?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
vxlan: support ndo_fill_metadata_dst also for IPv6
[linux-stable.git] / drivers / net / vxlan.c
blob14cfa4cdf903de27c4b1b1ba5873877468ff1632
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 #include <net/udp_tunnel.h>
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 */
65 static unsigned short vxlan_port __read_mostly = 8472;
66 module_param_named(udp_port, vxlan_port, ushort, 0444);
67 MODULE_PARM_DESC(udp_port, "Destination UDP port");
68
69 static bool log_ecn_error = true;
70 module_param(log_ecn_error, bool, 0644);
71 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
72
73 static int vxlan_net_id;
74 static struct rtnl_link_ops vxlan_link_ops;
75
76 static const u8 all_zeros_mac[ETH_ALEN];
77
78 static int vxlan_sock_add(struct vxlan_dev *vxlan);
79
80 /* per-network namespace private data for this module */
81 struct 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 */
88 struct 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 */
100 static u32 vxlan_salt __read_mostly;
101 static struct workqueue_struct *vxlan_wq;
102
103 static 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)
110 static inline
111 bool 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
121 static 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
129 static 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
137 static 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
152 static 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
163 static inline
164 bool 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
169 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
170 {
171 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
172 }
173
174 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
175 {
176 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
177 }
178
179 static 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
192 static 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 */
200 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
201 {
202 return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
203 }
204
205 /* Socket hash table head */
206 static 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 */
216 static 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
221 static 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 */
229 static 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
245 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
246 {
247 struct vxlan_dev *vxlan;
248
249 hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
250 if (vxlan->default_dst.remote_vni == id)
251 return vxlan;
252 }
253
254 return NULL;
255 }
256
257 /* Look up VNI in a per net namespace table */
258 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
259 sa_family_t family, __be16 port,
260 u32 flags)
261 {
262 struct vxlan_sock *vs;
263
264 vs = vxlan_find_sock(net, family, port, flags);
265 if (!vs)
266 return NULL;
267
268 return vxlan_vs_find_vni(vs, id);
269 }
270
271 /* Fill in neighbour message in skbuff. */
272 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
273 const struct vxlan_fdb *fdb,
274 u32 portid, u32 seq, int type, unsigned int flags,
275 const struct vxlan_rdst *rdst)
276 {
277 unsigned long now = jiffies;
278 struct nda_cacheinfo ci;
279 struct nlmsghdr *nlh;
280 struct ndmsg *ndm;
281 bool send_ip, send_eth;
282
283 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
284 if (nlh == NULL)
285 return -EMSGSIZE;
286
287 ndm = nlmsg_data(nlh);
288 memset(ndm, 0, sizeof(*ndm));
289
290 send_eth = send_ip = true;
291
292 if (type == RTM_GETNEIGH) {
293 ndm->ndm_family = AF_INET;
294 send_ip = !vxlan_addr_any(&rdst->remote_ip);
295 send_eth = !is_zero_ether_addr(fdb->eth_addr);
296 } else
297 ndm->ndm_family = AF_BRIDGE;
298 ndm->ndm_state = fdb->state;
299 ndm->ndm_ifindex = vxlan->dev->ifindex;
300 ndm->ndm_flags = fdb->flags;
301 ndm->ndm_type = RTN_UNICAST;
302
303 if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
304 nla_put_s32(skb, NDA_LINK_NETNSID,
305 peernet2id_alloc(dev_net(vxlan->dev), vxlan->net)))
306 goto nla_put_failure;
307
308 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
309 goto nla_put_failure;
310
311 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
312 goto nla_put_failure;
313
314 if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
315 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
316 goto nla_put_failure;
317 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
318 nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
319 goto nla_put_failure;
320 if (rdst->remote_ifindex &&
321 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
322 goto nla_put_failure;
323
324 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
325 ci.ndm_confirmed = 0;
326 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
327 ci.ndm_refcnt = 0;
328
329 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
330 goto nla_put_failure;
331
332 nlmsg_end(skb, nlh);
333 return 0;
334
335 nla_put_failure:
336 nlmsg_cancel(skb, nlh);
337 return -EMSGSIZE;
338 }
339
340 static inline size_t vxlan_nlmsg_size(void)
341 {
342 return NLMSG_ALIGN(sizeof(struct ndmsg))
343 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
344 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
345 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
346 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
347 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
348 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
349 + nla_total_size(sizeof(struct nda_cacheinfo));
350 }
351
352 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
353 struct vxlan_rdst *rd, int type)
354 {
355 struct net *net = dev_net(vxlan->dev);
356 struct sk_buff *skb;
357 int err = -ENOBUFS;
358
359 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
360 if (skb == NULL)
361 goto errout;
362
363 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
364 if (err < 0) {
365 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
366 WARN_ON(err == -EMSGSIZE);
367 kfree_skb(skb);
368 goto errout;
369 }
370
371 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
372 return;
373 errout:
374 if (err < 0)
375 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
376 }
377
378 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
379 {
380 struct vxlan_dev *vxlan = netdev_priv(dev);
381 struct vxlan_fdb f = {
382 .state = NUD_STALE,
383 };
384 struct vxlan_rdst remote = {
385 .remote_ip = *ipa, /* goes to NDA_DST */
386 .remote_vni = VXLAN_N_VID,
387 };
388
389 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
390 }
391
392 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
393 {
394 struct vxlan_fdb f = {
395 .state = NUD_STALE,
396 };
397 struct vxlan_rdst remote = { };
398
399 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
400
401 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
402 }
403
404 /* Hash Ethernet address */
405 static u32 eth_hash(const unsigned char *addr)
406 {
407 u64 value = get_unaligned((u64 *)addr);
408
409 /* only want 6 bytes */
410 #ifdef __BIG_ENDIAN
411 value >>= 16;
412 #else
413 value <<= 16;
414 #endif
415 return hash_64(value, FDB_HASH_BITS);
416 }
417
418 /* Hash chain to use given mac address */
419 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
420 const u8 *mac)
421 {
422 return &vxlan->fdb_head[eth_hash(mac)];
423 }
424
425 /* Look up Ethernet address in forwarding table */
426 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
427 const u8 *mac)
428 {
429 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
430 struct vxlan_fdb *f;
431
432 hlist_for_each_entry_rcu(f, head, hlist) {
433 if (ether_addr_equal(mac, f->eth_addr))
434 return f;
435 }
436
437 return NULL;
438 }
439
440 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
441 const u8 *mac)
442 {
443 struct vxlan_fdb *f;
444
445 f = __vxlan_find_mac(vxlan, mac);
446 if (f)
447 f->used = jiffies;
448
449 return f;
450 }
451
452 /* caller should hold vxlan->hash_lock */
453 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
454 union vxlan_addr *ip, __be16 port,
455 __u32 vni, __u32 ifindex)
456 {
457 struct vxlan_rdst *rd;
458
459 list_for_each_entry(rd, &f->remotes, list) {
460 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
461 rd->remote_port == port &&
462 rd->remote_vni == vni &&
463 rd->remote_ifindex == ifindex)
464 return rd;
465 }
466
467 return NULL;
468 }
469
470 /* Replace destination of unicast mac */
471 static int vxlan_fdb_replace(struct vxlan_fdb *f,
472 union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
473 {
474 struct vxlan_rdst *rd;
475
476 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
477 if (rd)
478 return 0;
479
480 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
481 if (!rd)
482 return 0;
483 rd->remote_ip = *ip;
484 rd->remote_port = port;
485 rd->remote_vni = vni;
486 rd->remote_ifindex = ifindex;
487 return 1;
488 }
489
490 /* Add/update destinations for multicast */
491 static int vxlan_fdb_append(struct vxlan_fdb *f,
492 union vxlan_addr *ip, __be16 port, __u32 vni,
493 __u32 ifindex, struct vxlan_rdst **rdp)
494 {
495 struct vxlan_rdst *rd;
496
497 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
498 if (rd)
499 return 0;
500
501 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
502 if (rd == NULL)
503 return -ENOBUFS;
504 rd->remote_ip = *ip;
505 rd->remote_port = port;
506 rd->remote_vni = vni;
507 rd->remote_ifindex = ifindex;
508
509 list_add_tail_rcu(&rd->list, &f->remotes);
510
511 *rdp = rd;
512 return 1;
513 }
514
515 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
516 unsigned int off,
517 struct vxlanhdr *vh, size_t hdrlen,
518 u32 data, struct gro_remcsum *grc,
519 bool nopartial)
520 {
521 size_t start, offset;
522
523 if (skb->remcsum_offload)
524 return vh;
525
526 if (!NAPI_GRO_CB(skb)->csum_valid)
527 return NULL;
528
529 start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
530 offset = start + ((data & VXLAN_RCO_UDP) ?
531 offsetof(struct udphdr, check) :
532 offsetof(struct tcphdr, check));
533
534 vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
535 start, offset, grc, nopartial);
536
537 skb->remcsum_offload = 1;
538
539 return vh;
540 }
541
542 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
543 struct sk_buff *skb,
544 struct udp_offload *uoff)
545 {
546 struct sk_buff *p, **pp = NULL;
547 struct vxlanhdr *vh, *vh2;
548 unsigned int hlen, off_vx;
549 int flush = 1;
550 struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
551 udp_offloads);
552 u32 flags;
553 struct gro_remcsum grc;
554
555 skb_gro_remcsum_init(&grc);
556
557 off_vx = skb_gro_offset(skb);
558 hlen = off_vx + sizeof(*vh);
559 vh = skb_gro_header_fast(skb, off_vx);
560 if (skb_gro_header_hard(skb, hlen)) {
561 vh = skb_gro_header_slow(skb, hlen, off_vx);
562 if (unlikely(!vh))
563 goto out;
564 }
565
566 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
567
568 flags = ntohl(vh->vx_flags);
569
570 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
571 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
572 ntohl(vh->vx_vni), &grc,
573 !!(vs->flags &
574 VXLAN_F_REMCSUM_NOPARTIAL));
575
576 if (!vh)
577 goto out;
578 }
579
580 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
581
582 flush = 0;
583
584 for (p = *head; p; p = p->next) {
585 if (!NAPI_GRO_CB(p)->same_flow)
586 continue;
587
588 vh2 = (struct vxlanhdr *)(p->data + off_vx);
589 if (vh->vx_flags != vh2->vx_flags ||
590 vh->vx_vni != vh2->vx_vni) {
591 NAPI_GRO_CB(p)->same_flow = 0;
592 continue;
593 }
594 }
595
596 pp = eth_gro_receive(head, skb);
597
598 out:
599 skb_gro_remcsum_cleanup(skb, &grc);
600 NAPI_GRO_CB(skb)->flush |= flush;
601
602 return pp;
603 }
604
605 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
606 struct udp_offload *uoff)
607 {
608 udp_tunnel_gro_complete(skb, nhoff);
609
610 return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
611 }
612
613 /* Notify netdevs that UDP port started listening */
614 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
615 {
616 struct net_device *dev;
617 struct sock *sk = vs->sock->sk;
618 struct net *net = sock_net(sk);
619 sa_family_t sa_family = vxlan_get_sk_family(vs);
620 __be16 port = inet_sk(sk)->inet_sport;
621 int err;
622
623 if (sa_family == AF_INET) {
624 err = udp_add_offload(&vs->udp_offloads);
625 if (err)
626 pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
627 }
628
629 rcu_read_lock();
630 for_each_netdev_rcu(net, dev) {
631 if (dev->netdev_ops->ndo_add_vxlan_port)
632 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
633 port);
634 }
635 rcu_read_unlock();
636 }
637
638 /* Notify netdevs that UDP port is no more listening */
639 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
640 {
641 struct net_device *dev;
642 struct sock *sk = vs->sock->sk;
643 struct net *net = sock_net(sk);
644 sa_family_t sa_family = vxlan_get_sk_family(vs);
645 __be16 port = inet_sk(sk)->inet_sport;
646
647 rcu_read_lock();
648 for_each_netdev_rcu(net, dev) {
649 if (dev->netdev_ops->ndo_del_vxlan_port)
650 dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
651 port);
652 }
653 rcu_read_unlock();
654
655 if (sa_family == AF_INET)
656 udp_del_offload(&vs->udp_offloads);
657 }
658
659 /* Add new entry to forwarding table -- assumes lock held */
660 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
661 const u8 *mac, union vxlan_addr *ip,
662 __u16 state, __u16 flags,
663 __be16 port, __u32 vni, __u32 ifindex,
664 __u8 ndm_flags)
665 {
666 struct vxlan_rdst *rd = NULL;
667 struct vxlan_fdb *f;
668 int notify = 0;
669
670 f = __vxlan_find_mac(vxlan, mac);
671 if (f) {
672 if (flags & NLM_F_EXCL) {
673 netdev_dbg(vxlan->dev,
674 "lost race to create %pM\n", mac);
675 return -EEXIST;
676 }
677 if (f->state != state) {
678 f->state = state;
679 f->updated = jiffies;
680 notify = 1;
681 }
682 if (f->flags != ndm_flags) {
683 f->flags = ndm_flags;
684 f->updated = jiffies;
685 notify = 1;
686 }
687 if ((flags & NLM_F_REPLACE)) {
688 /* Only change unicasts */
689 if (!(is_multicast_ether_addr(f->eth_addr) ||
690 is_zero_ether_addr(f->eth_addr))) {
691 notify |= vxlan_fdb_replace(f, ip, port, vni,
692 ifindex);
693 } else
694 return -EOPNOTSUPP;
695 }
696 if ((flags & NLM_F_APPEND) &&
697 (is_multicast_ether_addr(f->eth_addr) ||
698 is_zero_ether_addr(f->eth_addr))) {
699 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
700 &rd);
701
702 if (rc < 0)
703 return rc;
704 notify |= rc;
705 }
706 } else {
707 if (!(flags & NLM_F_CREATE))
708 return -ENOENT;
709
710 if (vxlan->cfg.addrmax &&
711 vxlan->addrcnt >= vxlan->cfg.addrmax)
712 return -ENOSPC;
713
714 /* Disallow replace to add a multicast entry */
715 if ((flags & NLM_F_REPLACE) &&
716 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
717 return -EOPNOTSUPP;
718
719 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
720 f = kmalloc(sizeof(*f), GFP_ATOMIC);
721 if (!f)
722 return -ENOMEM;
723
724 notify = 1;
725 f->state = state;
726 f->flags = ndm_flags;
727 f->updated = f->used = jiffies;
728 INIT_LIST_HEAD(&f->remotes);
729 memcpy(f->eth_addr, mac, ETH_ALEN);
730
731 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
732
733 ++vxlan->addrcnt;
734 hlist_add_head_rcu(&f->hlist,
735 vxlan_fdb_head(vxlan, mac));
736 }
737
738 if (notify) {
739 if (rd == NULL)
740 rd = first_remote_rtnl(f);
741 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
742 }
743
744 return 0;
745 }
746
747 static void vxlan_fdb_free(struct rcu_head *head)
748 {
749 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
750 struct vxlan_rdst *rd, *nd;
751
752 list_for_each_entry_safe(rd, nd, &f->remotes, list)
753 kfree(rd);
754 kfree(f);
755 }
756
757 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
758 {
759 netdev_dbg(vxlan->dev,
760 "delete %pM\n", f->eth_addr);
761
762 --vxlan->addrcnt;
763 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
764
765 hlist_del_rcu(&f->hlist);
766 call_rcu(&f->rcu, vxlan_fdb_free);
767 }
768
769 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
770 union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
771 {
772 struct net *net = dev_net(vxlan->dev);
773 int err;
774
775 if (tb[NDA_DST]) {
776 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
777 if (err)
778 return err;
779 } else {
780 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
781 if (remote->sa.sa_family == AF_INET) {
782 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
783 ip->sa.sa_family = AF_INET;
784 #if IS_ENABLED(CONFIG_IPV6)
785 } else {
786 ip->sin6.sin6_addr = in6addr_any;
787 ip->sa.sa_family = AF_INET6;
788 #endif
789 }
790 }
791
792 if (tb[NDA_PORT]) {
793 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
794 return -EINVAL;
795 *port = nla_get_be16(tb[NDA_PORT]);
796 } else {
797 *port = vxlan->cfg.dst_port;
798 }
799
800 if (tb[NDA_VNI]) {
801 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
802 return -EINVAL;
803 *vni = nla_get_u32(tb[NDA_VNI]);
804 } else {
805 *vni = vxlan->default_dst.remote_vni;
806 }
807
808 if (tb[NDA_IFINDEX]) {
809 struct net_device *tdev;
810
811 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
812 return -EINVAL;
813 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
814 tdev = __dev_get_by_index(net, *ifindex);
815 if (!tdev)
816 return -EADDRNOTAVAIL;
817 } else {
818 *ifindex = 0;
819 }
820
821 return 0;
822 }
823
824 /* Add static entry (via netlink) */
825 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
826 struct net_device *dev,
827 const unsigned char *addr, u16 vid, u16 flags)
828 {
829 struct vxlan_dev *vxlan = netdev_priv(dev);
830 /* struct net *net = dev_net(vxlan->dev); */
831 union vxlan_addr ip;
832 __be16 port;
833 u32 vni, ifindex;
834 int err;
835
836 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
837 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
838 ndm->ndm_state);
839 return -EINVAL;
840 }
841
842 if (tb[NDA_DST] == NULL)
843 return -EINVAL;
844
845 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
846 if (err)
847 return err;
848
849 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
850 return -EAFNOSUPPORT;
851
852 spin_lock_bh(&vxlan->hash_lock);
853 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
854 port, vni, ifindex, ndm->ndm_flags);
855 spin_unlock_bh(&vxlan->hash_lock);
856
857 return err;
858 }
859
860 /* Delete entry (via netlink) */
861 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
862 struct net_device *dev,
863 const unsigned char *addr, u16 vid)
864 {
865 struct vxlan_dev *vxlan = netdev_priv(dev);
866 struct vxlan_fdb *f;
867 struct vxlan_rdst *rd = NULL;
868 union vxlan_addr ip;
869 __be16 port;
870 u32 vni, ifindex;
871 int err;
872
873 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
874 if (err)
875 return err;
876
877 err = -ENOENT;
878
879 spin_lock_bh(&vxlan->hash_lock);
880 f = vxlan_find_mac(vxlan, addr);
881 if (!f)
882 goto out;
883
884 if (!vxlan_addr_any(&ip)) {
885 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
886 if (!rd)
887 goto out;
888 }
889
890 err = 0;
891
892 /* remove a destination if it's not the only one on the list,
893 * otherwise destroy the fdb entry
894 */
895 if (rd && !list_is_singular(&f->remotes)) {
896 list_del_rcu(&rd->list);
897 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
898 kfree_rcu(rd, rcu);
899 goto out;
900 }
901
902 vxlan_fdb_destroy(vxlan, f);
903
904 out:
905 spin_unlock_bh(&vxlan->hash_lock);
906
907 return err;
908 }
909
910 /* Dump forwarding table */
911 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
912 struct net_device *dev,
913 struct net_device *filter_dev, int idx)
914 {
915 struct vxlan_dev *vxlan = netdev_priv(dev);
916 unsigned int h;
917
918 for (h = 0; h < FDB_HASH_SIZE; ++h) {
919 struct vxlan_fdb *f;
920 int err;
921
922 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
923 struct vxlan_rdst *rd;
924
925 list_for_each_entry_rcu(rd, &f->remotes, list) {
926 if (idx < cb->args[0])
927 goto skip;
928
929 err = vxlan_fdb_info(skb, vxlan, f,
930 NETLINK_CB(cb->skb).portid,
931 cb->nlh->nlmsg_seq,
932 RTM_NEWNEIGH,
933 NLM_F_MULTI, rd);
934 if (err < 0)
935 goto out;
936 skip:
937 ++idx;
938 }
939 }
940 }
941 out:
942 return idx;
943 }
944
945 /* Watch incoming packets to learn mapping between Ethernet address
946 * and Tunnel endpoint.
947 * Return true if packet is bogus and should be dropped.
948 */
949 static bool vxlan_snoop(struct net_device *dev,
950 union vxlan_addr *src_ip, const u8 *src_mac)
951 {
952 struct vxlan_dev *vxlan = netdev_priv(dev);
953 struct vxlan_fdb *f;
954
955 f = vxlan_find_mac(vxlan, src_mac);
956 if (likely(f)) {
957 struct vxlan_rdst *rdst = first_remote_rcu(f);
958
959 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
960 return false;
961
962 /* Don't migrate static entries, drop packets */
963 if (f->state & NUD_NOARP)
964 return true;
965
966 if (net_ratelimit())
967 netdev_info(dev,
968 "%pM migrated from %pIS to %pIS\n",
969 src_mac, &rdst->remote_ip.sa, &src_ip->sa);
970
971 rdst->remote_ip = *src_ip;
972 f->updated = jiffies;
973 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
974 } else {
975 /* learned new entry */
976 spin_lock(&vxlan->hash_lock);
977
978 /* close off race between vxlan_flush and incoming packets */
979 if (netif_running(dev))
980 vxlan_fdb_create(vxlan, src_mac, src_ip,
981 NUD_REACHABLE,
982 NLM_F_EXCL|NLM_F_CREATE,
983 vxlan->cfg.dst_port,
984 vxlan->default_dst.remote_vni,
985 0, NTF_SELF);
986 spin_unlock(&vxlan->hash_lock);
987 }
988
989 return false;
990 }
991
992 /* See if multicast group is already in use by other ID */
993 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
994 {
995 struct vxlan_dev *vxlan;
996 unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
997
998 /* The vxlan_sock is only used by dev, leaving group has
999 * no effect on other vxlan devices.
1000 */
1001 if (family == AF_INET && dev->vn4_sock &&
1002 atomic_read(&dev->vn4_sock->refcnt) == 1)
1003 return false;
1004 #if IS_ENABLED(CONFIG_IPV6)
1005 if (family == AF_INET6 && dev->vn6_sock &&
1006 atomic_read(&dev->vn6_sock->refcnt) == 1)
1007 return false;
1008 #endif
1009
1010 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1011 if (!netif_running(vxlan->dev) || vxlan == dev)
1012 continue;
1013
1014 if (family == AF_INET && vxlan->vn4_sock != dev->vn4_sock)
1015 continue;
1016 #if IS_ENABLED(CONFIG_IPV6)
1017 if (family == AF_INET6 && vxlan->vn6_sock != dev->vn6_sock)
1018 continue;
1019 #endif
1020
1021 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1022 &dev->default_dst.remote_ip))
1023 continue;
1024
1025 if (vxlan->default_dst.remote_ifindex !=
1026 dev->default_dst.remote_ifindex)
1027 continue;
1028
1029 return true;
1030 }
1031
1032 return false;
1033 }
1034
1035 static void __vxlan_sock_release(struct vxlan_sock *vs)
1036 {
1037 struct vxlan_net *vn;
1038
1039 if (!vs)
1040 return;
1041 if (!atomic_dec_and_test(&vs->refcnt))
1042 return;
1043
1044 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1045 spin_lock(&vn->sock_lock);
1046 hlist_del_rcu(&vs->hlist);
1047 vxlan_notify_del_rx_port(vs);
1048 spin_unlock(&vn->sock_lock);
1049
1050 queue_work(vxlan_wq, &vs->del_work);
1051 }
1052
1053 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1054 {
1055 __vxlan_sock_release(vxlan->vn4_sock);
1056 #if IS_ENABLED(CONFIG_IPV6)
1057 __vxlan_sock_release(vxlan->vn6_sock);
1058 #endif
1059 }
1060
1061 /* Update multicast group membership when first VNI on
1062 * multicast address is brought up
1063 */
1064 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1065 {
1066 struct sock *sk;
1067 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1068 int ifindex = vxlan->default_dst.remote_ifindex;
1069 int ret = -EINVAL;
1070
1071 if (ip->sa.sa_family == AF_INET) {
1072 struct ip_mreqn mreq = {
1073 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1074 .imr_ifindex = ifindex,
1075 };
1076
1077 sk = vxlan->vn4_sock->sock->sk;
1078 lock_sock(sk);
1079 ret = ip_mc_join_group(sk, &mreq);
1080 release_sock(sk);
1081 #if IS_ENABLED(CONFIG_IPV6)
1082 } else {
1083 sk = vxlan->vn6_sock->sock->sk;
1084 lock_sock(sk);
1085 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1086 &ip->sin6.sin6_addr);
1087 release_sock(sk);
1088 #endif
1089 }
1090
1091 return ret;
1092 }
1093
1094 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1095 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1096 {
1097 struct sock *sk;
1098 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1099 int ifindex = vxlan->default_dst.remote_ifindex;
1100 int ret = -EINVAL;
1101
1102 if (ip->sa.sa_family == AF_INET) {
1103 struct ip_mreqn mreq = {
1104 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1105 .imr_ifindex = ifindex,
1106 };
1107
1108 sk = vxlan->vn4_sock->sock->sk;
1109 lock_sock(sk);
1110 ret = ip_mc_leave_group(sk, &mreq);
1111 release_sock(sk);
1112 #if IS_ENABLED(CONFIG_IPV6)
1113 } else {
1114 sk = vxlan->vn6_sock->sock->sk;
1115 lock_sock(sk);
1116 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1117 &ip->sin6.sin6_addr);
1118 release_sock(sk);
1119 #endif
1120 }
1121
1122 return ret;
1123 }
1124
1125 static struct vxlanhdr *vxlan_remcsum(struct sk_buff *skb, struct vxlanhdr *vh,
1126 size_t hdrlen, u32 data, bool nopartial)
1127 {
1128 size_t start, offset, plen;
1129
1130 if (skb->remcsum_offload)
1131 return vh;
1132
1133 start = (data & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
1134 offset = start + ((data & VXLAN_RCO_UDP) ?
1135 offsetof(struct udphdr, check) :
1136 offsetof(struct tcphdr, check));
1137
1138 plen = hdrlen + offset + sizeof(u16);
1139
1140 if (!pskb_may_pull(skb, plen))
1141 return NULL;
1142
1143 vh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1144
1145 skb_remcsum_process(skb, (void *)vh + hdrlen, start, offset,
1146 nopartial);
1147
1148 return vh;
1149 }
1150
1151 static void vxlan_rcv(struct vxlan_sock *vs, struct sk_buff *skb,
1152 struct vxlan_metadata *md, u32 vni,
1153 struct metadata_dst *tun_dst)
1154 {
1155 struct iphdr *oip = NULL;
1156 struct ipv6hdr *oip6 = NULL;
1157 struct vxlan_dev *vxlan;
1158 struct pcpu_sw_netstats *stats;
1159 union vxlan_addr saddr;
1160 int err = 0;
1161 union vxlan_addr *remote_ip;
1162
1163 /* For flow based devices, map all packets to VNI 0 */
1164 if (vs->flags & VXLAN_F_COLLECT_METADATA)
1165 vni = 0;
1166
1167 /* Is this VNI defined? */
1168 vxlan = vxlan_vs_find_vni(vs, vni);
1169 if (!vxlan)
1170 goto drop;
1171
1172 remote_ip = &vxlan->default_dst.remote_ip;
1173 skb_reset_mac_header(skb);
1174 skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1175 skb->protocol = eth_type_trans(skb, vxlan->dev);
1176 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1177
1178 /* Ignore packet loops (and multicast echo) */
1179 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1180 goto drop;
1181
1182 /* Re-examine inner Ethernet packet */
1183 if (remote_ip->sa.sa_family == AF_INET) {
1184 oip = ip_hdr(skb);
1185 saddr.sin.sin_addr.s_addr = oip->saddr;
1186 saddr.sa.sa_family = AF_INET;
1187 #if IS_ENABLED(CONFIG_IPV6)
1188 } else {
1189 oip6 = ipv6_hdr(skb);
1190 saddr.sin6.sin6_addr = oip6->saddr;
1191 saddr.sa.sa_family = AF_INET6;
1192 #endif
1193 }
1194
1195 if (tun_dst) {
1196 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1197 tun_dst = NULL;
1198 }
1199
1200 if ((vxlan->flags & VXLAN_F_LEARN) &&
1201 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1202 goto drop;
1203
1204 skb_reset_network_header(skb);
1205 /* In flow-based mode, GBP is carried in dst_metadata */
1206 if (!(vs->flags & VXLAN_F_COLLECT_METADATA))
1207 skb->mark = md->gbp;
1208
1209 if (oip6)
1210 err = IP6_ECN_decapsulate(oip6, skb);
1211 if (oip)
1212 err = IP_ECN_decapsulate(oip, skb);
1213
1214 if (unlikely(err)) {
1215 if (log_ecn_error) {
1216 if (oip6)
1217 net_info_ratelimited("non-ECT from %pI6\n",
1218 &oip6->saddr);
1219 if (oip)
1220 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1221 &oip->saddr, oip->tos);
1222 }
1223 if (err > 1) {
1224 ++vxlan->dev->stats.rx_frame_errors;
1225 ++vxlan->dev->stats.rx_errors;
1226 goto drop;
1227 }
1228 }
1229
1230 stats = this_cpu_ptr(vxlan->dev->tstats);
1231 u64_stats_update_begin(&stats->syncp);
1232 stats->rx_packets++;
1233 stats->rx_bytes += skb->len;
1234 u64_stats_update_end(&stats->syncp);
1235
1236 gro_cells_receive(&vxlan->gro_cells, skb);
1237
1238 return;
1239 drop:
1240 if (tun_dst)
1241 dst_release((struct dst_entry *)tun_dst);
1242
1243 /* Consume bad packet */
1244 kfree_skb(skb);
1245 }
1246
1247 /* Callback from net/ipv4/udp.c to receive packets */
1248 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1249 {
1250 struct metadata_dst *tun_dst = NULL;
1251 struct vxlan_sock *vs;
1252 struct vxlanhdr *vxh;
1253 u32 flags, vni;
1254 struct vxlan_metadata _md;
1255 struct vxlan_metadata *md = &_md;
1256
1257 /* Need Vxlan and inner Ethernet header to be present */
1258 if (!pskb_may_pull(skb, VXLAN_HLEN))
1259 goto error;
1260
1261 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1262 flags = ntohl(vxh->vx_flags);
1263 vni = ntohl(vxh->vx_vni);
1264
1265 if (flags & VXLAN_HF_VNI) {
1266 flags &= ~VXLAN_HF_VNI;
1267 } else {
1268 /* VNI flag always required to be set */
1269 goto bad_flags;
1270 }
1271
1272 if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1273 goto drop;
1274 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1275
1276 vs = rcu_dereference_sk_user_data(sk);
1277 if (!vs)
1278 goto drop;
1279
1280 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
1281 vxh = vxlan_remcsum(skb, vxh, sizeof(struct vxlanhdr), vni,
1282 !!(vs->flags & VXLAN_F_REMCSUM_NOPARTIAL));
1283 if (!vxh)
1284 goto drop;
1285
1286 flags &= ~VXLAN_HF_RCO;
1287 vni &= VXLAN_VNI_MASK;
1288 }
1289
1290 if (vxlan_collect_metadata(vs)) {
1291 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1292 cpu_to_be64(vni >> 8), sizeof(*md));
1293
1294 if (!tun_dst)
1295 goto drop;
1296
1297 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1298 } else {
1299 memset(md, 0, sizeof(*md));
1300 }
1301
1302 /* For backwards compatibility, only allow reserved fields to be
1303 * used by VXLAN extensions if explicitly requested.
1304 */
1305 if ((flags & VXLAN_HF_GBP) && (vs->flags & VXLAN_F_GBP)) {
1306 struct vxlanhdr_gbp *gbp;
1307
1308 gbp = (struct vxlanhdr_gbp *)vxh;
1309 md->gbp = ntohs(gbp->policy_id);
1310
1311 if (tun_dst)
1312 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1313
1314 if (gbp->dont_learn)
1315 md->gbp |= VXLAN_GBP_DONT_LEARN;
1316
1317 if (gbp->policy_applied)
1318 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1319
1320 flags &= ~VXLAN_GBP_USED_BITS;
1321 }
1322
1323 if (flags || vni & ~VXLAN_VNI_MASK) {
1324 /* If there are any unprocessed flags remaining treat
1325 * this as a malformed packet. This behavior diverges from
1326 * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1327 * in reserved fields are to be ignored. The approach here
1328 * maintains compatibility with previous stack code, and also
1329 * is more robust and provides a little more security in
1330 * adding extensions to VXLAN.
1331 */
1332
1333 goto bad_flags;
1334 }
1335
1336 vxlan_rcv(vs, skb, md, vni >> 8, tun_dst);
1337 return 0;
1338
1339 drop:
1340 /* Consume bad packet */
1341 kfree_skb(skb);
1342 return 0;
1343
1344 bad_flags:
1345 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1346 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1347
1348 error:
1349 if (tun_dst)
1350 dst_release((struct dst_entry *)tun_dst);
1351
1352 /* Return non vxlan pkt */
1353 return 1;
1354 }
1355
1356 static 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 }
1433 out:
1434 consume_skb(skb);
1435 return NETDEV_TX_OK;
1436 }
1437
1438 #if IS_ENABLED(CONFIG_IPV6)
1439 static 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_set_mac_header(reply, 0);
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_set_network_header(reply, 0);
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_set_transport_header(reply, 0);
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
1527 static 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
1591 out:
1592 consume_skb(skb);
1593 return NETDEV_TX_OK;
1594 }
1595 #endif
1596
1597 static 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
1669 static 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 |= htonl(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
1689 #if IS_ENABLED(CONFIG_IPV6)
1690 static int vxlan6_xmit_skb(struct dst_entry *dst, struct sock *sk,
1691 struct sk_buff *skb,
1692 struct net_device *dev, struct in6_addr *saddr,
1693 struct in6_addr *daddr, __u8 prio, __u8 ttl,
1694 __be16 src_port, __be16 dst_port, __be32 vni,
1695 struct vxlan_metadata *md, bool xnet, u32 vxflags)
1696 {
1697 struct vxlanhdr *vxh;
1698 int min_headroom;
1699 int err;
1700 bool udp_sum = !(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX);
1701 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1702 u16 hdrlen = sizeof(struct vxlanhdr);
1703
1704 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1705 skb->ip_summed == CHECKSUM_PARTIAL) {
1706 int csum_start = skb_checksum_start_offset(skb);
1707
1708 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1709 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1710 (skb->csum_offset == offsetof(struct udphdr, check) ||
1711 skb->csum_offset == offsetof(struct tcphdr, check))) {
1712 udp_sum = false;
1713 type |= SKB_GSO_TUNNEL_REMCSUM;
1714 }
1715 }
1716
1717 skb_scrub_packet(skb, xnet);
1718
1719 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1720 + VXLAN_HLEN + sizeof(struct ipv6hdr)
1721 + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1722
1723 /* Need space for new headers (invalidates iph ptr) */
1724 err = skb_cow_head(skb, min_headroom);
1725 if (unlikely(err)) {
1726 kfree_skb(skb);
1727 goto err;
1728 }
1729
1730 skb = vlan_hwaccel_push_inside(skb);
1731 if (WARN_ON(!skb)) {
1732 err = -ENOMEM;
1733 goto err;
1734 }
1735
1736 skb = iptunnel_handle_offloads(skb, udp_sum, type);
1737 if (IS_ERR(skb)) {
1738 err = -EINVAL;
1739 goto err;
1740 }
1741
1742 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1743 vxh->vx_flags = htonl(VXLAN_HF_VNI);
1744 vxh->vx_vni = vni;
1745
1746 if (type & SKB_GSO_TUNNEL_REMCSUM) {
1747 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1748 VXLAN_RCO_SHIFT;
1749
1750 if (skb->csum_offset == offsetof(struct udphdr, check))
1751 data |= VXLAN_RCO_UDP;
1752
1753 vxh->vx_vni |= htonl(data);
1754 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1755
1756 if (!skb_is_gso(skb)) {
1757 skb->ip_summed = CHECKSUM_NONE;
1758 skb->encapsulation = 0;
1759 }
1760 }
1761
1762 if (vxflags & VXLAN_F_GBP)
1763 vxlan_build_gbp_hdr(vxh, vxflags, md);
1764
1765 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1766
1767 udp_tunnel6_xmit_skb(dst, sk, skb, dev, saddr, daddr, prio,
1768 ttl, src_port, dst_port,
1769 !!(vxflags & VXLAN_F_UDP_ZERO_CSUM6_TX));
1770 return 0;
1771 err:
1772 dst_release(dst);
1773 return err;
1774 }
1775 #endif
1776
1777 static int vxlan_xmit_skb(struct rtable *rt, struct sock *sk, struct sk_buff *skb,
1778 __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1779 __be16 src_port, __be16 dst_port, __be32 vni,
1780 struct vxlan_metadata *md, bool xnet, u32 vxflags)
1781 {
1782 struct vxlanhdr *vxh;
1783 int min_headroom;
1784 int err;
1785 bool udp_sum = !!(vxflags & VXLAN_F_UDP_CSUM);
1786 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1787 u16 hdrlen = sizeof(struct vxlanhdr);
1788
1789 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1790 skb->ip_summed == CHECKSUM_PARTIAL) {
1791 int csum_start = skb_checksum_start_offset(skb);
1792
1793 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1794 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1795 (skb->csum_offset == offsetof(struct udphdr, check) ||
1796 skb->csum_offset == offsetof(struct tcphdr, check))) {
1797 udp_sum = false;
1798 type |= SKB_GSO_TUNNEL_REMCSUM;
1799 }
1800 }
1801
1802 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1803 + VXLAN_HLEN + sizeof(struct iphdr)
1804 + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1805
1806 /* Need space for new headers (invalidates iph ptr) */
1807 err = skb_cow_head(skb, min_headroom);
1808 if (unlikely(err)) {
1809 kfree_skb(skb);
1810 return err;
1811 }
1812
1813 skb = vlan_hwaccel_push_inside(skb);
1814 if (WARN_ON(!skb))
1815 return -ENOMEM;
1816
1817 skb = iptunnel_handle_offloads(skb, udp_sum, type);
1818 if (IS_ERR(skb))
1819 return PTR_ERR(skb);
1820
1821 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1822 vxh->vx_flags = htonl(VXLAN_HF_VNI);
1823 vxh->vx_vni = vni;
1824
1825 if (type & SKB_GSO_TUNNEL_REMCSUM) {
1826 u32 data = (skb_checksum_start_offset(skb) - hdrlen) >>
1827 VXLAN_RCO_SHIFT;
1828
1829 if (skb->csum_offset == offsetof(struct udphdr, check))
1830 data |= VXLAN_RCO_UDP;
1831
1832 vxh->vx_vni |= htonl(data);
1833 vxh->vx_flags |= htonl(VXLAN_HF_RCO);
1834
1835 if (!skb_is_gso(skb)) {
1836 skb->ip_summed = CHECKSUM_NONE;
1837 skb->encapsulation = 0;
1838 }
1839 }
1840
1841 if (vxflags & VXLAN_F_GBP)
1842 vxlan_build_gbp_hdr(vxh, vxflags, md);
1843
1844 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1845
1846 return udp_tunnel_xmit_skb(rt, sk, skb, src, dst, tos,
1847 ttl, df, src_port, dst_port, xnet,
1848 !(vxflags & VXLAN_F_UDP_CSUM));
1849 }
1850
1851 #if IS_ENABLED(CONFIG_IPV6)
1852 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1853 struct sk_buff *skb, int oif,
1854 const struct in6_addr *daddr,
1855 struct in6_addr *saddr)
1856 {
1857 struct dst_entry *ndst;
1858 struct flowi6 fl6;
1859 int err;
1860
1861 memset(&fl6, 0, sizeof(fl6));
1862 fl6.flowi6_oif = oif;
1863 fl6.daddr = *daddr;
1864 fl6.saddr = vxlan->cfg.saddr.sin6.sin6_addr;
1865 fl6.flowi6_mark = skb->mark;
1866 fl6.flowi6_proto = IPPROTO_UDP;
1867
1868 err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1869 vxlan->vn6_sock->sock->sk,
1870 &ndst, &fl6);
1871 if (err < 0)
1872 return ERR_PTR(err);
1873
1874 *saddr = fl6.saddr;
1875 return ndst;
1876 }
1877 #endif
1878
1879 /* Bypass encapsulation if the destination is local */
1880 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1881 struct vxlan_dev *dst_vxlan)
1882 {
1883 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1884 union vxlan_addr loopback;
1885 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1886 struct net_device *dev = skb->dev;
1887 int len = skb->len;
1888
1889 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1890 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1891 skb->pkt_type = PACKET_HOST;
1892 skb->encapsulation = 0;
1893 skb->dev = dst_vxlan->dev;
1894 __skb_pull(skb, skb_network_offset(skb));
1895
1896 if (remote_ip->sa.sa_family == AF_INET) {
1897 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1898 loopback.sa.sa_family = AF_INET;
1899 #if IS_ENABLED(CONFIG_IPV6)
1900 } else {
1901 loopback.sin6.sin6_addr = in6addr_loopback;
1902 loopback.sa.sa_family = AF_INET6;
1903 #endif
1904 }
1905
1906 if (dst_vxlan->flags & VXLAN_F_LEARN)
1907 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1908
1909 u64_stats_update_begin(&tx_stats->syncp);
1910 tx_stats->tx_packets++;
1911 tx_stats->tx_bytes += len;
1912 u64_stats_update_end(&tx_stats->syncp);
1913
1914 if (netif_rx(skb) == NET_RX_SUCCESS) {
1915 u64_stats_update_begin(&rx_stats->syncp);
1916 rx_stats->rx_packets++;
1917 rx_stats->rx_bytes += len;
1918 u64_stats_update_end(&rx_stats->syncp);
1919 } else {
1920 dev->stats.rx_dropped++;
1921 }
1922 }
1923
1924 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1925 struct vxlan_rdst *rdst, bool did_rsc)
1926 {
1927 struct ip_tunnel_info *info;
1928 struct vxlan_dev *vxlan = netdev_priv(dev);
1929 struct sock *sk;
1930 struct rtable *rt = NULL;
1931 const struct iphdr *old_iph;
1932 struct flowi4 fl4;
1933 union vxlan_addr *dst;
1934 union vxlan_addr remote_ip;
1935 struct vxlan_metadata _md;
1936 struct vxlan_metadata *md = &_md;
1937 __be16 src_port = 0, dst_port;
1938 u32 vni;
1939 __be16 df = 0;
1940 __u8 tos, ttl;
1941 int err;
1942 u32 flags = vxlan->flags;
1943
1944 info = skb_tunnel_info(skb);
1945
1946 if (rdst) {
1947 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1948 vni = rdst->remote_vni;
1949 dst = &rdst->remote_ip;
1950 } else {
1951 if (!info) {
1952 WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1953 dev->name);
1954 goto drop;
1955 }
1956 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1957 vni = be64_to_cpu(info->key.tun_id);
1958 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1959 if (remote_ip.sa.sa_family == AF_INET)
1960 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1961 else
1962 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1963 dst = &remote_ip;
1964 }
1965
1966 if (vxlan_addr_any(dst)) {
1967 if (did_rsc) {
1968 /* short-circuited back to local bridge */
1969 vxlan_encap_bypass(skb, vxlan, vxlan);
1970 return;
1971 }
1972 goto drop;
1973 }
1974
1975 old_iph = ip_hdr(skb);
1976
1977 ttl = vxlan->cfg.ttl;
1978 if (!ttl && vxlan_addr_multicast(dst))
1979 ttl = 1;
1980
1981 tos = vxlan->cfg.tos;
1982 if (tos == 1)
1983 tos = ip_tunnel_get_dsfield(old_iph, skb);
1984
1985 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1986 vxlan->cfg.port_max, true);
1987
1988 if (info) {
1989 if (info->key.tun_flags & TUNNEL_CSUM)
1990 flags |= VXLAN_F_UDP_CSUM;
1991 else
1992 flags &= ~VXLAN_F_UDP_CSUM;
1993
1994 ttl = info->key.ttl;
1995 tos = info->key.tos;
1996
1997 if (info->options_len)
1998 md = ip_tunnel_info_opts(info);
1999 } else {
2000 md->gbp = skb->mark;
2001 }
2002
2003 if (dst->sa.sa_family == AF_INET) {
2004 if (!vxlan->vn4_sock)
2005 goto drop;
2006 sk = vxlan->vn4_sock->sock->sk;
2007
2008 if (info && (info->key.tun_flags & TUNNEL_DONT_FRAGMENT))
2009 df = htons(IP_DF);
2010
2011 memset(&fl4, 0, sizeof(fl4));
2012 fl4.flowi4_oif = rdst ? rdst->remote_ifindex : 0;
2013 fl4.flowi4_tos = RT_TOS(tos);
2014 fl4.flowi4_mark = skb->mark;
2015 fl4.flowi4_proto = IPPROTO_UDP;
2016 fl4.daddr = dst->sin.sin_addr.s_addr;
2017 fl4.saddr = vxlan->cfg.saddr.sin.sin_addr.s_addr;
2018
2019 rt = ip_route_output_key(vxlan->net, &fl4);
2020 if (IS_ERR(rt)) {
2021 netdev_dbg(dev, "no route to %pI4\n",
2022 &dst->sin.sin_addr.s_addr);
2023 dev->stats.tx_carrier_errors++;
2024 goto tx_error;
2025 }
2026
2027 if (rt->dst.dev == dev) {
2028 netdev_dbg(dev, "circular route to %pI4\n",
2029 &dst->sin.sin_addr.s_addr);
2030 dev->stats.collisions++;
2031 goto rt_tx_error;
2032 }
2033
2034 /* Bypass encapsulation if the destination is local */
2035 if (rt->rt_flags & RTCF_LOCAL &&
2036 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2037 struct vxlan_dev *dst_vxlan;
2038
2039 ip_rt_put(rt);
2040 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2041 dst->sa.sa_family, dst_port,
2042 vxlan->flags);
2043 if (!dst_vxlan)
2044 goto tx_error;
2045 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2046 return;
2047 }
2048
2049 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2050 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2051 err = vxlan_xmit_skb(rt, sk, skb, fl4.saddr,
2052 dst->sin.sin_addr.s_addr, tos, ttl, df,
2053 src_port, dst_port, htonl(vni << 8), md,
2054 !net_eq(vxlan->net, dev_net(vxlan->dev)),
2055 flags);
2056 if (err < 0) {
2057 /* skb is already freed. */
2058 skb = NULL;
2059 goto rt_tx_error;
2060 }
2061
2062 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
2063 #if IS_ENABLED(CONFIG_IPV6)
2064 } else {
2065 struct dst_entry *ndst;
2066 struct in6_addr saddr;
2067 u32 rt6i_flags;
2068
2069 if (!vxlan->vn6_sock)
2070 goto drop;
2071 sk = vxlan->vn6_sock->sock->sk;
2072
2073 ndst = vxlan6_get_route(vxlan, skb,
2074 rdst ? rdst->remote_ifindex : 0,
2075 &dst->sin6.sin6_addr, &saddr);
2076 if (IS_ERR(ndst)) {
2077 netdev_dbg(dev, "no route to %pI6\n",
2078 &dst->sin6.sin6_addr);
2079 dev->stats.tx_carrier_errors++;
2080 goto tx_error;
2081 }
2082
2083 if (ndst->dev == dev) {
2084 netdev_dbg(dev, "circular route to %pI6\n",
2085 &dst->sin6.sin6_addr);
2086 dst_release(ndst);
2087 dev->stats.collisions++;
2088 goto tx_error;
2089 }
2090
2091 /* Bypass encapsulation if the destination is local */
2092 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2093 if (rt6i_flags & RTF_LOCAL &&
2094 !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2095 struct vxlan_dev *dst_vxlan;
2096
2097 dst_release(ndst);
2098 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2099 dst->sa.sa_family, dst_port,
2100 vxlan->flags);
2101 if (!dst_vxlan)
2102 goto tx_error;
2103 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2104 return;
2105 }
2106
2107 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2108 err = vxlan6_xmit_skb(ndst, sk, skb, dev, &saddr, &dst->sin6.sin6_addr,
2109 0, ttl, src_port, dst_port, htonl(vni << 8), md,
2110 !net_eq(vxlan->net, dev_net(vxlan->dev)),
2111 flags);
2112 #endif
2113 }
2114
2115 return;
2116
2117 drop:
2118 dev->stats.tx_dropped++;
2119 goto tx_free;
2120
2121 rt_tx_error:
2122 ip_rt_put(rt);
2123 tx_error:
2124 dev->stats.tx_errors++;
2125 tx_free:
2126 dev_kfree_skb(skb);
2127 }
2128
2129 /* Transmit local packets over Vxlan
2130 *
2131 * Outer IP header inherits ECN and DF from inner header.
2132 * Outer UDP destination is the VXLAN assigned port.
2133 * source port is based on hash of flow
2134 */
2135 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2136 {
2137 struct vxlan_dev *vxlan = netdev_priv(dev);
2138 const struct ip_tunnel_info *info;
2139 struct ethhdr *eth;
2140 bool did_rsc = false;
2141 struct vxlan_rdst *rdst, *fdst = NULL;
2142 struct vxlan_fdb *f;
2143
2144 info = skb_tunnel_info(skb);
2145
2146 skb_reset_mac_header(skb);
2147 eth = eth_hdr(skb);
2148
2149 if ((vxlan->flags & VXLAN_F_PROXY)) {
2150 if (ntohs(eth->h_proto) == ETH_P_ARP)
2151 return arp_reduce(dev, skb);
2152 #if IS_ENABLED(CONFIG_IPV6)
2153 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2154 pskb_may_pull(skb, sizeof(struct ipv6hdr)
2155 + sizeof(struct nd_msg)) &&
2156 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2157 struct nd_msg *msg;
2158
2159 msg = (struct nd_msg *)skb_transport_header(skb);
2160 if (msg->icmph.icmp6_code == 0 &&
2161 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2162 return neigh_reduce(dev, skb);
2163 }
2164 eth = eth_hdr(skb);
2165 #endif
2166 }
2167
2168 if (vxlan->flags & VXLAN_F_COLLECT_METADATA &&
2169 info && info->mode & IP_TUNNEL_INFO_TX) {
2170 vxlan_xmit_one(skb, dev, NULL, false);
2171 return NETDEV_TX_OK;
2172 }
2173
2174 f = vxlan_find_mac(vxlan, eth->h_dest);
2175 did_rsc = false;
2176
2177 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2178 (ntohs(eth->h_proto) == ETH_P_IP ||
2179 ntohs(eth->h_proto) == ETH_P_IPV6)) {
2180 did_rsc = route_shortcircuit(dev, skb);
2181 if (did_rsc)
2182 f = vxlan_find_mac(vxlan, eth->h_dest);
2183 }
2184
2185 if (f == NULL) {
2186 f = vxlan_find_mac(vxlan, all_zeros_mac);
2187 if (f == NULL) {
2188 if ((vxlan->flags & VXLAN_F_L2MISS) &&
2189 !is_multicast_ether_addr(eth->h_dest))
2190 vxlan_fdb_miss(vxlan, eth->h_dest);
2191
2192 dev->stats.tx_dropped++;
2193 kfree_skb(skb);
2194 return NETDEV_TX_OK;
2195 }
2196 }
2197
2198 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2199 struct sk_buff *skb1;
2200
2201 if (!fdst) {
2202 fdst = rdst;
2203 continue;
2204 }
2205 skb1 = skb_clone(skb, GFP_ATOMIC);
2206 if (skb1)
2207 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2208 }
2209
2210 if (fdst)
2211 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2212 else
2213 kfree_skb(skb);
2214 return NETDEV_TX_OK;
2215 }
2216
2217 /* Walk the forwarding table and purge stale entries */
2218 static void vxlan_cleanup(unsigned long arg)
2219 {
2220 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2221 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2222 unsigned int h;
2223
2224 if (!netif_running(vxlan->dev))
2225 return;
2226
2227 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2228 struct hlist_node *p, *n;
2229
2230 spin_lock_bh(&vxlan->hash_lock);
2231 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2232 struct vxlan_fdb *f
2233 = container_of(p, struct vxlan_fdb, hlist);
2234 unsigned long timeout;
2235
2236 if (f->state & NUD_PERMANENT)
2237 continue;
2238
2239 timeout = f->used + vxlan->cfg.age_interval * HZ;
2240 if (time_before_eq(timeout, jiffies)) {
2241 netdev_dbg(vxlan->dev,
2242 "garbage collect %pM\n",
2243 f->eth_addr);
2244 f->state = NUD_STALE;
2245 vxlan_fdb_destroy(vxlan, f);
2246 } else if (time_before(timeout, next_timer))
2247 next_timer = timeout;
2248 }
2249 spin_unlock_bh(&vxlan->hash_lock);
2250 }
2251
2252 mod_timer(&vxlan->age_timer, next_timer);
2253 }
2254
2255 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2256 {
2257 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2258 __u32 vni = vxlan->default_dst.remote_vni;
2259
2260 spin_lock(&vn->sock_lock);
2261 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2262 spin_unlock(&vn->sock_lock);
2263 }
2264
2265 /* Setup stats when device is created */
2266 static int vxlan_init(struct net_device *dev)
2267 {
2268 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2269 if (!dev->tstats)
2270 return -ENOMEM;
2271
2272 return 0;
2273 }
2274
2275 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2276 {
2277 struct vxlan_fdb *f;
2278
2279 spin_lock_bh(&vxlan->hash_lock);
2280 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2281 if (f)
2282 vxlan_fdb_destroy(vxlan, f);
2283 spin_unlock_bh(&vxlan->hash_lock);
2284 }
2285
2286 static void vxlan_uninit(struct net_device *dev)
2287 {
2288 struct vxlan_dev *vxlan = netdev_priv(dev);
2289
2290 vxlan_fdb_delete_default(vxlan);
2291
2292 free_percpu(dev->tstats);
2293 }
2294
2295 /* Start ageing timer and join group when device is brought up */
2296 static int vxlan_open(struct net_device *dev)
2297 {
2298 struct vxlan_dev *vxlan = netdev_priv(dev);
2299 int ret;
2300
2301 ret = vxlan_sock_add(vxlan);
2302 if (ret < 0)
2303 return ret;
2304
2305 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2306 ret = vxlan_igmp_join(vxlan);
2307 if (ret == -EADDRINUSE)
2308 ret = 0;
2309 if (ret) {
2310 vxlan_sock_release(vxlan);
2311 return ret;
2312 }
2313 }
2314
2315 if (vxlan->cfg.age_interval)
2316 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2317
2318 return ret;
2319 }
2320
2321 /* Purge the forwarding table */
2322 static void vxlan_flush(struct vxlan_dev *vxlan)
2323 {
2324 unsigned int h;
2325
2326 spin_lock_bh(&vxlan->hash_lock);
2327 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2328 struct hlist_node *p, *n;
2329 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2330 struct vxlan_fdb *f
2331 = container_of(p, struct vxlan_fdb, hlist);
2332 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2333 if (!is_zero_ether_addr(f->eth_addr))
2334 vxlan_fdb_destroy(vxlan, f);
2335 }
2336 }
2337 spin_unlock_bh(&vxlan->hash_lock);
2338 }
2339
2340 /* Cleanup timer and forwarding table on shutdown */
2341 static int vxlan_stop(struct net_device *dev)
2342 {
2343 struct vxlan_dev *vxlan = netdev_priv(dev);
2344 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2345 int ret = 0;
2346
2347 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2348 !vxlan_group_used(vn, vxlan))
2349 ret = vxlan_igmp_leave(vxlan);
2350
2351 del_timer_sync(&vxlan->age_timer);
2352
2353 vxlan_flush(vxlan);
2354 vxlan_sock_release(vxlan);
2355
2356 return ret;
2357 }
2358
2359 /* Stub, nothing needs to be done. */
2360 static void vxlan_set_multicast_list(struct net_device *dev)
2361 {
2362 }
2363
2364 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2365 {
2366 struct vxlan_dev *vxlan = netdev_priv(dev);
2367 struct vxlan_rdst *dst = &vxlan->default_dst;
2368 struct net_device *lowerdev;
2369 int max_mtu;
2370
2371 lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2372 if (lowerdev == NULL)
2373 return eth_change_mtu(dev, new_mtu);
2374
2375 if (dst->remote_ip.sa.sa_family == AF_INET6)
2376 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2377 else
2378 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2379
2380 if (new_mtu < 68 || new_mtu > max_mtu)
2381 return -EINVAL;
2382
2383 dev->mtu = new_mtu;
2384 return 0;
2385 }
2386
2387 static int egress_ipv4_tun_info(struct net_device *dev, struct sk_buff *skb,
2388 struct ip_tunnel_info *info,
2389 __be16 sport, __be16 dport)
2390 {
2391 struct vxlan_dev *vxlan = netdev_priv(dev);
2392 struct rtable *rt;
2393 struct flowi4 fl4;
2394
2395 memset(&fl4, 0, sizeof(fl4));
2396 fl4.flowi4_tos = RT_TOS(info->key.tos);
2397 fl4.flowi4_mark = skb->mark;
2398 fl4.flowi4_proto = IPPROTO_UDP;
2399 fl4.daddr = info->key.u.ipv4.dst;
2400
2401 rt = ip_route_output_key(vxlan->net, &fl4);
2402 if (IS_ERR(rt))
2403 return PTR_ERR(rt);
2404 ip_rt_put(rt);
2405
2406 info->key.u.ipv4.src = fl4.saddr;
2407 info->key.tp_src = sport;
2408 info->key.tp_dst = dport;
2409 return 0;
2410 }
2411
2412 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2413 {
2414 struct vxlan_dev *vxlan = netdev_priv(dev);
2415 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2416 __be16 sport, dport;
2417
2418 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2419 vxlan->cfg.port_max, true);
2420 dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2421
2422 if (ip_tunnel_info_af(info) == AF_INET) {
2423 if (!vxlan->vn4_sock)
2424 return -EINVAL;
2425 return egress_ipv4_tun_info(dev, skb, info, sport, dport);
2426 } else {
2427 #if IS_ENABLED(CONFIG_IPV6)
2428 struct dst_entry *ndst;
2429
2430 if (!vxlan->vn6_sock)
2431 return -EINVAL;
2432 ndst = vxlan6_get_route(vxlan, skb, 0,
2433 &info->key.u.ipv6.dst,
2434 &info->key.u.ipv6.src);
2435 if (IS_ERR(ndst))
2436 return PTR_ERR(ndst);
2437 dst_release(ndst);
2438
2439 info->key.tp_src = sport;
2440 info->key.tp_dst = dport;
2441 #else /* !CONFIG_IPV6 */
2442 return -EPFNOSUPPORT;
2443 #endif
2444 }
2445 return 0;
2446 }
2447
2448 static const struct net_device_ops vxlan_netdev_ops = {
2449 .ndo_init = vxlan_init,
2450 .ndo_uninit = vxlan_uninit,
2451 .ndo_open = vxlan_open,
2452 .ndo_stop = vxlan_stop,
2453 .ndo_start_xmit = vxlan_xmit,
2454 .ndo_get_stats64 = ip_tunnel_get_stats64,
2455 .ndo_set_rx_mode = vxlan_set_multicast_list,
2456 .ndo_change_mtu = vxlan_change_mtu,
2457 .ndo_validate_addr = eth_validate_addr,
2458 .ndo_set_mac_address = eth_mac_addr,
2459 .ndo_fdb_add = vxlan_fdb_add,
2460 .ndo_fdb_del = vxlan_fdb_delete,
2461 .ndo_fdb_dump = vxlan_fdb_dump,
2462 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2463 };
2464
2465 /* Info for udev, that this is a virtual tunnel endpoint */
2466 static struct device_type vxlan_type = {
2467 .name = "vxlan",
2468 };
2469
2470 /* Calls the ndo_add_vxlan_port of the caller in order to
2471 * supply the listening VXLAN udp ports. Callers are expected
2472 * to implement the ndo_add_vxlan_port.
2473 */
2474 void vxlan_get_rx_port(struct net_device *dev)
2475 {
2476 struct vxlan_sock *vs;
2477 struct net *net = dev_net(dev);
2478 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2479 sa_family_t sa_family;
2480 __be16 port;
2481 unsigned int i;
2482
2483 spin_lock(&vn->sock_lock);
2484 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2485 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2486 port = inet_sk(vs->sock->sk)->inet_sport;
2487 sa_family = vxlan_get_sk_family(vs);
2488 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2489 port);
2490 }
2491 }
2492 spin_unlock(&vn->sock_lock);
2493 }
2494 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2495
2496 /* Initialize the device structure. */
2497 static void vxlan_setup(struct net_device *dev)
2498 {
2499 struct vxlan_dev *vxlan = netdev_priv(dev);
2500 unsigned int h;
2501
2502 eth_hw_addr_random(dev);
2503 ether_setup(dev);
2504
2505 dev->netdev_ops = &vxlan_netdev_ops;
2506 dev->destructor = free_netdev;
2507 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2508
2509 dev->features |= NETIF_F_LLTX;
2510 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2511 dev->features |= NETIF_F_RXCSUM;
2512 dev->features |= NETIF_F_GSO_SOFTWARE;
2513
2514 dev->vlan_features = dev->features;
2515 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2516 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2517 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2518 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2519 netif_keep_dst(dev);
2520 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
2521
2522 INIT_LIST_HEAD(&vxlan->next);
2523 spin_lock_init(&vxlan->hash_lock);
2524
2525 init_timer_deferrable(&vxlan->age_timer);
2526 vxlan->age_timer.function = vxlan_cleanup;
2527 vxlan->age_timer.data = (unsigned long) vxlan;
2528
2529 vxlan->cfg.dst_port = htons(vxlan_port);
2530
2531 vxlan->dev = dev;
2532
2533 gro_cells_init(&vxlan->gro_cells, dev);
2534
2535 for (h = 0; h < FDB_HASH_SIZE; ++h)
2536 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2537 }
2538
2539 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2540 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2541 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2542 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2543 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2544 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2545 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2546 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2547 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2548 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2549 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2550 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2551 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2552 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2553 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2554 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2555 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2556 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 },
2557 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2558 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 },
2559 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
2560 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
2561 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2562 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2563 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, },
2564 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG },
2565 };
2566
2567 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2568 {
2569 if (tb[IFLA_ADDRESS]) {
2570 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2571 pr_debug("invalid link address (not ethernet)\n");
2572 return -EINVAL;
2573 }
2574
2575 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2576 pr_debug("invalid all zero ethernet address\n");
2577 return -EADDRNOTAVAIL;
2578 }
2579 }
2580
2581 if (!data)
2582 return -EINVAL;
2583
2584 if (data[IFLA_VXLAN_ID]) {
2585 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2586 if (id >= VXLAN_VID_MASK)
2587 return -ERANGE;
2588 }
2589
2590 if (data[IFLA_VXLAN_PORT_RANGE]) {
2591 const struct ifla_vxlan_port_range *p
2592 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2593
2594 if (ntohs(p->high) < ntohs(p->low)) {
2595 pr_debug("port range %u .. %u not valid\n",
2596 ntohs(p->low), ntohs(p->high));
2597 return -EINVAL;
2598 }
2599 }
2600
2601 return 0;
2602 }
2603
2604 static void vxlan_get_drvinfo(struct net_device *netdev,
2605 struct ethtool_drvinfo *drvinfo)
2606 {
2607 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2608 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2609 }
2610
2611 static const struct ethtool_ops vxlan_ethtool_ops = {
2612 .get_drvinfo = vxlan_get_drvinfo,
2613 .get_link = ethtool_op_get_link,
2614 };
2615
2616 static void vxlan_del_work(struct work_struct *work)
2617 {
2618 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2619 udp_tunnel_sock_release(vs->sock);
2620 kfree_rcu(vs, rcu);
2621 }
2622
2623 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2624 __be16 port, u32 flags)
2625 {
2626 struct socket *sock;
2627 struct udp_port_cfg udp_conf;
2628 int err;
2629
2630 memset(&udp_conf, 0, sizeof(udp_conf));
2631
2632 if (ipv6) {
2633 udp_conf.family = AF_INET6;
2634 udp_conf.use_udp6_rx_checksums =
2635 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2636 udp_conf.ipv6_v6only = 1;
2637 } else {
2638 udp_conf.family = AF_INET;
2639 }
2640
2641 udp_conf.local_udp_port = port;
2642
2643 /* Open UDP socket */
2644 err = udp_sock_create(net, &udp_conf, &sock);
2645 if (err < 0)
2646 return ERR_PTR(err);
2647
2648 return sock;
2649 }
2650
2651 /* Create new listen socket if needed */
2652 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2653 __be16 port, u32 flags)
2654 {
2655 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2656 struct vxlan_sock *vs;
2657 struct socket *sock;
2658 unsigned int h;
2659 struct udp_tunnel_sock_cfg tunnel_cfg;
2660
2661 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2662 if (!vs)
2663 return ERR_PTR(-ENOMEM);
2664
2665 for (h = 0; h < VNI_HASH_SIZE; ++h)
2666 INIT_HLIST_HEAD(&vs->vni_list[h]);
2667
2668 INIT_WORK(&vs->del_work, vxlan_del_work);
2669
2670 sock = vxlan_create_sock(net, ipv6, port, flags);
2671 if (IS_ERR(sock)) {
2672 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2673 PTR_ERR(sock));
2674 kfree(vs);
2675 return ERR_CAST(sock);
2676 }
2677
2678 vs->sock = sock;
2679 atomic_set(&vs->refcnt, 1);
2680 vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2681
2682 /* Initialize the vxlan udp offloads structure */
2683 vs->udp_offloads.port = port;
2684 vs->udp_offloads.callbacks.gro_receive = vxlan_gro_receive;
2685 vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2686
2687 spin_lock(&vn->sock_lock);
2688 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2689 vxlan_notify_add_rx_port(vs);
2690 spin_unlock(&vn->sock_lock);
2691
2692 /* Mark socket as an encapsulation socket. */
2693 tunnel_cfg.sk_user_data = vs;
2694 tunnel_cfg.encap_type = 1;
2695 tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2696 tunnel_cfg.encap_destroy = NULL;
2697
2698 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2699
2700 return vs;
2701 }
2702
2703 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2704 {
2705 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2706 struct vxlan_sock *vs = NULL;
2707
2708 if (!vxlan->cfg.no_share) {
2709 spin_lock(&vn->sock_lock);
2710 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2711 vxlan->cfg.dst_port, vxlan->flags);
2712 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2713 spin_unlock(&vn->sock_lock);
2714 return -EBUSY;
2715 }
2716 spin_unlock(&vn->sock_lock);
2717 }
2718 if (!vs)
2719 vs = vxlan_socket_create(vxlan->net, ipv6,
2720 vxlan->cfg.dst_port, vxlan->flags);
2721 if (IS_ERR(vs))
2722 return PTR_ERR(vs);
2723 #if IS_ENABLED(CONFIG_IPV6)
2724 if (ipv6)
2725 vxlan->vn6_sock = vs;
2726 else
2727 #endif
2728 vxlan->vn4_sock = vs;
2729 vxlan_vs_add_dev(vs, vxlan);
2730 return 0;
2731 }
2732
2733 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2734 {
2735 bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2736 bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2737 int ret = 0;
2738
2739 vxlan->vn4_sock = NULL;
2740 #if IS_ENABLED(CONFIG_IPV6)
2741 vxlan->vn6_sock = NULL;
2742 if (ipv6 || metadata)
2743 ret = __vxlan_sock_add(vxlan, true);
2744 #endif
2745 if (!ret && (!ipv6 || metadata))
2746 ret = __vxlan_sock_add(vxlan, false);
2747 if (ret < 0)
2748 vxlan_sock_release(vxlan);
2749 return ret;
2750 }
2751
2752 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2753 struct vxlan_config *conf)
2754 {
2755 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2756 struct vxlan_dev *vxlan = netdev_priv(dev);
2757 struct vxlan_rdst *dst = &vxlan->default_dst;
2758 unsigned short needed_headroom = ETH_HLEN;
2759 int err;
2760 bool use_ipv6 = false;
2761 __be16 default_port = vxlan->cfg.dst_port;
2762
2763 vxlan->net = src_net;
2764
2765 dst->remote_vni = conf->vni;
2766
2767 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2768
2769 /* Unless IPv6 is explicitly requested, assume IPv4 */
2770 if (!dst->remote_ip.sa.sa_family)
2771 dst->remote_ip.sa.sa_family = AF_INET;
2772
2773 if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2774 vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2775 if (!IS_ENABLED(CONFIG_IPV6))
2776 return -EPFNOSUPPORT;
2777 use_ipv6 = true;
2778 vxlan->flags |= VXLAN_F_IPV6;
2779 }
2780
2781 if (conf->remote_ifindex) {
2782 struct net_device *lowerdev
2783 = __dev_get_by_index(src_net, conf->remote_ifindex);
2784
2785 dst->remote_ifindex = conf->remote_ifindex;
2786
2787 if (!lowerdev) {
2788 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2789 return -ENODEV;
2790 }
2791
2792 #if IS_ENABLED(CONFIG_IPV6)
2793 if (use_ipv6) {
2794 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2795 if (idev && idev->cnf.disable_ipv6) {
2796 pr_info("IPv6 is disabled via sysctl\n");
2797 return -EPERM;
2798 }
2799 }
2800 #endif
2801
2802 if (!conf->mtu)
2803 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2804
2805 needed_headroom = lowerdev->hard_header_len;
2806 }
2807
2808 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2809 needed_headroom += VXLAN6_HEADROOM;
2810 else
2811 needed_headroom += VXLAN_HEADROOM;
2812 dev->needed_headroom = needed_headroom;
2813
2814 memcpy(&vxlan->cfg, conf, sizeof(*conf));
2815 if (!vxlan->cfg.dst_port)
2816 vxlan->cfg.dst_port = default_port;
2817 vxlan->flags |= conf->flags;
2818
2819 if (!vxlan->cfg.age_interval)
2820 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2821
2822 if (vxlan_find_vni(src_net, conf->vni, use_ipv6 ? AF_INET6 : AF_INET,
2823 vxlan->cfg.dst_port, vxlan->flags))
2824 return -EEXIST;
2825
2826 dev->ethtool_ops = &vxlan_ethtool_ops;
2827
2828 /* create an fdb entry for a valid default destination */
2829 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2830 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2831 &vxlan->default_dst.remote_ip,
2832 NUD_REACHABLE|NUD_PERMANENT,
2833 NLM_F_EXCL|NLM_F_CREATE,
2834 vxlan->cfg.dst_port,
2835 vxlan->default_dst.remote_vni,
2836 vxlan->default_dst.remote_ifindex,
2837 NTF_SELF);
2838 if (err)
2839 return err;
2840 }
2841
2842 err = register_netdevice(dev);
2843 if (err) {
2844 vxlan_fdb_delete_default(vxlan);
2845 return err;
2846 }
2847
2848 list_add(&vxlan->next, &vn->vxlan_list);
2849
2850 return 0;
2851 }
2852
2853 struct net_device *vxlan_dev_create(struct net *net, const char *name,
2854 u8 name_assign_type, struct vxlan_config *conf)
2855 {
2856 struct nlattr *tb[IFLA_MAX+1];
2857 struct net_device *dev;
2858 int err;
2859
2860 memset(&tb, 0, sizeof(tb));
2861
2862 dev = rtnl_create_link(net, name, name_assign_type,
2863 &vxlan_link_ops, tb);
2864 if (IS_ERR(dev))
2865 return dev;
2866
2867 err = vxlan_dev_configure(net, dev, conf);
2868 if (err < 0) {
2869 free_netdev(dev);
2870 return ERR_PTR(err);
2871 }
2872
2873 return dev;
2874 }
2875 EXPORT_SYMBOL_GPL(vxlan_dev_create);
2876
2877 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2878 struct nlattr *tb[], struct nlattr *data[])
2879 {
2880 struct vxlan_config conf;
2881 int err;
2882
2883 memset(&conf, 0, sizeof(conf));
2884
2885 if (data[IFLA_VXLAN_ID])
2886 conf.vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2887
2888 if (data[IFLA_VXLAN_GROUP]) {
2889 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2890 } else if (data[IFLA_VXLAN_GROUP6]) {
2891 if (!IS_ENABLED(CONFIG_IPV6))
2892 return -EPFNOSUPPORT;
2893
2894 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2895 conf.remote_ip.sa.sa_family = AF_INET6;
2896 }
2897
2898 if (data[IFLA_VXLAN_LOCAL]) {
2899 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2900 conf.saddr.sa.sa_family = AF_INET;
2901 } else if (data[IFLA_VXLAN_LOCAL6]) {
2902 if (!IS_ENABLED(CONFIG_IPV6))
2903 return -EPFNOSUPPORT;
2904
2905 /* TODO: respect scope id */
2906 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2907 conf.saddr.sa.sa_family = AF_INET6;
2908 }
2909
2910 if (data[IFLA_VXLAN_LINK])
2911 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2912
2913 if (data[IFLA_VXLAN_TOS])
2914 conf.tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2915
2916 if (data[IFLA_VXLAN_TTL])
2917 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2918
2919 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2920 conf.flags |= VXLAN_F_LEARN;
2921
2922 if (data[IFLA_VXLAN_AGEING])
2923 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2924
2925 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2926 conf.flags |= VXLAN_F_PROXY;
2927
2928 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2929 conf.flags |= VXLAN_F_RSC;
2930
2931 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2932 conf.flags |= VXLAN_F_L2MISS;
2933
2934 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2935 conf.flags |= VXLAN_F_L3MISS;
2936
2937 if (data[IFLA_VXLAN_LIMIT])
2938 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2939
2940 if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2941 nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2942 conf.flags |= VXLAN_F_COLLECT_METADATA;
2943
2944 if (data[IFLA_VXLAN_PORT_RANGE]) {
2945 const struct ifla_vxlan_port_range *p
2946 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2947 conf.port_min = ntohs(p->low);
2948 conf.port_max = ntohs(p->high);
2949 }
2950
2951 if (data[IFLA_VXLAN_PORT])
2952 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2953
2954 if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2955 conf.flags |= VXLAN_F_UDP_CSUM;
2956
2957 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2958 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2959 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2960
2961 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2962 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2963 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2964
2965 if (data[IFLA_VXLAN_REMCSUM_TX] &&
2966 nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
2967 conf.flags |= VXLAN_F_REMCSUM_TX;
2968
2969 if (data[IFLA_VXLAN_REMCSUM_RX] &&
2970 nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
2971 conf.flags |= VXLAN_F_REMCSUM_RX;
2972
2973 if (data[IFLA_VXLAN_GBP])
2974 conf.flags |= VXLAN_F_GBP;
2975
2976 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
2977 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
2978
2979 err = vxlan_dev_configure(src_net, dev, &conf);
2980 switch (err) {
2981 case -ENODEV:
2982 pr_info("ifindex %d does not exist\n", conf.remote_ifindex);
2983 break;
2984
2985 case -EPERM:
2986 pr_info("IPv6 is disabled via sysctl\n");
2987 break;
2988
2989 case -EEXIST:
2990 pr_info("duplicate VNI %u\n", conf.vni);
2991 break;
2992 }
2993
2994 return err;
2995 }
2996
2997 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2998 {
2999 struct vxlan_dev *vxlan = netdev_priv(dev);
3000 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3001
3002 spin_lock(&vn->sock_lock);
3003 if (!hlist_unhashed(&vxlan->hlist))
3004 hlist_del_rcu(&vxlan->hlist);
3005 spin_unlock(&vn->sock_lock);
3006
3007 gro_cells_destroy(&vxlan->gro_cells);
3008 list_del(&vxlan->next);
3009 unregister_netdevice_queue(dev, head);
3010 }
3011
3012 static size_t vxlan_get_size(const struct net_device *dev)
3013 {
3014
3015 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
3016 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3017 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3018 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3019 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
3020 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
3021 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
3022 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
3023 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
3024 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
3025 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
3026 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */
3027 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3028 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3029 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3030 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3031 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3032 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3033 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3034 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3035 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3036 0;
3037 }
3038
3039 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3040 {
3041 const struct vxlan_dev *vxlan = netdev_priv(dev);
3042 const struct vxlan_rdst *dst = &vxlan->default_dst;
3043 struct ifla_vxlan_port_range ports = {
3044 .low = htons(vxlan->cfg.port_min),
3045 .high = htons(vxlan->cfg.port_max),
3046 };
3047
3048 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
3049 goto nla_put_failure;
3050
3051 if (!vxlan_addr_any(&dst->remote_ip)) {
3052 if (dst->remote_ip.sa.sa_family == AF_INET) {
3053 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3054 dst->remote_ip.sin.sin_addr.s_addr))
3055 goto nla_put_failure;
3056 #if IS_ENABLED(CONFIG_IPV6)
3057 } else {
3058 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3059 &dst->remote_ip.sin6.sin6_addr))
3060 goto nla_put_failure;
3061 #endif
3062 }
3063 }
3064
3065 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3066 goto nla_put_failure;
3067
3068 if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3069 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3070 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3071 vxlan->cfg.saddr.sin.sin_addr.s_addr))
3072 goto nla_put_failure;
3073 #if IS_ENABLED(CONFIG_IPV6)
3074 } else {
3075 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3076 &vxlan->cfg.saddr.sin6.sin6_addr))
3077 goto nla_put_failure;
3078 #endif
3079 }
3080 }
3081
3082 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3083 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3084 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3085 !!(vxlan->flags & VXLAN_F_LEARN)) ||
3086 nla_put_u8(skb, IFLA_VXLAN_PROXY,
3087 !!(vxlan->flags & VXLAN_F_PROXY)) ||
3088 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3089 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3090 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3091 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3092 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3093 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3094 !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3095 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3096 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3097 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3098 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3099 !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
3100 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3101 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3102 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3103 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3104 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3105 !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3106 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3107 !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3108 goto nla_put_failure;
3109
3110 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3111 goto nla_put_failure;
3112
3113 if (vxlan->flags & VXLAN_F_GBP &&
3114 nla_put_flag(skb, IFLA_VXLAN_GBP))
3115 goto nla_put_failure;
3116
3117 if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3118 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3119 goto nla_put_failure;
3120
3121 return 0;
3122
3123 nla_put_failure:
3124 return -EMSGSIZE;
3125 }
3126
3127 static struct net *vxlan_get_link_net(const struct net_device *dev)
3128 {
3129 struct vxlan_dev *vxlan = netdev_priv(dev);
3130
3131 return vxlan->net;
3132 }
3133
3134 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3135 .kind = "vxlan",
3136 .maxtype = IFLA_VXLAN_MAX,
3137 .policy = vxlan_policy,
3138 .priv_size = sizeof(struct vxlan_dev),
3139 .setup = vxlan_setup,
3140 .validate = vxlan_validate,
3141 .newlink = vxlan_newlink,
3142 .dellink = vxlan_dellink,
3143 .get_size = vxlan_get_size,
3144 .fill_info = vxlan_fill_info,
3145 .get_link_net = vxlan_get_link_net,
3146 };
3147
3148 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3149 struct net_device *dev)
3150 {
3151 struct vxlan_dev *vxlan, *next;
3152 LIST_HEAD(list_kill);
3153
3154 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3155 struct vxlan_rdst *dst = &vxlan->default_dst;
3156
3157 /* In case we created vxlan device with carrier
3158 * and we loose the carrier due to module unload
3159 * we also need to remove vxlan device. In other
3160 * cases, it's not necessary and remote_ifindex
3161 * is 0 here, so no matches.
3162 */
3163 if (dst->remote_ifindex == dev->ifindex)
3164 vxlan_dellink(vxlan->dev, &list_kill);
3165 }
3166
3167 unregister_netdevice_many(&list_kill);
3168 }
3169
3170 static int vxlan_lowerdev_event(struct notifier_block *unused,
3171 unsigned long event, void *ptr)
3172 {
3173 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3174 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3175
3176 if (event == NETDEV_UNREGISTER)
3177 vxlan_handle_lowerdev_unregister(vn, dev);
3178
3179 return NOTIFY_DONE;
3180 }
3181
3182 static struct notifier_block vxlan_notifier_block __read_mostly = {
3183 .notifier_call = vxlan_lowerdev_event,
3184 };
3185
3186 static __net_init int vxlan_init_net(struct net *net)
3187 {
3188 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3189 unsigned int h;
3190
3191 INIT_LIST_HEAD(&vn->vxlan_list);
3192 spin_lock_init(&vn->sock_lock);
3193
3194 for (h = 0; h < PORT_HASH_SIZE; ++h)
3195 INIT_HLIST_HEAD(&vn->sock_list[h]);
3196
3197 return 0;
3198 }
3199
3200 static void __net_exit vxlan_exit_net(struct net *net)
3201 {
3202 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3203 struct vxlan_dev *vxlan, *next;
3204 struct net_device *dev, *aux;
3205 LIST_HEAD(list);
3206
3207 rtnl_lock();
3208 for_each_netdev_safe(net, dev, aux)
3209 if (dev->rtnl_link_ops == &vxlan_link_ops)
3210 unregister_netdevice_queue(dev, &list);
3211
3212 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3213 /* If vxlan->dev is in the same netns, it has already been added
3214 * to the list by the previous loop.
3215 */
3216 if (!net_eq(dev_net(vxlan->dev), net)) {
3217 gro_cells_destroy(&vxlan->gro_cells);
3218 unregister_netdevice_queue(vxlan->dev, &list);
3219 }
3220 }
3221
3222 unregister_netdevice_many(&list);
3223 rtnl_unlock();
3224 }
3225
3226 static struct pernet_operations vxlan_net_ops = {
3227 .init = vxlan_init_net,
3228 .exit = vxlan_exit_net,
3229 .id = &vxlan_net_id,
3230 .size = sizeof(struct vxlan_net),
3231 };
3232
3233 static int __init vxlan_init_module(void)
3234 {
3235 int rc;
3236
3237 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
3238 if (!vxlan_wq)
3239 return -ENOMEM;
3240
3241 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3242
3243 rc = register_pernet_subsys(&vxlan_net_ops);
3244 if (rc)
3245 goto out1;
3246
3247 rc = register_netdevice_notifier(&vxlan_notifier_block);
3248 if (rc)
3249 goto out2;
3250
3251 rc = rtnl_link_register(&vxlan_link_ops);
3252 if (rc)
3253 goto out3;
3254
3255 return 0;
3256 out3:
3257 unregister_netdevice_notifier(&vxlan_notifier_block);
3258 out2:
3259 unregister_pernet_subsys(&vxlan_net_ops);
3260 out1:
3261 destroy_workqueue(vxlan_wq);
3262 return rc;
3263 }
3264 late_initcall(vxlan_init_module);
3265
3266 static void __exit vxlan_cleanup_module(void)
3267 {
3268 rtnl_link_unregister(&vxlan_link_ops);
3269 unregister_netdevice_notifier(&vxlan_notifier_block);
3270 destroy_workqueue(vxlan_wq);
3271 unregister_pernet_subsys(&vxlan_net_ops);
3272 /* rcu_barrier() is called by netns */
3273 }
3274 module_exit(vxlan_cleanup_module);
3275
3276 MODULE_LICENSE("GPL");
3277 MODULE_VERSION(VXLAN_VERSION);
3278 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3279 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3280 MODULE_ALIAS_RTNL_LINK("vxlan");
Linux kernel stable tree