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1795 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1796 return false;
1797
1798 return true;
1799}
1800
1801static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1802 struct sk_buff *skb)
1803{
1804 int err = 0;
1805
1806 if (vxlan_get_sk_family(vs) == AF_INET)
1807 err = IP_ECN_decapsulate(oiph, skb);
1808#if IS_ENABLED(CONFIG_IPV6)
1809 else
1810 err = IP6_ECN_decapsulate(oiph, skb);
1811#endif
1812
1813 if (unlikely(err) && log_ecn_error) {
1814 if (vxlan_get_sk_family(vs) == AF_INET)
1815 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1816 &((struct iphdr *)oiph)->saddr,
1817 ((struct iphdr *)oiph)->tos);
1818 else
1819 net_info_ratelimited("non-ECT from %pI6\n",
1820 &((struct ipv6hdr *)oiph)->saddr);
1821 }
1822 return err <= 1;
1823}
1824
1825/* Callback from net/ipv4/udp.c to receive packets */
1826static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1827{
1828 struct pcpu_sw_netstats *stats;
1829 struct vxlan_dev *vxlan;
1830 struct vxlan_sock *vs;
1831 struct vxlanhdr unparsed;
1832 struct vxlan_metadata _md;
1833 struct vxlan_metadata *md = &_md;
1834 __be16 protocol = htons(ETH_P_TEB);
1835 bool raw_proto = false;
1836 void *oiph;
1837 __be32 vni = 0;
1838
1839 /* Need UDP and VXLAN header to be present */
1840 if (!pskb_may_pull(skb, VXLAN_HLEN))
1841 goto drop;
1842
1843 unparsed = *vxlan_hdr(skb);
1844 /* VNI flag always required to be set */
1845 if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1846 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1847 ntohl(vxlan_hdr(skb)->vx_flags),
1848 ntohl(vxlan_hdr(skb)->vx_vni));
1849 /* Return non vxlan pkt */
1850 goto drop;
1851 }
1852 unparsed.vx_flags &= ~VXLAN_HF_VNI;
1853 unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1854
1855 vs = rcu_dereference_sk_user_data(sk);
1856 if (!vs)
1857 goto drop;
1858
1859 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1860
1861 vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1862 if (!vxlan)
1863 goto drop;
1864
1865 /* For backwards compatibility, only allow reserved fields to be
1866 * used by VXLAN extensions if explicitly requested.
1867 */
1868 if (vs->flags & VXLAN_F_GPE) {
1869 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1870 goto drop;
1871 raw_proto = true;
1872 }
1873
1874 if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1875 !net_eq(vxlan->net, dev_net(vxlan->dev))))
1876 goto drop;
1877
1878 if (vxlan_collect_metadata(vs)) {
1879 struct metadata_dst *tun_dst;
1880
1881 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1882 key32_to_tunnel_id(vni), sizeof(*md));
1883
1884 if (!tun_dst)
1885 goto drop;
1886
1887 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1888
1889 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1890 } else {
1891 memset(md, 0, sizeof(*md));
1892 }
1893
1894 if (vs->flags & VXLAN_F_REMCSUM_RX)
1895 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1896 goto drop;
1897 if (vs->flags & VXLAN_F_GBP)
1898 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1899 /* Note that GBP and GPE can never be active together. This is
1900 * ensured in vxlan_dev_configure.
1901 */
1902
1903 if (unparsed.vx_flags || unparsed.vx_vni) {
1904 /* If there are any unprocessed flags remaining treat
1905 * this as a malformed packet. This behavior diverges from
1906 * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1907 * in reserved fields are to be ignored. The approach here
1908 * maintains compatibility with previous stack code, and also
1909 * is more robust and provides a little more security in
1910 * adding extensions to VXLAN.
1911 */
1912 goto drop;
1913 }
1914
1915 if (!raw_proto) {
1916 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1917 goto drop;
1918 } else {
1919 skb_reset_mac_header(skb);
1920 skb->dev = vxlan->dev;
1921 skb->pkt_type = PACKET_HOST;
1922 }
1923
1924 oiph = skb_network_header(skb);
1925 skb_reset_network_header(skb);
1926
1927 if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1928 ++vxlan->dev->stats.rx_frame_errors;
1929 ++vxlan->dev->stats.rx_errors;
1930 goto drop;
1931 }
1932
1933 rcu_read_lock();
1934
1935 if (unlikely(!(vxlan->dev->flags & IFF_UP))) {
1936 rcu_read_unlock();
1937 atomic_long_inc(&vxlan->dev->rx_dropped);
1938 goto drop;
1939 }
1940
1941 stats = this_cpu_ptr(vxlan->dev->tstats);
1942 u64_stats_update_begin(&stats->syncp);
1943 stats->rx_packets++;
1944 stats->rx_bytes += skb->len;
1945 u64_stats_update_end(&stats->syncp);
1946
1947 gro_cells_receive(&vxlan->gro_cells, skb);
1948
1949 rcu_read_unlock();
1950
1951 return 0;
1952
1953drop:
1954 /* Consume bad packet */
1955 kfree_skb(skb);
1956 return 0;
1957}
1958
1959/* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */
1960static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb)
1961{
1962 struct vxlan_dev *vxlan;
1963 struct vxlan_sock *vs;
1964 struct vxlanhdr *hdr;
1965 __be32 vni;
1966
1967 if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN))
1968 return -EINVAL;
1969
1970 hdr = vxlan_hdr(skb);
1971
1972 if (!(hdr->vx_flags & VXLAN_HF_VNI))
1973 return -EINVAL;
1974
1975 vs = rcu_dereference_sk_user_data(sk);
1976 if (!vs)
1977 return -ENOENT;
1978
1979 vni = vxlan_vni(hdr->vx_vni);
1980 vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1981 if (!vxlan)
1982 return -ENOENT;
1983
1984 return 0;
1985}
1986
1987static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1988{
1989 struct vxlan_dev *vxlan = netdev_priv(dev);
1990 struct arphdr *parp;
1991 u8 *arpptr, *sha;
1992 __be32 sip, tip;
1993 struct neighbour *n;
1994
1995 if (dev->flags & IFF_NOARP)
1996 goto out;
1997
1998 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1999 dev->stats.tx_dropped++;
2000 goto out;
2001 }
2002 parp = arp_hdr(skb);
2003
2004 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
2005 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
2006 parp->ar_pro != htons(ETH_P_IP) ||
2007 parp->ar_op != htons(ARPOP_REQUEST) ||
2008 parp->ar_hln != dev->addr_len ||
2009 parp->ar_pln != 4)
2010 goto out;
2011 arpptr = (u8 *)parp + sizeof(struct arphdr);
2012 sha = arpptr;
2013 arpptr += dev->addr_len; /* sha */
2014 memcpy(&sip, arpptr, sizeof(sip));
2015 arpptr += sizeof(sip);
2016 arpptr += dev->addr_len; /* tha */
2017 memcpy(&tip, arpptr, sizeof(tip));
2018
2019 if (ipv4_is_loopback(tip) ||
2020 ipv4_is_multicast(tip))
2021 goto out;
2022
2023 n = neigh_lookup(&arp_tbl, &tip, dev);
2024
2025 if (n) {
2026 struct vxlan_fdb *f;
2027 struct sk_buff *reply;
2028
2029 if (!(n->nud_state & NUD_CONNECTED)) {
2030 neigh_release(n);
2031 goto out;
2032 }
2033
2034 f = vxlan_find_mac(vxlan, n->ha, vni);
2035 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2036 /* bridge-local neighbor */
2037 neigh_release(n);
2038 goto out;
2039 }
2040
2041 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
2042 n->ha, sha);
2043
2044 neigh_release(n);
2045
2046 if (reply == NULL)
2047 goto out;
2048
2049 skb_reset_mac_header(reply);
2050 __skb_pull(reply, skb_network_offset(reply));
2051 reply->ip_summed = CHECKSUM_UNNECESSARY;
2052 reply->pkt_type = PACKET_HOST;
2053
2054 if (netif_rx_ni(reply) == NET_RX_DROP)
2055 dev->stats.rx_dropped++;
2056 } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2057 union vxlan_addr ipa = {
2058 .sin.sin_addr.s_addr = tip,
2059 .sin.sin_family = AF_INET,
2060 };
2061
2062 vxlan_ip_miss(dev, &ipa);
2063 }
2064out:
2065 consume_skb(skb);
2066 return NETDEV_TX_OK;
2067}
2068
2069#if IS_ENABLED(CONFIG_IPV6)
2070static struct sk_buff *vxlan_na_create(struct sk_buff *request,
2071 struct neighbour *n, bool isrouter)
2072{
2073 struct net_device *dev = request->dev;
2074 struct sk_buff *reply;
2075 struct nd_msg *ns, *na;
2076 struct ipv6hdr *pip6;
2077 u8 *daddr;
2078 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
2079 int ns_olen;
2080 int i, len;
2081
2082 if (dev == NULL || !pskb_may_pull(request, request->len))
2083 return NULL;
2084
2085 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
2086 sizeof(*na) + na_olen + dev->needed_tailroom;
2087 reply = alloc_skb(len, GFP_ATOMIC);
2088 if (reply == NULL)
2089 return NULL;
2090
2091 reply->protocol = htons(ETH_P_IPV6);
2092 reply->dev = dev;
2093 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
2094 skb_push(reply, sizeof(struct ethhdr));
2095 skb_reset_mac_header(reply);
2096
2097 ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
2098
2099 daddr = eth_hdr(request)->h_source;
2100 ns_olen = request->len - skb_network_offset(request) -
2101 sizeof(struct ipv6hdr) - sizeof(*ns);
2102 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
2103 if (!ns->opt[i + 1]) {
2104 kfree_skb(reply);
2105 return NULL;
2106 }
2107 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
2108 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
2109 break;
2110 }
2111 }
2112
2113 /* Ethernet header */
2114 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
2115 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
2116 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
2117 reply->protocol = htons(ETH_P_IPV6);
2118
2119 skb_pull(reply, sizeof(struct ethhdr));
2120 skb_reset_network_header(reply);
2121 skb_put(reply, sizeof(struct ipv6hdr));
2122
2123 /* IPv6 header */
2124
2125 pip6 = ipv6_hdr(reply);
2126 memset(pip6, 0, sizeof(struct ipv6hdr));
2127 pip6->version = 6;
2128 pip6->priority = ipv6_hdr(request)->priority;
2129 pip6->nexthdr = IPPROTO_ICMPV6;
2130 pip6->hop_limit = 255;
2131 pip6->daddr = ipv6_hdr(request)->saddr;
2132 pip6->saddr = *(struct in6_addr *)n->primary_key;
2133
2134 skb_pull(reply, sizeof(struct ipv6hdr));
2135 skb_reset_transport_header(reply);
2136
2137 /* Neighbor Advertisement */
2138 na = skb_put_zero(reply, sizeof(*na) + na_olen);
2139 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
2140 na->icmph.icmp6_router = isrouter;
2141 na->icmph.icmp6_override = 1;
2142 na->icmph.icmp6_solicited = 1;
2143 na->target = ns->target;
2144 ether_addr_copy(&na->opt[2], n->ha);
2145 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
2146 na->opt[1] = na_olen >> 3;
2147
2148 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
2149 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
2150 csum_partial(na, sizeof(*na)+na_olen, 0));
2151
2152 pip6->payload_len = htons(sizeof(*na)+na_olen);
2153
2154 skb_push(reply, sizeof(struct ipv6hdr));
2155
2156 reply->ip_summed = CHECKSUM_UNNECESSARY;
2157
2158 return reply;
2159}
2160
2161static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
2162{
2163 struct vxlan_dev *vxlan = netdev_priv(dev);
2164 const struct in6_addr *daddr;
2165 const struct ipv6hdr *iphdr;
2166 struct inet6_dev *in6_dev;
2167 struct neighbour *n;
2168 struct nd_msg *msg;
2169
2170 in6_dev = __in6_dev_get(dev);
2171 if (!in6_dev)
2172 goto out;
2173
2174 iphdr = ipv6_hdr(skb);
2175 daddr = &iphdr->daddr;
2176 msg = (struct nd_msg *)(iphdr + 1);
2177
2178 if (ipv6_addr_loopback(daddr) ||
2179 ipv6_addr_is_multicast(&msg->target))
2180 goto out;
2181
2182 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
2183
2184 if (n) {
2185 struct vxlan_fdb *f;
2186 struct sk_buff *reply;
2187
2188 if (!(n->nud_state & NUD_CONNECTED)) {
2189 neigh_release(n);
2190 goto out;
2191 }
2192
2193 f = vxlan_find_mac(vxlan, n->ha, vni);
2194 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2195 /* bridge-local neighbor */
2196 neigh_release(n);
2197 goto out;
2198 }
2199
2200 reply = vxlan_na_create(skb, n,
2201 !!(f ? f->flags & NTF_ROUTER : 0));
2202
2203 neigh_release(n);
2204
2205 if (reply == NULL)
2206 goto out;
2207
2208 if (netif_rx_ni(reply) == NET_RX_DROP)
2209 dev->stats.rx_dropped++;
2210
2211 } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2212 union vxlan_addr ipa = {
2213 .sin6.sin6_addr = msg->target,
2214 .sin6.sin6_family = AF_INET6,
2215 };
2216
2217 vxlan_ip_miss(dev, &ipa);
2218 }
2219
2220out:
2221 consume_skb(skb);
2222 return NETDEV_TX_OK;
2223}
2224#endif
2225
2226static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
2227{
2228 struct vxlan_dev *vxlan = netdev_priv(dev);
2229 struct neighbour *n;
2230
2231 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
2232 return false;
2233
2234 n = NULL;
2235 switch (ntohs(eth_hdr(skb)->h_proto)) {
2236 case ETH_P_IP:
2237 {
2238 struct iphdr *pip;
2239
2240 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
2241 return false;
2242 pip = ip_hdr(skb);
2243 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
2244 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2245 union vxlan_addr ipa = {
2246 .sin.sin_addr.s_addr = pip->daddr,
2247 .sin.sin_family = AF_INET,
2248 };
2249
2250 vxlan_ip_miss(dev, &ipa);
2251 return false;
2252 }
2253
2254 break;
2255 }
2256#if IS_ENABLED(CONFIG_IPV6)
2257 case ETH_P_IPV6:
2258 {
2259 struct ipv6hdr *pip6;
2260
2261 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
2262 return false;
2263 pip6 = ipv6_hdr(skb);
2264 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
2265 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2266 union vxlan_addr ipa = {
2267 .sin6.sin6_addr = pip6->daddr,
2268 .sin6.sin6_family = AF_INET6,
2269 };
2270
2271 vxlan_ip_miss(dev, &ipa);
2272 return false;
2273 }
2274
2275 break;
2276 }
2277#endif
2278 default:
2279 return false;
2280 }
2281
2282 if (n) {
2283 bool diff;
2284
2285 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
2286 if (diff) {
2287 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
2288 dev->addr_len);
2289 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
2290 }
2291 neigh_release(n);
2292 return diff;
2293 }
2294
2295 return false;
2296}
2297
2298static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
2299 struct vxlan_metadata *md)
2300{
2301 struct vxlanhdr_gbp *gbp;
2302
2303 if (!md->gbp)
2304 return;
2305
2306 gbp = (struct vxlanhdr_gbp *)vxh;
2307 vxh->vx_flags |= VXLAN_HF_GBP;
2308
2309 if (md->gbp & VXLAN_GBP_DONT_LEARN)
2310 gbp->dont_learn = 1;
2311
2312 if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
2313 gbp->policy_applied = 1;
2314
2315 gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
2316}
2317
2318static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
2319 __be16 protocol)
2320{
2321 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
2322
2323 gpe->np_applied = 1;
2324 gpe->next_protocol = tun_p_from_eth_p(protocol);
2325 if (!gpe->next_protocol)
2326 return -EPFNOSUPPORT;
2327 return 0;
2328}
2329
2330static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
2331 int iphdr_len, __be32 vni,
2332 struct vxlan_metadata *md, u32 vxflags,
2333 bool udp_sum)
2334{
2335 struct vxlanhdr *vxh;
2336 int min_headroom;
2337 int err;
2338 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
2339 __be16 inner_protocol = htons(ETH_P_TEB);
2340
2341 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
2342 skb->ip_summed == CHECKSUM_PARTIAL) {
2343 int csum_start = skb_checksum_start_offset(skb);
2344
2345 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
2346 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
2347 (skb->csum_offset == offsetof(struct udphdr, check) ||
2348 skb->csum_offset == offsetof(struct tcphdr, check)))
2349 type |= SKB_GSO_TUNNEL_REMCSUM;
2350 }
2351
2352 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
2353 + VXLAN_HLEN + iphdr_len;
2354
2355 /* Need space for new headers (invalidates iph ptr) */
2356 err = skb_cow_head(skb, min_headroom);
2357 if (unlikely(err))
2358 return err;
2359
2360 err = iptunnel_handle_offloads(skb, type);
2361 if (err)
2362 return err;
2363
2364 vxh = __skb_push(skb, sizeof(*vxh));
2365 vxh->vx_flags = VXLAN_HF_VNI;
2366 vxh->vx_vni = vxlan_vni_field(vni);
2367
2368 if (type & SKB_GSO_TUNNEL_REMCSUM) {
2369 unsigned int start;
2370
2371 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
2372 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
2373 vxh->vx_flags |= VXLAN_HF_RCO;
2374
2375 if (!skb_is_gso(skb)) {
2376 skb->ip_summed = CHECKSUM_NONE;
2377 skb->encapsulation = 0;
2378 }
2379 }
2380
2381 if (vxflags & VXLAN_F_GBP)
2382 vxlan_build_gbp_hdr(vxh, vxflags, md);
2383 if (vxflags & VXLAN_F_GPE) {
2384 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
2385 if (err < 0)
2386 return err;
2387 inner_protocol = skb->protocol;
2388 }
2389
2390 skb_set_inner_protocol(skb, inner_protocol);
2391 return 0;
2392}
2393
2394static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
2395 struct vxlan_sock *sock4,
2396 struct sk_buff *skb, int oif, u8 tos,
2397 __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
2398 struct dst_cache *dst_cache,
2399 const struct ip_tunnel_info *info)
2400{
2401 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2402 struct rtable *rt = NULL;
2403 struct flowi4 fl4;
2404
2405 if (!sock4)
2406 return ERR_PTR(-EIO);
2407
2408 if (tos && !info)
2409 use_cache = false;
2410 if (use_cache) {
2411 rt = dst_cache_get_ip4(dst_cache, saddr);
2412 if (rt)
2413 return rt;
2414 }
2415
2416 memset(&fl4, 0, sizeof(fl4));
2417 fl4.flowi4_oif = oif;
2418 fl4.flowi4_tos = RT_TOS(tos);
2419 fl4.flowi4_mark = skb->mark;
2420 fl4.flowi4_proto = IPPROTO_UDP;
2421 fl4.daddr = daddr;
2422 fl4.saddr = *saddr;
2423 fl4.fl4_dport = dport;
2424 fl4.fl4_sport = sport;
2425
2426 rt = ip_route_output_key(vxlan->net, &fl4);
2427 if (!IS_ERR(rt)) {
2428 if (rt->dst.dev == dev) {
2429 netdev_dbg(dev, "circular route to %pI4\n", &daddr);
2430 ip_rt_put(rt);
2431 return ERR_PTR(-ELOOP);
2432 }
2433
2434 *saddr = fl4.saddr;
2435 if (use_cache)
2436 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2437 } else {
2438 netdev_dbg(dev, "no route to %pI4\n", &daddr);
2439 return ERR_PTR(-ENETUNREACH);
2440 }
2441 return rt;
2442}
2443
2444#if IS_ENABLED(CONFIG_IPV6)
2445static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
2446 struct net_device *dev,
2447 struct vxlan_sock *sock6,
2448 struct sk_buff *skb, int oif, u8 tos,
2449 __be32 label,
2450 const struct in6_addr *daddr,
2451 struct in6_addr *saddr,
2452 __be16 dport, __be16 sport,
2453 struct dst_cache *dst_cache,
2454 const struct ip_tunnel_info *info)
2455{
2456 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2457 struct dst_entry *ndst;
2458 struct flowi6 fl6;
2459
2460 if (!sock6)
2461 return ERR_PTR(-EIO);
2462
2463 if (tos && !info)
2464 use_cache = false;
2465 if (use_cache) {
2466 ndst = dst_cache_get_ip6(dst_cache, saddr);
2467 if (ndst)
2468 return ndst;
2469 }
2470
2471 memset(&fl6, 0, sizeof(fl6));
2472 fl6.flowi6_oif = oif;
2473 fl6.daddr = *daddr;
2474 fl6.saddr = *saddr;
2475 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
2476 fl6.flowi6_mark = skb->mark;
2477 fl6.flowi6_proto = IPPROTO_UDP;
2478 fl6.fl6_dport = dport;
2479 fl6.fl6_sport = sport;
2480
2481 ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk,
2482 &fl6, NULL);
2483 if (unlikely(IS_ERR(ndst))) {
2484 netdev_dbg(dev, "no route to %pI6\n", daddr);
2485 return ERR_PTR(-ENETUNREACH);
2486 }
2487
2488 if (unlikely(ndst->dev == dev)) {
2489 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2490 dst_release(ndst);
2491 return ERR_PTR(-ELOOP);
2492 }
2493
2494 *saddr = fl6.saddr;
2495 if (use_cache)
2496 dst_cache_set_ip6(dst_cache, ndst, saddr);
2497 return ndst;
2498}
2499#endif
2500
2501/* Bypass encapsulation if the destination is local */
2502static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2503 struct vxlan_dev *dst_vxlan, __be32 vni,
2504 bool snoop)
2505{
2506 struct pcpu_sw_netstats *tx_stats, *rx_stats;
2507 union vxlan_addr loopback;
2508 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2509 struct net_device *dev;
2510 int len = skb->len;
2511
2512 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2513 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2514 skb->pkt_type = PACKET_HOST;
2515 skb->encapsulation = 0;
2516 skb->dev = dst_vxlan->dev;
2517 __skb_pull(skb, skb_network_offset(skb));
2518
2519 if (remote_ip->sa.sa_family == AF_INET) {
2520 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2521 loopback.sa.sa_family = AF_INET;
2522#if IS_ENABLED(CONFIG_IPV6)
2523 } else {
2524 loopback.sin6.sin6_addr = in6addr_loopback;
2525 loopback.sa.sa_family = AF_INET6;
2526#endif
2527 }
2528
2529 rcu_read_lock();
2530 dev = skb->dev;
2531 if (unlikely(!(dev->flags & IFF_UP))) {
2532 kfree_skb(skb);
2533 goto drop;
2534 }
2535
2536 if ((dst_vxlan->cfg.flags & VXLAN_F_LEARN) && snoop)
2537 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2538
2539 u64_stats_update_begin(&tx_stats->syncp);
2540 tx_stats->tx_packets++;
2541 tx_stats->tx_bytes += len;
2542 u64_stats_update_end(&tx_stats->syncp);
2543
2544 if (netif_rx(skb) == NET_RX_SUCCESS) {
2545 u64_stats_update_begin(&rx_stats->syncp);
2546 rx_stats->rx_packets++;
2547 rx_stats->rx_bytes += len;
2548 u64_stats_update_end(&rx_stats->syncp);
2549 } else {
2550drop:
2551 dev->stats.rx_dropped++;
2552 }
2553 rcu_read_unlock();
2554}
2555
2556static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2557 struct vxlan_dev *vxlan,
2558 union vxlan_addr *daddr,
2559 __be16 dst_port, int dst_ifindex, __be32 vni,
2560 struct dst_entry *dst,
2561 u32 rt_flags)
2562{
2563#if IS_ENABLED(CONFIG_IPV6)
2564 /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2565 * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2566 * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2567 */
2568 BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2569#endif
2570 /* Bypass encapsulation if the destination is local */
2571 if (rt_flags & RTCF_LOCAL &&
2572 !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2573 struct vxlan_dev *dst_vxlan;
2574
2575 dst_release(dst);
2576 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2577 daddr->sa.sa_family, dst_port,
2578 vxlan->cfg.flags);
2579 if (!dst_vxlan) {
2580 dev->stats.tx_errors++;
2581 kfree_skb(skb);
2582
2583 return -ENOENT;
2584 }
2585 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni, true);
2586 return 1;
2587 }
2588
2589 return 0;
2590}
2591
2592static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2593 __be32 default_vni, struct vxlan_rdst *rdst,
2594 bool did_rsc)
2595{
2596 struct dst_cache *dst_cache;
2597 struct ip_tunnel_info *info;
2598 struct vxlan_dev *vxlan = netdev_priv(dev);
2599 const struct iphdr *old_iph = ip_hdr(skb);
2600 union vxlan_addr *dst;
2601 union vxlan_addr remote_ip, local_ip;
2602 struct vxlan_metadata _md;
2603 struct vxlan_metadata *md = &_md;
2604 __be16 src_port = 0, dst_port;
2605 struct dst_entry *ndst = NULL;
2606 __be32 vni, label;
2607 __u8 tos, ttl;
2608 int ifindex;
2609 int err;
2610 u32 flags = vxlan->cfg.flags;
2611 bool udp_sum = false;
2612 bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2613
2614 info = skb_tunnel_info(skb);
2615
2616 if (rdst) {
2617 dst = &rdst->remote_ip;
2618 if (vxlan_addr_any(dst)) {
2619 if (did_rsc) {
2620 /* short-circuited back to local bridge */
2621 vxlan_encap_bypass(skb, vxlan, vxlan,
2622 default_vni, true);
2623 return;
2624 }
2625 goto drop;
2626 }
2627
2628 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2629 vni = (rdst->remote_vni) ? : default_vni;
2630 ifindex = rdst->remote_ifindex;
2631 local_ip = vxlan->cfg.saddr;
2632 dst_cache = &rdst->dst_cache;
2633 md->gbp = skb->mark;
2634 if (flags & VXLAN_F_TTL_INHERIT) {
2635 ttl = ip_tunnel_get_ttl(old_iph, skb);
2636 } else {
2637 ttl = vxlan->cfg.ttl;
2638 if (!ttl && vxlan_addr_multicast(dst))
2639 ttl = 1;
2640 }
2641
2642 tos = vxlan->cfg.tos;
2643 if (tos == 1)
2644 tos = ip_tunnel_get_dsfield(old_iph, skb);
2645
2646 if (dst->sa.sa_family == AF_INET)
2647 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2648 else
2649 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2650 label = vxlan->cfg.label;
2651 } else {
2652 if (!info) {
2653 WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2654 dev->name);
2655 goto drop;
2656 }
2657 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2658 if (remote_ip.sa.sa_family == AF_INET) {
2659 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2660 local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2661 } else {
2662 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2663 local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2664 }
2665 dst = &remote_ip;
2666 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2667 vni = tunnel_id_to_key32(info->key.tun_id);
2668 ifindex = 0;
2669 dst_cache = &info->dst_cache;
2670 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) {
2671 if (info->options_len < sizeof(*md))
2672 goto drop;
2673 md = ip_tunnel_info_opts(info);
2674 }
2675 ttl = info->key.ttl;
2676 tos = info->key.tos;
2677 label = info->key.label;
2678 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2679 }
2680 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2681 vxlan->cfg.port_max, true);
2682
2683 rcu_read_lock();
2684 if (dst->sa.sa_family == AF_INET) {
2685 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2686 struct rtable *rt;
2687 __be16 df = 0;
2688
2689 if (!ifindex)
2690 ifindex = sock4->sock->sk->sk_bound_dev_if;
2691
2692 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2693 dst->sin.sin_addr.s_addr,
2694 &local_ip.sin.sin_addr.s_addr,
2695 dst_port, src_port,
2696 dst_cache, info);
2697 if (IS_ERR(rt)) {
2698 err = PTR_ERR(rt);
2699 goto tx_error;
2700 }
2701
2702 if (!info) {
2703 /* Bypass encapsulation if the destination is local */
2704 err = encap_bypass_if_local(skb, dev, vxlan, dst,
2705 dst_port, ifindex, vni,
2706 &rt->dst, rt->rt_flags);
2707 if (err)
2708 goto out_unlock;
2709
2710 if (vxlan->cfg.df == VXLAN_DF_SET) {
2711 df = htons(IP_DF);
2712 } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) {
2713 struct ethhdr *eth = eth_hdr(skb);
2714
2715 if (ntohs(eth->h_proto) == ETH_P_IPV6 ||
2716 (ntohs(eth->h_proto) == ETH_P_IP &&
2717 old_iph->frag_off & htons(IP_DF)))
2718 df = htons(IP_DF);
2719 }
2720 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2721 df = htons(IP_DF);
2722 }
2723
2724 ndst = &rt->dst;
2725 err = skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM,
2726 netif_is_any_bridge_port(dev));
2727 if (err < 0) {
2728 goto tx_error;
2729 } else if (err) {
2730 if (info) {
2731 struct in_addr src, dst;
2732
2733 src = remote_ip.sin.sin_addr;
2734 dst = local_ip.sin.sin_addr;
2735 info->key.u.ipv4.src = src.s_addr;
2736 info->key.u.ipv4.dst = dst.s_addr;
2737 }
2738 vxlan_encap_bypass(skb, vxlan, vxlan, vni, false);
2739 dst_release(ndst);
2740 goto out_unlock;
2741 }
2742
2743 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2744 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2745 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2746 vni, md, flags, udp_sum);
2747 if (err < 0)
2748 goto tx_error;
2749
2750 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2751 dst->sin.sin_addr.s_addr, tos, ttl, df,
2752 src_port, dst_port, xnet, !udp_sum);
2753#if IS_ENABLED(CONFIG_IPV6)
2754 } else {
2755 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2756
2757 if (!ifindex)
2758 ifindex = sock6->sock->sk->sk_bound_dev_if;
2759
2760 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2761 label, &dst->sin6.sin6_addr,
2762 &local_ip.sin6.sin6_addr,
2763 dst_port, src_port,
2764 dst_cache, info);
2765 if (IS_ERR(ndst)) {
2766 err = PTR_ERR(ndst);
2767 ndst = NULL;
2768 goto tx_error;
2769 }
2770
2771 if (!info) {
2772 u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2773
2774 err = encap_bypass_if_local(skb, dev, vxlan, dst,
2775 dst_port, ifindex, vni,
2776 ndst, rt6i_flags);
2777 if (err)
2778 goto out_unlock;
2779 }
2780
2781 err = skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM,
2782 netif_is_any_bridge_port(dev));
2783 if (err < 0) {
2784 goto tx_error;
2785 } else if (err) {
2786 if (info) {
2787 struct in6_addr src, dst;
2788
2789 src = remote_ip.sin6.sin6_addr;
2790 dst = local_ip.sin6.sin6_addr;
2791 info->key.u.ipv6.src = src;
2792 info->key.u.ipv6.dst = dst;
2793 }
2794
2795 vxlan_encap_bypass(skb, vxlan, vxlan, vni, false);
2796 dst_release(ndst);
2797 goto out_unlock;
2798 }
2799
2800 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2801 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2802 skb_scrub_packet(skb, xnet);
2803 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2804 vni, md, flags, udp_sum);
2805 if (err < 0)
2806 goto tx_error;
2807
2808 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2809 &local_ip.sin6.sin6_addr,
2810 &dst->sin6.sin6_addr, tos, ttl,
2811 label, src_port, dst_port, !udp_sum);
2812#endif
2813 }
2814out_unlock:
2815 rcu_read_unlock();
2816 return;
2817
2818drop:
2819 dev->stats.tx_dropped++;
2820 dev_kfree_skb(skb);
2821 return;
2822
2823tx_error:
2824 rcu_read_unlock();
2825 if (err == -ELOOP)
2826 dev->stats.collisions++;
2827 else if (err == -ENETUNREACH)
2828 dev->stats.tx_carrier_errors++;
2829 dst_release(ndst);
2830 dev->stats.tx_errors++;
2831 kfree_skb(skb);
2832}
2833
2834static void vxlan_xmit_nh(struct sk_buff *skb, struct net_device *dev,
2835 struct vxlan_fdb *f, __be32 vni, bool did_rsc)
2836{
2837 struct vxlan_rdst nh_rdst;
2838 struct nexthop *nh;
2839 bool do_xmit;
2840 u32 hash;
2841
2842 memset(&nh_rdst, 0, sizeof(struct vxlan_rdst));
2843 hash = skb_get_hash(skb);
2844
2845 rcu_read_lock();
2846 nh = rcu_dereference(f->nh);
2847 if (!nh) {
2848 rcu_read_unlock();
2849 goto drop;
2850 }
2851 do_xmit = vxlan_fdb_nh_path_select(nh, hash, &nh_rdst);
2852 rcu_read_unlock();
2853
2854 if (likely(do_xmit))
2855 vxlan_xmit_one(skb, dev, vni, &nh_rdst, did_rsc);
2856 else
2857 goto drop;
2858
2859 return;
2860
2861drop:
2862 dev->stats.tx_dropped++;
2863 dev_kfree_skb(skb);
2864}
2865
2866/* Transmit local packets over Vxlan
2867 *
2868 * Outer IP header inherits ECN and DF from inner header.
2869 * Outer UDP destination is the VXLAN assigned port.
2870 * source port is based on hash of flow
2871 */
2872static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2873{
2874 struct vxlan_dev *vxlan = netdev_priv(dev);
2875 struct vxlan_rdst *rdst, *fdst = NULL;
2876 const struct ip_tunnel_info *info;
2877 bool did_rsc = false;
2878 struct vxlan_fdb *f;
2879 struct ethhdr *eth;
2880 __be32 vni = 0;
2881
2882 info = skb_tunnel_info(skb);
2883
2884 skb_reset_mac_header(skb);
2885
2886 if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2887 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2888 info->mode & IP_TUNNEL_INFO_TX) {
2889 vni = tunnel_id_to_key32(info->key.tun_id);
2890 } else {
2891 if (info && info->mode & IP_TUNNEL_INFO_TX)
2892 vxlan_xmit_one(skb, dev, vni, NULL, false);
2893 else
2894 kfree_skb(skb);
2895 return NETDEV_TX_OK;
2896 }
2897 }
2898
2899 if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2900 eth = eth_hdr(skb);
2901 if (ntohs(eth->h_proto) == ETH_P_ARP)
2902 return arp_reduce(dev, skb, vni);
2903#if IS_ENABLED(CONFIG_IPV6)
2904 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2905 pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2906 sizeof(struct nd_msg)) &&
2907 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2908 struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2909
2910 if (m->icmph.icmp6_code == 0 &&
2911 m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2912 return neigh_reduce(dev, skb, vni);
2913 }
2914#endif
2915 }
2916
2917 eth = eth_hdr(skb);
2918 f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2919 did_rsc = false;
2920
2921 if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2922 (ntohs(eth->h_proto) == ETH_P_IP ||
2923 ntohs(eth->h_proto) == ETH_P_IPV6)) {
2924 did_rsc = route_shortcircuit(dev, skb);
2925 if (did_rsc)
2926 f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2927 }
2928
2929 if (f == NULL) {
2930 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2931 if (f == NULL) {
2932 if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2933 !is_multicast_ether_addr(eth->h_dest))
2934 vxlan_fdb_miss(vxlan, eth->h_dest);
2935
2936 dev->stats.tx_dropped++;
2937 kfree_skb(skb);
2938 return NETDEV_TX_OK;
2939 }
2940 }
2941
2942 if (rcu_access_pointer(f->nh)) {
2943 vxlan_xmit_nh(skb, dev, f,
2944 (vni ? : vxlan->default_dst.remote_vni), did_rsc);
2945 } else {
2946 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2947 struct sk_buff *skb1;
2948
2949 if (!fdst) {
2950 fdst = rdst;
2951 continue;
2952 }
2953 skb1 = skb_clone(skb, GFP_ATOMIC);
2954 if (skb1)
2955 vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2956 }
2957 if (fdst)
2958 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2959 else
2960 kfree_skb(skb);
2961 }
2962
2963 return NETDEV_TX_OK;
2964}
2965
2966/* Walk the forwarding table and purge stale entries */
2967static void vxlan_cleanup(struct timer_list *t)
2968{
2969 struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2970 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2971 unsigned int h;
2972
2973 if (!netif_running(vxlan->dev))
2974 return;
2975
2976 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2977 struct hlist_node *p, *n;
2978
2979 spin_lock(&vxlan->hash_lock[h]);
2980 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2981 struct vxlan_fdb *f
2982 = container_of(p, struct vxlan_fdb, hlist);
2983 unsigned long timeout;
2984
2985 if (f->state & (NUD_PERMANENT | NUD_NOARP))
2986 continue;
2987
2988 if (f->flags & NTF_EXT_LEARNED)
2989 continue;
2990
2991 timeout = f->used + vxlan->cfg.age_interval * HZ;
2992 if (time_before_eq(timeout, jiffies)) {
2993 netdev_dbg(vxlan->dev,
2994 "garbage collect %pM\n",
2995 f->eth_addr);
2996 f->state = NUD_STALE;
2997 vxlan_fdb_destroy(vxlan, f, true, true);
2998 } else if (time_before(timeout, next_timer))
2999 next_timer = timeout;
3000 }
3001 spin_unlock(&vxlan->hash_lock[h]);
3002 }
3003
3004 mod_timer(&vxlan->age_timer, next_timer);
3005}
3006
3007static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
3008{
3009 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3010
3011 spin_lock(&vn->sock_lock);
3012 hlist_del_init_rcu(&vxlan->hlist4.hlist);
3013#if IS_ENABLED(CONFIG_IPV6)
3014 hlist_del_init_rcu(&vxlan->hlist6.hlist);
3015#endif
3016 spin_unlock(&vn->sock_lock);
3017}
3018
3019static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
3020 struct vxlan_dev_node *node)
3021{
3022 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3023 __be32 vni = vxlan->default_dst.remote_vni;
3024
3025 node->vxlan = vxlan;
3026 spin_lock(&vn->sock_lock);
3027 hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
3028 spin_unlock(&vn->sock_lock);
3029}
3030
3031/* Setup stats when device is created */
3032static int vxlan_init(struct net_device *dev)
3033{
3034 struct vxlan_dev *vxlan = netdev_priv(dev);
3035 int err;
3036
3037 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
3038 if (!dev->tstats)
3039 return -ENOMEM;
3040
3041 err = gro_cells_init(&vxlan->gro_cells, dev);
3042 if (err) {
3043 free_percpu(dev->tstats);
3044 return err;
3045 }
3046
3047 return 0;
3048}
3049
3050static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
3051{
3052 struct vxlan_fdb *f;
3053 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni);
3054
3055 spin_lock_bh(&vxlan->hash_lock[hash_index]);
3056 f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
3057 if (f)
3058 vxlan_fdb_destroy(vxlan, f, true, true);
3059 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
3060}
3061
3062static void vxlan_uninit(struct net_device *dev)
3063{
3064 struct vxlan_dev *vxlan = netdev_priv(dev);
3065
3066 gro_cells_destroy(&vxlan->gro_cells);
3067
3068 vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
3069
3070 free_percpu(dev->tstats);
3071}
3072
3073/* Start ageing timer and join group when device is brought up */
3074static int vxlan_open(struct net_device *dev)
3075{
3076 struct vxlan_dev *vxlan = netdev_priv(dev);
3077 int ret;
3078
3079 ret = vxlan_sock_add(vxlan);
3080 if (ret < 0)
3081 return ret;
3082
3083 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
3084 ret = vxlan_igmp_join(vxlan);
3085 if (ret == -EADDRINUSE)
3086 ret = 0;
3087 if (ret) {
3088 vxlan_sock_release(vxlan);
3089 return ret;
3090 }
3091 }
3092
3093 if (vxlan->cfg.age_interval)
3094 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
3095
3096 return ret;
3097}
3098
3099/* Purge the forwarding table */
3100static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
3101{
3102 unsigned int h;
3103
3104 for (h = 0; h < FDB_HASH_SIZE; ++h) {
3105 struct hlist_node *p, *n;
3106
3107 spin_lock_bh(&vxlan->hash_lock[h]);
3108 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
3109 struct vxlan_fdb *f
3110 = container_of(p, struct vxlan_fdb, hlist);
3111 if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
3112 continue;
3113 /* the all_zeros_mac entry is deleted at vxlan_uninit */
3114 if (is_zero_ether_addr(f->eth_addr) &&
3115 f->vni == vxlan->cfg.vni)
3116 continue;
3117 vxlan_fdb_destroy(vxlan, f, true, true);
3118 }
3119 spin_unlock_bh(&vxlan->hash_lock[h]);
3120 }
3121}
3122
3123/* Cleanup timer and forwarding table on shutdown */
3124static int vxlan_stop(struct net_device *dev)
3125{
3126 struct vxlan_dev *vxlan = netdev_priv(dev);
3127 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3128 int ret = 0;
3129
3130 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
3131 !vxlan_group_used(vn, vxlan))
3132 ret = vxlan_igmp_leave(vxlan);
3133
3134 del_timer_sync(&vxlan->age_timer);
3135
3136 vxlan_flush(vxlan, false);
3137 vxlan_sock_release(vxlan);
3138
3139 return ret;
3140}
3141
3142/* Stub, nothing needs to be done. */
3143static void vxlan_set_multicast_list(struct net_device *dev)
3144{
3145}
3146
3147static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
3148{
3149 struct vxlan_dev *vxlan = netdev_priv(dev);
3150 struct vxlan_rdst *dst = &vxlan->default_dst;
3151 struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3152 dst->remote_ifindex);
3153 bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
3154
3155 /* This check is different than dev->max_mtu, because it looks at
3156 * the lowerdev->mtu, rather than the static dev->max_mtu
3157 */
3158 if (lowerdev) {
3159 int max_mtu = lowerdev->mtu -
3160 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
3161 if (new_mtu > max_mtu)
3162 return -EINVAL;
3163 }
3164
3165 dev->mtu = new_mtu;
3166 return 0;
3167}
3168
3169static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
3170{
3171 struct vxlan_dev *vxlan = netdev_priv(dev);
3172 struct ip_tunnel_info *info = skb_tunnel_info(skb);
3173 __be16 sport, dport;
3174
3175 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
3176 vxlan->cfg.port_max, true);
3177 dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
3178
3179 if (ip_tunnel_info_af(info) == AF_INET) {
3180 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
3181 struct rtable *rt;
3182
3183 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
3184 info->key.u.ipv4.dst,
3185 &info->key.u.ipv4.src, dport, sport,
3186 &info->dst_cache, info);
3187 if (IS_ERR(rt))
3188 return PTR_ERR(rt);
3189 ip_rt_put(rt);
3190 } else {
3191#if IS_ENABLED(CONFIG_IPV6)
3192 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
3193 struct dst_entry *ndst;
3194
3195 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
3196 info->key.label, &info->key.u.ipv6.dst,
3197 &info->key.u.ipv6.src, dport, sport,
3198 &info->dst_cache, info);
3199 if (IS_ERR(ndst))
3200 return PTR_ERR(ndst);
3201 dst_release(ndst);
3202#else /* !CONFIG_IPV6 */
3203 return -EPFNOSUPPORT;
3204#endif
3205 }
3206 info->key.tp_src = sport;
3207 info->key.tp_dst = dport;
3208 return 0;
3209}
3210
3211static const struct net_device_ops vxlan_netdev_ether_ops = {
3212 .ndo_init = vxlan_init,
3213 .ndo_uninit = vxlan_uninit,
3214 .ndo_open = vxlan_open,
3215 .ndo_stop = vxlan_stop,
3216 .ndo_start_xmit = vxlan_xmit,
3217 .ndo_get_stats64 = ip_tunnel_get_stats64,
3218 .ndo_set_rx_mode = vxlan_set_multicast_list,
3219 .ndo_change_mtu = vxlan_change_mtu,
3220 .ndo_validate_addr = eth_validate_addr,
3221 .ndo_set_mac_address = eth_mac_addr,
3222 .ndo_fdb_add = vxlan_fdb_add,
3223 .ndo_fdb_del = vxlan_fdb_delete,
3224 .ndo_fdb_dump = vxlan_fdb_dump,
3225 .ndo_fdb_get = vxlan_fdb_get,
3226 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
3227 .ndo_change_proto_down = dev_change_proto_down_generic,
3228};
3229
3230static const struct net_device_ops vxlan_netdev_raw_ops = {
3231 .ndo_init = vxlan_init,
3232 .ndo_uninit = vxlan_uninit,
3233 .ndo_open = vxlan_open,
3234 .ndo_stop = vxlan_stop,
3235 .ndo_start_xmit = vxlan_xmit,
3236 .ndo_get_stats64 = ip_tunnel_get_stats64,
3237 .ndo_change_mtu = vxlan_change_mtu,
3238 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
3239};
3240
3241/* Info for udev, that this is a virtual tunnel endpoint */
3242static struct device_type vxlan_type = {
3243 .name = "vxlan",
3244};
3245
3246/* Calls the ndo_udp_tunnel_add of the caller in order to
3247 * supply the listening VXLAN udp ports. Callers are expected
3248 * to implement the ndo_udp_tunnel_add.
3249 */
3250static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
3251{
3252 struct vxlan_sock *vs;
3253 struct net *net = dev_net(dev);
3254 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3255 unsigned int i;
3256
3257 spin_lock(&vn->sock_lock);
3258 for (i = 0; i < PORT_HASH_SIZE; ++i) {
3259 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
3260 unsigned short type;
3261
3262 if (vs->flags & VXLAN_F_GPE)
3263 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
3264 else
3265 type = UDP_TUNNEL_TYPE_VXLAN;
3266
3267 if (push)
3268 udp_tunnel_push_rx_port(dev, vs->sock, type);
3269 else
3270 udp_tunnel_drop_rx_port(dev, vs->sock, type);
3271 }
3272 }
3273 spin_unlock(&vn->sock_lock);
3274}
3275
3276/* Initialize the device structure. */
3277static void vxlan_setup(struct net_device *dev)
3278{
3279 struct vxlan_dev *vxlan = netdev_priv(dev);
3280 unsigned int h;
3281
3282 eth_hw_addr_random(dev);
3283 ether_setup(dev);
3284
3285 dev->needs_free_netdev = true;
3286 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
3287
3288 dev->features |= NETIF_F_LLTX;
3289 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
3290 dev->features |= NETIF_F_RXCSUM;
3291 dev->features |= NETIF_F_GSO_SOFTWARE;
3292
3293 dev->vlan_features = dev->features;
3294 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
3295 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
3296 netif_keep_dst(dev);
3297 dev->priv_flags |= IFF_NO_QUEUE;
3298
3299 /* MTU range: 68 - 65535 */
3300 dev->min_mtu = ETH_MIN_MTU;
3301 dev->max_mtu = ETH_MAX_MTU;
3302
3303 INIT_LIST_HEAD(&vxlan->next);
3304
3305 timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
3306
3307 vxlan->dev = dev;
3308
3309 for (h = 0; h < FDB_HASH_SIZE; ++h) {
3310 spin_lock_init(&vxlan->hash_lock[h]);
3311 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
3312 }
3313}
3314
3315static void vxlan_ether_setup(struct net_device *dev)
3316{
3317 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
3318 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3319 dev->netdev_ops = &vxlan_netdev_ether_ops;
3320}
3321
3322static void vxlan_raw_setup(struct net_device *dev)
3323{
3324 dev->header_ops = NULL;
3325 dev->type = ARPHRD_NONE;
3326 dev->hard_header_len = 0;
3327 dev->addr_len = 0;
3328 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
3329 dev->netdev_ops = &vxlan_netdev_raw_ops;
3330}
3331
3332static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
3333 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
3334 [IFLA_VXLAN_GROUP] = { .len = sizeof_field(struct iphdr, daddr) },
3335 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
3336 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
3337 [IFLA_VXLAN_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) },
3338 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
3339 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
3340 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
3341 [IFLA_VXLAN_LABEL] = { .type = NLA_U32 },
3342 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
3343 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
3344 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
3345 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
3346 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
3347 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
3348 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
3349 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
3350 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 },
3351 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
3352 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 },
3353 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
3354 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
3355 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
3356 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
3357 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, },
3358 [IFLA_VXLAN_GPE] = { .type = NLA_FLAG, },
3359 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG },
3360 [IFLA_VXLAN_TTL_INHERIT] = { .type = NLA_FLAG },
3361 [IFLA_VXLAN_DF] = { .type = NLA_U8 },
3362};
3363
3364static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
3365 struct netlink_ext_ack *extack)
3366{
3367 if (tb[IFLA_ADDRESS]) {
3368 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
3369 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3370 "Provided link layer address is not Ethernet");
3371 return -EINVAL;
3372 }
3373
3374 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
3375 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3376 "Provided Ethernet address is not unicast");
3377 return -EADDRNOTAVAIL;
3378 }
3379 }
3380
3381 if (tb[IFLA_MTU]) {
3382 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
3383
3384 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
3385 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3386 "MTU must be between 68 and 65535");
3387 return -EINVAL;
3388 }
3389 }
3390
3391 if (!data) {
3392 NL_SET_ERR_MSG(extack,
3393 "Required attributes not provided to perform the operation");
3394 return -EINVAL;
3395 }
3396
3397 if (data[IFLA_VXLAN_ID]) {
3398 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
3399
3400 if (id >= VXLAN_N_VID) {
3401 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_ID],
3402 "VXLAN ID must be lower than 16777216");
3403 return -ERANGE;
3404 }
3405 }
3406
3407 if (data[IFLA_VXLAN_PORT_RANGE]) {
3408 const struct ifla_vxlan_port_range *p
3409 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3410
3411 if (ntohs(p->high) < ntohs(p->low)) {
3412 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_PORT_RANGE],
3413 "Invalid source port range");
3414 return -EINVAL;
3415 }
3416 }
3417
3418 if (data[IFLA_VXLAN_DF]) {
3419 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]);
3420
3421 if (df < 0 || df > VXLAN_DF_MAX) {
3422 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_DF],
3423 "Invalid DF attribute");
3424 return -EINVAL;
3425 }
3426 }
3427
3428 return 0;
3429}
3430
3431static void vxlan_get_drvinfo(struct net_device *netdev,
3432 struct ethtool_drvinfo *drvinfo)
3433{
3434 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
3435 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
3436}
3437
3438static int vxlan_get_link_ksettings(struct net_device *dev,
3439 struct ethtool_link_ksettings *cmd)
3440{
3441 struct vxlan_dev *vxlan = netdev_priv(dev);
3442 struct vxlan_rdst *dst = &vxlan->default_dst;
3443 struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3444 dst->remote_ifindex);
3445
3446 if (!lowerdev) {
3447 cmd->base.duplex = DUPLEX_UNKNOWN;
3448 cmd->base.port = PORT_OTHER;
3449 cmd->base.speed = SPEED_UNKNOWN;
3450
3451 return 0;
3452 }
3453
3454 return __ethtool_get_link_ksettings(lowerdev, cmd);
3455}
3456
3457static const struct ethtool_ops vxlan_ethtool_ops = {
3458 .get_drvinfo = vxlan_get_drvinfo,
3459 .get_link = ethtool_op_get_link,
3460 .get_link_ksettings = vxlan_get_link_ksettings,
3461};
3462
3463static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
3464 __be16 port, u32 flags, int ifindex)
3465{
3466 struct socket *sock;
3467 struct udp_port_cfg udp_conf;
3468 int err;
3469
3470 memset(&udp_conf, 0, sizeof(udp_conf));
3471
3472 if (ipv6) {
3473 udp_conf.family = AF_INET6;
3474 udp_conf.use_udp6_rx_checksums =
3475 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
3476 udp_conf.ipv6_v6only = 1;
3477 } else {
3478 udp_conf.family = AF_INET;
3479 }
3480
3481 udp_conf.local_udp_port = port;
3482 udp_conf.bind_ifindex = ifindex;
3483
3484 /* Open UDP socket */
3485 err = udp_sock_create(net, &udp_conf, &sock);
3486 if (err < 0)
3487 return ERR_PTR(err);
3488
3489 return sock;
3490}
3491
3492/* Create new listen socket if needed */
3493static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
3494 __be16 port, u32 flags,
3495 int ifindex)
3496{
3497 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3498 struct vxlan_sock *vs;
3499 struct socket *sock;
3500 unsigned int h;
3501 struct udp_tunnel_sock_cfg tunnel_cfg;
3502
3503 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
3504 if (!vs)
3505 return ERR_PTR(-ENOMEM);
3506
3507 for (h = 0; h < VNI_HASH_SIZE; ++h)
3508 INIT_HLIST_HEAD(&vs->vni_list[h]);
3509
3510 sock = vxlan_create_sock(net, ipv6, port, flags, ifindex);
3511 if (IS_ERR(sock)) {
3512 kfree(vs);
3513 return ERR_CAST(sock);
3514 }
3515
3516 vs->sock = sock;
3517 refcount_set(&vs->refcnt, 1);
3518 vs->flags = (flags & VXLAN_F_RCV_FLAGS);
3519
3520 spin_lock(&vn->sock_lock);
3521 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
3522 udp_tunnel_notify_add_rx_port(sock,
3523 (vs->flags & VXLAN_F_GPE) ?
3524 UDP_TUNNEL_TYPE_VXLAN_GPE :
3525 UDP_TUNNEL_TYPE_VXLAN);
3526 spin_unlock(&vn->sock_lock);
3527
3528 /* Mark socket as an encapsulation socket. */
3529 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
3530 tunnel_cfg.sk_user_data = vs;
3531 tunnel_cfg.encap_type = 1;
3532 tunnel_cfg.encap_rcv = vxlan_rcv;
3533 tunnel_cfg.encap_err_lookup = vxlan_err_lookup;
3534 tunnel_cfg.encap_destroy = NULL;
3535 tunnel_cfg.gro_receive = vxlan_gro_receive;
3536 tunnel_cfg.gro_complete = vxlan_gro_complete;
3537
3538 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
3539
3540 return vs;
3541}
3542
3543static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
3544{
3545 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3546 struct vxlan_sock *vs = NULL;
3547 struct vxlan_dev_node *node;
3548 int l3mdev_index = 0;
3549
3550 if (vxlan->cfg.remote_ifindex)
3551 l3mdev_index = l3mdev_master_upper_ifindex_by_index(
3552 vxlan->net, vxlan->cfg.remote_ifindex);
3553
3554 if (!vxlan->cfg.no_share) {
3555 spin_lock(&vn->sock_lock);
3556 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
3557 vxlan->cfg.dst_port, vxlan->cfg.flags,
3558 l3mdev_index);
3559 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
3560 spin_unlock(&vn->sock_lock);
3561 return -EBUSY;
3562 }
3563 spin_unlock(&vn->sock_lock);
3564 }
3565 if (!vs)
3566 vs = vxlan_socket_create(vxlan->net, ipv6,
3567 vxlan->cfg.dst_port, vxlan->cfg.flags,
3568 l3mdev_index);
3569 if (IS_ERR(vs))
3570 return PTR_ERR(vs);
3571#if IS_ENABLED(CONFIG_IPV6)
3572 if (ipv6) {
3573 rcu_assign_pointer(vxlan->vn6_sock, vs);
3574 node = &vxlan->hlist6;
3575 } else
3576#endif
3577 {
3578 rcu_assign_pointer(vxlan->vn4_sock, vs);
3579 node = &vxlan->hlist4;
3580 }
3581 vxlan_vs_add_dev(vs, vxlan, node);
3582 return 0;
3583}
3584
3585static int vxlan_sock_add(struct vxlan_dev *vxlan)
3586{
3587 bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
3588 bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
3589 bool ipv4 = !ipv6 || metadata;
3590 int ret = 0;
3591
3592 RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
3593#if IS_ENABLED(CONFIG_IPV6)
3594 RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
3595 if (ipv6) {
3596 ret = __vxlan_sock_add(vxlan, true);
3597 if (ret < 0 && ret != -EAFNOSUPPORT)
3598 ipv4 = false;
3599 }
3600#endif
3601 if (ipv4)
3602 ret = __vxlan_sock_add(vxlan, false);
3603 if (ret < 0)
3604 vxlan_sock_release(vxlan);
3605 return ret;
3606}
3607
3608static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
3609 struct net_device **lower,
3610 struct vxlan_dev *old,
3611 struct netlink_ext_ack *extack)
3612{
3613 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
3614 struct vxlan_dev *tmp;
3615 bool use_ipv6 = false;
3616
3617 if (conf->flags & VXLAN_F_GPE) {
3618 /* For now, allow GPE only together with
3619 * COLLECT_METADATA. This can be relaxed later; in such
3620 * case, the other side of the PtP link will have to be
3621 * provided.
3622 */
3623 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3624 !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3625 NL_SET_ERR_MSG(extack,
3626 "VXLAN GPE does not support this combination of attributes");
3627 return -EINVAL;
3628 }
3629 }
3630
3631 if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3632 /* Unless IPv6 is explicitly requested, assume IPv4 */
3633 conf->remote_ip.sa.sa_family = AF_INET;
3634 conf->saddr.sa.sa_family = AF_INET;
3635 } else if (!conf->remote_ip.sa.sa_family) {
3636 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3637 } else if (!conf->saddr.sa.sa_family) {
3638 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3639 }
3640
3641 if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3642 NL_SET_ERR_MSG(extack,
3643 "Local and remote address must be from the same family");
3644 return -EINVAL;
3645 }
3646
3647 if (vxlan_addr_multicast(&conf->saddr)) {
3648 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3649 return -EINVAL;
3650 }
3651
3652 if (conf->saddr.sa.sa_family == AF_INET6) {
3653 if (!IS_ENABLED(CONFIG_IPV6)) {
3654 NL_SET_ERR_MSG(extack,
3655 "IPv6 support not enabled in the kernel");
3656 return -EPFNOSUPPORT;
3657 }
3658 use_ipv6 = true;
3659 conf->flags |= VXLAN_F_IPV6;
3660
3661 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3662 int local_type =
3663 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3664 int remote_type =
3665 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3666
3667 if (local_type & IPV6_ADDR_LINKLOCAL) {
3668 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3669 (remote_type != IPV6_ADDR_ANY)) {
3670 NL_SET_ERR_MSG(extack,
3671 "Invalid combination of local and remote address scopes");
3672 return -EINVAL;
3673 }
3674
3675 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3676 } else {
3677 if (remote_type ==
3678 (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3679 NL_SET_ERR_MSG(extack,
3680 "Invalid combination of local and remote address scopes");
3681 return -EINVAL;
3682 }
3683
3684 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3685 }
3686 }
3687 }
3688
3689 if (conf->label && !use_ipv6) {
3690 NL_SET_ERR_MSG(extack,
3691 "Label attribute only applies to IPv6 VXLAN devices");
3692 return -EINVAL;
3693 }
3694
3695 if (conf->remote_ifindex) {
3696 struct net_device *lowerdev;
3697
3698 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3699 if (!lowerdev) {
3700 NL_SET_ERR_MSG(extack,
3701 "Invalid local interface, device not found");
3702 return -ENODEV;
3703 }
3704
3705#if IS_ENABLED(CONFIG_IPV6)
3706 if (use_ipv6) {
3707 struct inet6_dev *idev = __in6_dev_get(lowerdev);
3708 if (idev && idev->cnf.disable_ipv6) {
3709 NL_SET_ERR_MSG(extack,
3710 "IPv6 support disabled by administrator");
3711 return -EPERM;
3712 }
3713 }
3714#endif
3715
3716 *lower = lowerdev;
3717 } else {
3718 if (vxlan_addr_multicast(&conf->remote_ip)) {
3719 NL_SET_ERR_MSG(extack,
3720 "Local interface required for multicast remote destination");
3721
3722 return -EINVAL;
3723 }
3724
3725#if IS_ENABLED(CONFIG_IPV6)
3726 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3727 NL_SET_ERR_MSG(extack,
3728 "Local interface required for link-local local/remote addresses");
3729 return -EINVAL;
3730 }
3731#endif
3732
3733 *lower = NULL;
3734 }
3735
3736 if (!conf->dst_port) {
3737 if (conf->flags & VXLAN_F_GPE)
3738 conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3739 else
3740 conf->dst_port = htons(vxlan_port);
3741 }
3742
3743 if (!conf->age_interval)
3744 conf->age_interval = FDB_AGE_DEFAULT;
3745
3746 list_for_each_entry(tmp, &vn->vxlan_list, next) {
3747 if (tmp == old)
3748 continue;
3749
3750 if (tmp->cfg.vni != conf->vni)
3751 continue;
3752 if (tmp->cfg.dst_port != conf->dst_port)
3753 continue;
3754 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3755 (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3756 continue;
3757
3758 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3759 tmp->cfg.remote_ifindex != conf->remote_ifindex)
3760 continue;
3761
3762 NL_SET_ERR_MSG(extack,
3763 "A VXLAN device with the specified VNI already exists");
3764 return -EEXIST;
3765 }
3766
3767 return 0;
3768}
3769
3770static void vxlan_config_apply(struct net_device *dev,
3771 struct vxlan_config *conf,
3772 struct net_device *lowerdev,
3773 struct net *src_net,
3774 bool changelink)
3775{
3776 struct vxlan_dev *vxlan = netdev_priv(dev);
3777 struct vxlan_rdst *dst = &vxlan->default_dst;
3778 unsigned short needed_headroom = ETH_HLEN;
3779 bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3780 int max_mtu = ETH_MAX_MTU;
3781
3782 if (!changelink) {
3783 if (conf->flags & VXLAN_F_GPE)
3784 vxlan_raw_setup(dev);
3785 else
3786 vxlan_ether_setup(dev);
3787
3788 if (conf->mtu)
3789 dev->mtu = conf->mtu;
3790
3791 vxlan->net = src_net;
3792 }
3793
3794 dst->remote_vni = conf->vni;
3795
3796 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3797
3798 if (lowerdev) {
3799 dst->remote_ifindex = conf->remote_ifindex;
3800
3801 dev->gso_max_size = lowerdev->gso_max_size;
3802 dev->gso_max_segs = lowerdev->gso_max_segs;
3803
3804 needed_headroom = lowerdev->hard_header_len;
3805
3806 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3807 VXLAN_HEADROOM);
3808 if (max_mtu < ETH_MIN_MTU)
3809 max_mtu = ETH_MIN_MTU;
3810
3811 if (!changelink && !conf->mtu)
3812 dev->mtu = max_mtu;
3813 }
3814
3815 if (dev->mtu > max_mtu)
3816 dev->mtu = max_mtu;
3817
3818 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3819 needed_headroom += VXLAN6_HEADROOM;
3820 else
3821 needed_headroom += VXLAN_HEADROOM;
3822 dev->needed_headroom = needed_headroom;
3823
3824 memcpy(&vxlan->cfg, conf, sizeof(*conf));
3825}
3826
3827static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3828 struct vxlan_config *conf, bool changelink,
3829 struct netlink_ext_ack *extack)
3830{
3831 struct vxlan_dev *vxlan = netdev_priv(dev);
3832 struct net_device *lowerdev;
3833 int ret;
3834
3835 ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3836 if (ret)
3837 return ret;
3838
3839 vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3840
3841 return 0;
3842}
3843
3844static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3845 struct vxlan_config *conf,
3846 struct netlink_ext_ack *extack)
3847{
3848 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3849 struct vxlan_dev *vxlan = netdev_priv(dev);
3850 struct net_device *remote_dev = NULL;
3851 struct vxlan_fdb *f = NULL;
3852 bool unregister = false;
3853 struct vxlan_rdst *dst;
3854 int err;
3855
3856 dst = &vxlan->default_dst;
3857 err = vxlan_dev_configure(net, dev, conf, false, extack);
3858 if (err)
3859 return err;
3860
3861 dev->ethtool_ops = &vxlan_ethtool_ops;
3862
3863 /* create an fdb entry for a valid default destination */
3864 if (!vxlan_addr_any(&dst->remote_ip)) {
3865 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3866 &dst->remote_ip,
3867 NUD_REACHABLE | NUD_PERMANENT,
3868 vxlan->cfg.dst_port,
3869 dst->remote_vni,
3870 dst->remote_vni,
3871 dst->remote_ifindex,
3872 NTF_SELF, 0, &f, extack);
3873 if (err)
3874 return err;
3875 }
3876
3877 err = register_netdevice(dev);
3878 if (err)
3879 goto errout;
3880 unregister = true;
3881
3882 if (dst->remote_ifindex) {
3883 remote_dev = __dev_get_by_index(net, dst->remote_ifindex);
3884 if (!remote_dev)
3885 goto errout;
3886
3887 err = netdev_upper_dev_link(remote_dev, dev, extack);
3888 if (err)
3889 goto errout;
3890 }
3891
3892 err = rtnl_configure_link(dev, NULL);
3893 if (err)
3894 goto unlink;
3895
3896 if (f) {
3897 vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f);
3898
3899 /* notify default fdb entry */
3900 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f),
3901 RTM_NEWNEIGH, true, extack);
3902 if (err) {
3903 vxlan_fdb_destroy(vxlan, f, false, false);
3904 if (remote_dev)
3905 netdev_upper_dev_unlink(remote_dev, dev);
3906 goto unregister;
3907 }
3908 }
3909
3910 list_add(&vxlan->next, &vn->vxlan_list);
3911 if (remote_dev)
3912 dst->remote_dev = remote_dev;
3913 return 0;
3914unlink:
3915 if (remote_dev)
3916 netdev_upper_dev_unlink(remote_dev, dev);
3917errout:
3918 /* unregister_netdevice() destroys the default FDB entry with deletion
3919 * notification. But the addition notification was not sent yet, so
3920 * destroy the entry by hand here.
3921 */
3922 if (f)
3923 __vxlan_fdb_free(f);
3924unregister:
3925 if (unregister)
3926 unregister_netdevice(dev);
3927 return err;
3928}
3929
3930/* Set/clear flags based on attribute */
3931static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[],
3932 int attrtype, unsigned long mask, bool changelink,
3933 bool changelink_supported,
3934 struct netlink_ext_ack *extack)
3935{
3936 unsigned long flags;
3937
3938 if (!tb[attrtype])
3939 return 0;
3940
3941 if (changelink && !changelink_supported) {
3942 vxlan_flag_attr_error(attrtype, extack);
3943 return -EOPNOTSUPP;
3944 }
3945
3946 if (vxlan_policy[attrtype].type == NLA_FLAG)
3947 flags = conf->flags | mask;
3948 else if (nla_get_u8(tb[attrtype]))
3949 flags = conf->flags | mask;
3950 else
3951 flags = conf->flags & ~mask;
3952
3953 conf->flags = flags;
3954
3955 return 0;
3956}
3957
3958static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3959 struct net_device *dev, struct vxlan_config *conf,
3960 bool changelink, struct netlink_ext_ack *extack)
3961{
3962 struct vxlan_dev *vxlan = netdev_priv(dev);
3963 int err = 0;
3964
3965 memset(conf, 0, sizeof(*conf));
3966
3967 /* if changelink operation, start with old existing cfg */
3968 if (changelink)
3969 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3970
3971 if (data[IFLA_VXLAN_ID]) {
3972 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3973
3974 if (changelink && (vni != conf->vni)) {
3975 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI");
3976 return -EOPNOTSUPP;
3977 }
3978 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3979 }
3980
3981 if (data[IFLA_VXLAN_GROUP]) {
3982 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) {
3983 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group");
3984 return -EOPNOTSUPP;
3985 }
3986
3987 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3988 conf->remote_ip.sa.sa_family = AF_INET;
3989 } else if (data[IFLA_VXLAN_GROUP6]) {
3990 if (!IS_ENABLED(CONFIG_IPV6)) {
3991 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel");
3992 return -EPFNOSUPPORT;
3993 }
3994
3995 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) {
3996 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group");
3997 return -EOPNOTSUPP;
3998 }
3999
4000 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
4001 conf->remote_ip.sa.sa_family = AF_INET6;
4002 }
4003
4004 if (data[IFLA_VXLAN_LOCAL]) {
4005 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) {
4006 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old");
4007 return -EOPNOTSUPP;
4008 }
4009
4010 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
4011 conf->saddr.sa.sa_family = AF_INET;
4012 } else if (data[IFLA_VXLAN_LOCAL6]) {
4013 if (!IS_ENABLED(CONFIG_IPV6)) {
4014 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel");
4015 return -EPFNOSUPPORT;
4016 }
4017
4018 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) {
4019 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old");
4020 return -EOPNOTSUPP;
4021 }
4022
4023 /* TODO: respect scope id */
4024 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
4025 conf->saddr.sa.sa_family = AF_INET6;
4026 }
4027
4028 if (data[IFLA_VXLAN_LINK])
4029 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
4030
4031 if (data[IFLA_VXLAN_TOS])
4032 conf->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
4033
4034 if (data[IFLA_VXLAN_TTL])
4035 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
4036
4037 if (data[IFLA_VXLAN_TTL_INHERIT]) {
4038 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT,
4039 VXLAN_F_TTL_INHERIT, changelink, false,
4040 extack);
4041 if (err)
4042 return err;
4043
4044 }
4045
4046 if (data[IFLA_VXLAN_LABEL])
4047 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
4048 IPV6_FLOWLABEL_MASK;
4049
4050 if (data[IFLA_VXLAN_LEARNING]) {
4051 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING,
4052 VXLAN_F_LEARN, changelink, true,
4053 extack);
4054 if (err)
4055 return err;
4056 } else if (!changelink) {
4057 /* default to learn on a new device */
4058 conf->flags |= VXLAN_F_LEARN;
4059 }
4060
4061 if (data[IFLA_VXLAN_AGEING])
4062 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
4063
4064 if (data[IFLA_VXLAN_PROXY]) {
4065 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY,
4066 VXLAN_F_PROXY, changelink, false,
4067 extack);
4068 if (err)
4069 return err;
4070 }
4071
4072 if (data[IFLA_VXLAN_RSC]) {
4073 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC,
4074 VXLAN_F_RSC, changelink, false,
4075 extack);
4076 if (err)
4077 return err;
4078 }
4079
4080 if (data[IFLA_VXLAN_L2MISS]) {
4081 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS,
4082 VXLAN_F_L2MISS, changelink, false,
4083 extack);
4084 if (err)
4085 return err;
4086 }
4087
4088 if (data[IFLA_VXLAN_L3MISS]) {
4089 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS,
4090 VXLAN_F_L3MISS, changelink, false,
4091 extack);
4092 if (err)
4093 return err;
4094 }
4095
4096 if (data[IFLA_VXLAN_LIMIT]) {
4097 if (changelink) {
4098 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT],
4099 "Cannot change limit");
4100 return -EOPNOTSUPP;
4101 }
4102 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
4103 }
4104
4105 if (data[IFLA_VXLAN_COLLECT_METADATA]) {
4106 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA,
4107 VXLAN_F_COLLECT_METADATA, changelink, false,
4108 extack);
4109 if (err)
4110 return err;
4111 }
4112
4113 if (data[IFLA_VXLAN_PORT_RANGE]) {
4114 if (!changelink) {
4115 const struct ifla_vxlan_port_range *p
4116 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
4117 conf->port_min = ntohs(p->low);
4118 conf->port_max = ntohs(p->high);
4119 } else {
4120 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
4121 "Cannot change port range");
4122 return -EOPNOTSUPP;
4123 }
4124 }
4125
4126 if (data[IFLA_VXLAN_PORT]) {
4127 if (changelink) {
4128 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT],
4129 "Cannot change port");
4130 return -EOPNOTSUPP;
4131 }
4132 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
4133 }
4134
4135 if (data[IFLA_VXLAN_UDP_CSUM]) {
4136 if (changelink) {
4137 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM],
4138 "Cannot change UDP_CSUM flag");
4139 return -EOPNOTSUPP;
4140 }
4141 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
4142 conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
4143 }
4144
4145 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
4146 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4147 VXLAN_F_UDP_ZERO_CSUM6_TX, changelink,
4148 false, extack);
4149 if (err)
4150 return err;
4151 }
4152
4153 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
4154 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4155 VXLAN_F_UDP_ZERO_CSUM6_RX, changelink,
4156 false, extack);
4157 if (err)
4158 return err;
4159 }
4160
4161 if (data[IFLA_VXLAN_REMCSUM_TX]) {
4162 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX,
4163 VXLAN_F_REMCSUM_TX, changelink, false,
4164 extack);
4165 if (err)
4166 return err;
4167 }
4168
4169 if (data[IFLA_VXLAN_REMCSUM_RX]) {
4170 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX,
4171 VXLAN_F_REMCSUM_RX, changelink, false,
4172 extack);
4173 if (err)
4174 return err;
4175 }
4176
4177 if (data[IFLA_VXLAN_GBP]) {
4178 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP,
4179 VXLAN_F_GBP, changelink, false, extack);
4180 if (err)
4181 return err;
4182 }
4183
4184 if (data[IFLA_VXLAN_GPE]) {
4185 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE,
4186 VXLAN_F_GPE, changelink, false,
4187 extack);
4188 if (err)
4189 return err;
4190 }
4191
4192 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
4193 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL,
4194 VXLAN_F_REMCSUM_NOPARTIAL, changelink,
4195 false, extack);
4196 if (err)
4197 return err;
4198 }
4199
4200 if (tb[IFLA_MTU]) {
4201 if (changelink) {
4202 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
4203 "Cannot change mtu");
4204 return -EOPNOTSUPP;
4205 }
4206 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
4207 }
4208
4209 if (data[IFLA_VXLAN_DF])
4210 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]);
4211
4212 return 0;
4213}
4214
4215static int vxlan_newlink(struct net *src_net, struct net_device *dev,
4216 struct nlattr *tb[], struct nlattr *data[],
4217 struct netlink_ext_ack *extack)
4218{
4219 struct vxlan_config conf;
4220 int err;
4221
4222 err = vxlan_nl2conf(tb, data, dev, &conf, false, extack);
4223 if (err)
4224 return err;
4225
4226 return __vxlan_dev_create(src_net, dev, &conf, extack);
4227}
4228
4229static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
4230 struct nlattr *data[],
4231 struct netlink_ext_ack *extack)
4232{
4233 struct vxlan_dev *vxlan = netdev_priv(dev);
4234 struct net_device *lowerdev;
4235 struct vxlan_config conf;
4236 struct vxlan_rdst *dst;
4237 int err;
4238
4239 dst = &vxlan->default_dst;
4240 err = vxlan_nl2conf(tb, data, dev, &conf, true, extack);
4241 if (err)
4242 return err;
4243
4244 err = vxlan_config_validate(vxlan->net, &conf, &lowerdev,
4245 vxlan, extack);
4246 if (err)
4247 return err;
4248
4249 if (dst->remote_dev == lowerdev)
4250 lowerdev = NULL;
4251
4252 err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev,
4253 extack);
4254 if (err)
4255 return err;
4256
4257 /* handle default dst entry */
4258 if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) {
4259 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni);
4260
4261 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4262 if (!vxlan_addr_any(&conf.remote_ip)) {
4263 err = vxlan_fdb_update(vxlan, all_zeros_mac,
4264 &conf.remote_ip,
4265 NUD_REACHABLE | NUD_PERMANENT,
4266 NLM_F_APPEND | NLM_F_CREATE,
4267 vxlan->cfg.dst_port,
4268 conf.vni, conf.vni,
4269 conf.remote_ifindex,
4270 NTF_SELF, 0, true, extack);
4271 if (err) {
4272 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4273 netdev_adjacent_change_abort(dst->remote_dev,
4274 lowerdev, dev);
4275 return err;
4276 }
4277 }
4278 if (!vxlan_addr_any(&dst->remote_ip))
4279 __vxlan_fdb_delete(vxlan, all_zeros_mac,
4280 dst->remote_ip,
4281 vxlan->cfg.dst_port,
4282 dst->remote_vni,
4283 dst->remote_vni,
4284 dst->remote_ifindex,
4285 true);
4286 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4287 }
4288
4289 if (conf.age_interval != vxlan->cfg.age_interval)
4290 mod_timer(&vxlan->age_timer, jiffies);
4291
4292 netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev);
4293 if (lowerdev && lowerdev != dst->remote_dev)
4294 dst->remote_dev = lowerdev;
4295 vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true);
4296 return 0;
4297}
4298
4299static void vxlan_dellink(struct net_device *dev, struct list_head *head)
4300{
4301 struct vxlan_dev *vxlan = netdev_priv(dev);
4302
4303 vxlan_flush(vxlan, true);
4304
4305 list_del(&vxlan->next);
4306 unregister_netdevice_queue(dev, head);
4307 if (vxlan->default_dst.remote_dev)
4308 netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev);
4309}
4310
4311static size_t vxlan_get_size(const struct net_device *dev)
4312{
4313
4314 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
4315 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
4316 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
4317 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
4318 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
4319 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL_INHERIT */
4320 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
4321 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_DF */
4322 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
4323 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
4324 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
4325 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
4326 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
4327 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
4328 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */
4329 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
4330 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
4331 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
4332 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
4333 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
4334 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
4335 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
4336 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
4337 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
4338 0;
4339}
4340
4341static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
4342{
4343 const struct vxlan_dev *vxlan = netdev_priv(dev);
4344 const struct vxlan_rdst *dst = &vxlan->default_dst;
4345 struct ifla_vxlan_port_range ports = {
4346 .low = htons(vxlan->cfg.port_min),
4347 .high = htons(vxlan->cfg.port_max),
4348 };
4349
4350 if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
4351 goto nla_put_failure;
4352
4353 if (!vxlan_addr_any(&dst->remote_ip)) {
4354 if (dst->remote_ip.sa.sa_family == AF_INET) {
4355 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
4356 dst->remote_ip.sin.sin_addr.s_addr))
4357 goto nla_put_failure;
4358#if IS_ENABLED(CONFIG_IPV6)
4359 } else {
4360 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
4361 &dst->remote_ip.sin6.sin6_addr))
4362 goto nla_put_failure;
4363#endif
4364 }
4365 }
4366
4367 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
4368 goto nla_put_failure;
4369
4370 if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
4371 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
4372 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
4373 vxlan->cfg.saddr.sin.sin_addr.s_addr))
4374 goto nla_put_failure;
4375#if IS_ENABLED(CONFIG_IPV6)
4376 } else {
4377 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
4378 &vxlan->cfg.saddr.sin6.sin6_addr))
4379 goto nla_put_failure;
4380#endif
4381 }
4382 }
4383
4384 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
4385 nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
4386 !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
4387 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
4388 nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) ||
4389 nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
4390 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
4391 !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
4392 nla_put_u8(skb, IFLA_VXLAN_PROXY,
4393 !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
4394 nla_put_u8(skb, IFLA_VXLAN_RSC,
4395 !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
4396 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
4397 !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
4398 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
4399 !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
4400 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
4401 !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
4402 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
4403 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
4404 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
4405 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
4406 !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
4407 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4408 !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
4409 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4410 !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
4411 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
4412 !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
4413 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
4414 !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
4415 goto nla_put_failure;
4416
4417 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
4418 goto nla_put_failure;
4419
4420 if (vxlan->cfg.flags & VXLAN_F_GBP &&
4421 nla_put_flag(skb, IFLA_VXLAN_GBP))
4422 goto nla_put_failure;
4423
4424 if (vxlan->cfg.flags & VXLAN_F_GPE &&
4425 nla_put_flag(skb, IFLA_VXLAN_GPE))
4426 goto nla_put_failure;
4427
4428 if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
4429 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
4430 goto nla_put_failure;
4431
4432 return 0;
4433
4434nla_put_failure:
4435 return -EMSGSIZE;
4436}
4437
4438static struct net *vxlan_get_link_net(const struct net_device *dev)
4439{
4440 struct vxlan_dev *vxlan = netdev_priv(dev);
4441
4442 return vxlan->net;
4443}
4444
4445static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
4446 .kind = "vxlan",
4447 .maxtype = IFLA_VXLAN_MAX,
4448 .policy = vxlan_policy,
4449 .priv_size = sizeof(struct vxlan_dev),
4450 .setup = vxlan_setup,
4451 .validate = vxlan_validate,
4452 .newlink = vxlan_newlink,
4453 .changelink = vxlan_changelink,
4454 .dellink = vxlan_dellink,
4455 .get_size = vxlan_get_size,
4456 .fill_info = vxlan_fill_info,
4457 .get_link_net = vxlan_get_link_net,
4458};
4459
4460struct net_device *vxlan_dev_create(struct net *net, const char *name,
4461 u8 name_assign_type,
4462 struct vxlan_config *conf)
4463{
4464 struct nlattr *tb[IFLA_MAX + 1];
4465 struct net_device *dev;
4466 int err;
4467
4468 memset(&tb, 0, sizeof(tb));
4469
4470 dev = rtnl_create_link(net, name, name_assign_type,
4471 &vxlan_link_ops, tb, NULL);
4472 if (IS_ERR(dev))
4473 return dev;
4474
4475 err = __vxlan_dev_create(net, dev, conf, NULL);
4476 if (err < 0) {
4477 free_netdev(dev);
4478 return ERR_PTR(err);
4479 }
4480
4481 err = rtnl_configure_link(dev, NULL);
4482 if (err < 0) {
4483 LIST_HEAD(list_kill);
4484
4485 vxlan_dellink(dev, &list_kill);
4486 unregister_netdevice_many(&list_kill);
4487 return ERR_PTR(err);
4488 }
4489
4490 return dev;
4491}
4492EXPORT_SYMBOL_GPL(vxlan_dev_create);
4493
4494static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
4495 struct net_device *dev)
4496{
4497 struct vxlan_dev *vxlan, *next;
4498 LIST_HEAD(list_kill);
4499
4500 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4501 struct vxlan_rdst *dst = &vxlan->default_dst;
4502
4503 /* In case we created vxlan device with carrier
4504 * and we loose the carrier due to module unload
4505 * we also need to remove vxlan device. In other
4506 * cases, it's not necessary and remote_ifindex
4507 * is 0 here, so no matches.
4508 */
4509 if (dst->remote_ifindex == dev->ifindex)
4510 vxlan_dellink(vxlan->dev, &list_kill);
4511 }
4512
4513 unregister_netdevice_many(&list_kill);
4514}
4515
4516static int vxlan_netdevice_event(struct notifier_block *unused,
4517 unsigned long event, void *ptr)
4518{
4519 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4520 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
4521
4522 if (event == NETDEV_UNREGISTER) {
4523 if (!dev->udp_tunnel_nic_info)
4524 vxlan_offload_rx_ports(dev, false);
4525 vxlan_handle_lowerdev_unregister(vn, dev);
4526 } else if (event == NETDEV_REGISTER) {
4527 if (!dev->udp_tunnel_nic_info)
4528 vxlan_offload_rx_ports(dev, true);
4529 } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
4530 event == NETDEV_UDP_TUNNEL_DROP_INFO) {
4531 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
4532 }
4533
4534 return NOTIFY_DONE;
4535}
4536
4537static struct notifier_block vxlan_notifier_block __read_mostly = {
4538 .notifier_call = vxlan_netdevice_event,
4539};
4540
4541static void
4542vxlan_fdb_offloaded_set(struct net_device *dev,
4543 struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4544{
4545 struct vxlan_dev *vxlan = netdev_priv(dev);
4546 struct vxlan_rdst *rdst;
4547 struct vxlan_fdb *f;
4548 u32 hash_index;
4549
4550 hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4551
4552 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4553
4554 f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4555 if (!f)
4556 goto out;
4557
4558 rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip,
4559 fdb_info->remote_port,
4560 fdb_info->remote_vni,
4561 fdb_info->remote_ifindex);
4562 if (!rdst)
4563 goto out;
4564
4565 rdst->offloaded = fdb_info->offloaded;
4566
4567out:
4568 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4569}
4570
4571static int
4572vxlan_fdb_external_learn_add(struct net_device *dev,
4573 struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4574{
4575 struct vxlan_dev *vxlan = netdev_priv(dev);
4576 struct netlink_ext_ack *extack;
4577 u32 hash_index;
4578 int err;
4579
4580 hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4581 extack = switchdev_notifier_info_to_extack(&fdb_info->info);
4582
4583 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4584 err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip,
4585 NUD_REACHABLE,
4586 NLM_F_CREATE | NLM_F_REPLACE,
4587 fdb_info->remote_port,
4588 fdb_info->vni,
4589 fdb_info->remote_vni,
4590 fdb_info->remote_ifindex,
4591 NTF_USE | NTF_SELF | NTF_EXT_LEARNED,
4592 0, false, extack);
4593 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4594
4595 return err;
4596}
4597
4598static int
4599vxlan_fdb_external_learn_del(struct net_device *dev,
4600 struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4601{
4602 struct vxlan_dev *vxlan = netdev_priv(dev);
4603 struct vxlan_fdb *f;
4604 u32 hash_index;
4605 int err = 0;
4606
4607 hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4608 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4609
4610 f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4611 if (!f)
4612 err = -ENOENT;
4613 else if (f->flags & NTF_EXT_LEARNED)
4614 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr,
4615 fdb_info->remote_ip,
4616 fdb_info->remote_port,
4617 fdb_info->vni,
4618 fdb_info->remote_vni,
4619 fdb_info->remote_ifindex,
4620 false);
4621
4622 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4623
4624 return err;
4625}
4626
4627static int vxlan_switchdev_event(struct notifier_block *unused,
4628 unsigned long event, void *ptr)
4629{
4630 struct net_device *dev = switchdev_notifier_info_to_dev(ptr);
4631 struct switchdev_notifier_vxlan_fdb_info *fdb_info;
4632 int err = 0;
4633
4634 switch (event) {
4635 case SWITCHDEV_VXLAN_FDB_OFFLOADED:
4636 vxlan_fdb_offloaded_set(dev, ptr);
4637 break;
4638 case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE:
4639 fdb_info = ptr;
4640 err = vxlan_fdb_external_learn_add(dev, fdb_info);
4641 if (err) {
4642 err = notifier_from_errno(err);
4643 break;
4644 }
4645 fdb_info->offloaded = true;
4646 vxlan_fdb_offloaded_set(dev, fdb_info);
4647 break;
4648 case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE:
4649 fdb_info = ptr;
4650 err = vxlan_fdb_external_learn_del(dev, fdb_info);
4651 if (err) {
4652 err = notifier_from_errno(err);
4653 break;
4654 }
4655 fdb_info->offloaded = false;
4656 vxlan_fdb_offloaded_set(dev, fdb_info);
4657 break;
4658 }
4659
4660 return err;
4661}
4662
4663static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = {
4664 .notifier_call = vxlan_switchdev_event,
4665};
4666
4667static void vxlan_fdb_nh_flush(struct nexthop *nh)
4668{
4669 struct vxlan_fdb *fdb;
4670 struct vxlan_dev *vxlan;
4671 u32 hash_index;
4672
4673 rcu_read_lock();
4674 list_for_each_entry_rcu(fdb, &nh->fdb_list, nh_list) {
4675 vxlan = rcu_dereference(fdb->vdev);
4676 WARN_ON(!vxlan);
4677 hash_index = fdb_head_index(vxlan, fdb->eth_addr,
4678 vxlan->default_dst.remote_vni);
4679 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4680 if (!hlist_unhashed(&fdb->hlist))
4681 vxlan_fdb_destroy(vxlan, fdb, false, false);
4682 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4683 }
4684 rcu_read_unlock();
4685}
4686
4687static int vxlan_nexthop_event(struct notifier_block *nb,
4688 unsigned long event, void *ptr)
4689{
4690 struct nexthop *nh = ptr;
4691
4692 if (!nh || event != NEXTHOP_EVENT_DEL)
4693 return NOTIFY_DONE;
4694
4695 vxlan_fdb_nh_flush(nh);
4696
4697 return NOTIFY_DONE;
4698}
4699
4700static struct notifier_block vxlan_nexthop_notifier_block __read_mostly = {
4701 .notifier_call = vxlan_nexthop_event,
4702};
4703
4704static __net_init int vxlan_init_net(struct net *net)
4705{
4706 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4707 unsigned int h;
4708
4709 INIT_LIST_HEAD(&vn->vxlan_list);
4710 spin_lock_init(&vn->sock_lock);
4711
4712 for (h = 0; h < PORT_HASH_SIZE; ++h)
4713 INIT_HLIST_HEAD(&vn->sock_list[h]);
4714
4715 return register_nexthop_notifier(net, &vxlan_nexthop_notifier_block);
4716}
4717
4718static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
4719{
4720 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4721 struct vxlan_dev *vxlan, *next;
4722 struct net_device *dev, *aux;
4723 unsigned int h;
4724
4725 for_each_netdev_safe(net, dev, aux)
4726 if (dev->rtnl_link_ops == &vxlan_link_ops)
4727 unregister_netdevice_queue(dev, head);
4728
4729 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4730 /* If vxlan->dev is in the same netns, it has already been added
4731 * to the list by the previous loop.
4732 */
4733 if (!net_eq(dev_net(vxlan->dev), net))
4734 unregister_netdevice_queue(vxlan->dev, head);
4735 }
4736
4737 for (h = 0; h < PORT_HASH_SIZE; ++h)
4738 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
4739}
4740
4741static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
4742{
4743 struct net *net;
4744 LIST_HEAD(list);
4745
4746 rtnl_lock();
4747 list_for_each_entry(net, net_list, exit_list)
4748 unregister_nexthop_notifier(net, &vxlan_nexthop_notifier_block);
4749 list_for_each_entry(net, net_list, exit_list)
4750 vxlan_destroy_tunnels(net, &list);
4751
4752 unregister_netdevice_many(&list);
4753 rtnl_unlock();
4754}
4755
4756static struct pernet_operations vxlan_net_ops = {
4757 .init = vxlan_init_net,
4758 .exit_batch = vxlan_exit_batch_net,
4759 .id = &vxlan_net_id,
4760 .size = sizeof(struct vxlan_net),
4761};
4762
4763static int __init vxlan_init_module(void)
4764{
4765 int rc;
4766
4767 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
4768
4769 rc = register_pernet_subsys(&vxlan_net_ops);
4770 if (rc)
4771 goto out1;
4772
4773 rc = register_netdevice_notifier(&vxlan_notifier_block);
4774 if (rc)
4775 goto out2;
4776
4777 rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block);
4778 if (rc)
4779 goto out3;
4780
4781 rc = rtnl_link_register(&vxlan_link_ops);
4782 if (rc)
4783 goto out4;
4784
4785 return 0;
4786out4:
4787 unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4788out3:
4789 unregister_netdevice_notifier(&vxlan_notifier_block);
4790out2:
4791 unregister_pernet_subsys(&vxlan_net_ops);
4792out1:
4793 return rc;
4794}
4795late_initcall(vxlan_init_module);
4796
4797static void __exit vxlan_cleanup_module(void)
4798{
4799 rtnl_link_unregister(&vxlan_link_ops);
4800 unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4801 unregister_netdevice_notifier(&vxlan_notifier_block);
4802 unregister_pernet_subsys(&vxlan_net_ops);
4803 /* rcu_barrier() is called by netns */
4804}
4805module_exit(vxlan_cleanup_module);
4806
4807MODULE_LICENSE("GPL");
4808MODULE_VERSION(VXLAN_VERSION);
4809MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
4810MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
4811MODULE_ALIAS_RTNL_LINK("vxlan");