projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: get rid of SET_ETHTOOL_OPS
[deliverable/linux.git] / drivers / net / vxlan.c
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/rtnetlink.h>
37 #include <net/route.h>
38 #include <net/dsfield.h>
39 #include <net/inet_ecn.h>
40 #include <net/net_namespace.h>
41 #include <net/netns/generic.h>
42 #include <net/vxlan.h>
43 #include <net/protocol.h>
44 #if IS_ENABLED(CONFIG_IPV6)
45 #include <net/ipv6.h>
46 #include <net/addrconf.h>
47 #include <net/ip6_tunnel.h>
48 #include <net/ip6_checksum.h>
49 #endif
50
51 #define VXLAN_VERSION "0.1"
52
53 #define PORT_HASH_BITS 8
54 #define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
55 #define VNI_HASH_BITS 10
56 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
57 #define FDB_HASH_BITS 8
58 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
59 #define FDB_AGE_DEFAULT 300 /* 5 min */
60 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
61
62 #define VXLAN_N_VID (1u << 24)
63 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
64 #define VXLAN_HLEN (sizeof(struct udphdr) + sizeof(struct vxlanhdr))
65
66 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
67
68 /* VXLAN protocol header */
69 struct vxlanhdr {
70 __be32 vx_flags;
71 __be32 vx_vni;
72 };
73
74 /* UDP port for VXLAN traffic.
75 * The IANA assigned port is 4789, but the Linux default is 8472
76 * for compatibility with early adopters.
77 */
78 static unsigned short vxlan_port __read_mostly = 8472;
79 module_param_named(udp_port, vxlan_port, ushort, 0444);
80 MODULE_PARM_DESC(udp_port, "Destination UDP port");
81
82 static bool log_ecn_error = true;
83 module_param(log_ecn_error, bool, 0644);
84 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
85
86 static int vxlan_net_id;
87
88 static const u8 all_zeros_mac[ETH_ALEN];
89
90 /* per-network namespace private data for this module */
91 struct vxlan_net {
92 struct list_head vxlan_list;
93 struct hlist_head sock_list[PORT_HASH_SIZE];
94 spinlock_t sock_lock;
95 };
96
97 union vxlan_addr {
98 struct sockaddr_in sin;
99 struct sockaddr_in6 sin6;
100 struct sockaddr sa;
101 };
102
103 struct vxlan_rdst {
104 union vxlan_addr remote_ip;
105 __be16 remote_port;
106 u32 remote_vni;
107 u32 remote_ifindex;
108 struct list_head list;
109 struct rcu_head rcu;
110 };
111
112 /* Forwarding table entry */
113 struct vxlan_fdb {
114 struct hlist_node hlist; /* linked list of entries */
115 struct rcu_head rcu;
116 unsigned long updated; /* jiffies */
117 unsigned long used;
118 struct list_head remotes;
119 u16 state; /* see ndm_state */
120 u8 flags; /* see ndm_flags */
121 u8 eth_addr[ETH_ALEN];
122 };
123
124 /* Pseudo network device */
125 struct vxlan_dev {
126 struct hlist_node hlist; /* vni hash table */
127 struct list_head next; /* vxlan's per namespace list */
128 struct vxlan_sock *vn_sock; /* listening socket */
129 struct net_device *dev;
130 struct net *net; /* netns for packet i/o */
131 struct vxlan_rdst default_dst; /* default destination */
132 union vxlan_addr saddr; /* source address */
133 __be16 dst_port;
134 __u16 port_min; /* source port range */
135 __u16 port_max;
136 __u8 tos; /* TOS override */
137 __u8 ttl;
138 u32 flags; /* VXLAN_F_* below */
139
140 struct work_struct sock_work;
141 struct work_struct igmp_join;
142 struct work_struct igmp_leave;
143
144 unsigned long age_interval;
145 struct timer_list age_timer;
146 spinlock_t hash_lock;
147 unsigned int addrcnt;
148 unsigned int addrmax;
149
150 struct hlist_head fdb_head[FDB_HASH_SIZE];
151 };
152
153 #define VXLAN_F_LEARN 0x01
154 #define VXLAN_F_PROXY 0x02
155 #define VXLAN_F_RSC 0x04
156 #define VXLAN_F_L2MISS 0x08
157 #define VXLAN_F_L3MISS 0x10
158 #define VXLAN_F_IPV6 0x20 /* internal flag */
159
160 /* salt for hash table */
161 static u32 vxlan_salt __read_mostly;
162 static struct workqueue_struct *vxlan_wq;
163
164 static void vxlan_sock_work(struct work_struct *work);
165
166 #if IS_ENABLED(CONFIG_IPV6)
167 static inline
168 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
169 {
170 if (a->sa.sa_family != b->sa.sa_family)
171 return false;
172 if (a->sa.sa_family == AF_INET6)
173 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
174 else
175 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
176 }
177
178 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
179 {
180 if (ipa->sa.sa_family == AF_INET6)
181 return ipv6_addr_any(&ipa->sin6.sin6_addr);
182 else
183 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
184 }
185
186 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
187 {
188 if (ipa->sa.sa_family == AF_INET6)
189 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
190 else
191 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
192 }
193
194 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
195 {
196 if (nla_len(nla) >= sizeof(struct in6_addr)) {
197 nla_memcpy(&ip->sin6.sin6_addr, nla, sizeof(struct in6_addr));
198 ip->sa.sa_family = AF_INET6;
199 return 0;
200 } else if (nla_len(nla) >= sizeof(__be32)) {
201 ip->sin.sin_addr.s_addr = nla_get_be32(nla);
202 ip->sa.sa_family = AF_INET;
203 return 0;
204 } else {
205 return -EAFNOSUPPORT;
206 }
207 }
208
209 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
210 const union vxlan_addr *ip)
211 {
212 if (ip->sa.sa_family == AF_INET6)
213 return nla_put(skb, attr, sizeof(struct in6_addr), &ip->sin6.sin6_addr);
214 else
215 return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
216 }
217
218 #else /* !CONFIG_IPV6 */
219
220 static inline
221 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
222 {
223 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
224 }
225
226 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
227 {
228 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
229 }
230
231 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
232 {
233 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
234 }
235
236 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
237 {
238 if (nla_len(nla) >= sizeof(struct in6_addr)) {
239 return -EAFNOSUPPORT;
240 } else if (nla_len(nla) >= sizeof(__be32)) {
241 ip->sin.sin_addr.s_addr = nla_get_be32(nla);
242 ip->sa.sa_family = AF_INET;
243 return 0;
244 } else {
245 return -EAFNOSUPPORT;
246 }
247 }
248
249 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
250 const union vxlan_addr *ip)
251 {
252 return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
253 }
254 #endif
255
256 /* Virtual Network hash table head */
257 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
258 {
259 return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
260 }
261
262 /* Socket hash table head */
263 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
264 {
265 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
266
267 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
268 }
269
270 /* First remote destination for a forwarding entry.
271 * Guaranteed to be non-NULL because remotes are never deleted.
272 */
273 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
274 {
275 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
276 }
277
278 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
279 {
280 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
281 }
282
283 /* Find VXLAN socket based on network namespace and UDP port */
284 static struct vxlan_sock *vxlan_find_sock(struct net *net, __be16 port)
285 {
286 struct vxlan_sock *vs;
287
288 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
289 if (inet_sk(vs->sock->sk)->inet_sport == port)
290 return vs;
291 }
292 return NULL;
293 }
294
295 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
296 {
297 struct vxlan_dev *vxlan;
298
299 hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
300 if (vxlan->default_dst.remote_vni == id)
301 return vxlan;
302 }
303
304 return NULL;
305 }
306
307 /* Look up VNI in a per net namespace table */
308 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id, __be16 port)
309 {
310 struct vxlan_sock *vs;
311
312 vs = vxlan_find_sock(net, port);
313 if (!vs)
314 return NULL;
315
316 return vxlan_vs_find_vni(vs, id);
317 }
318
319 /* Fill in neighbour message in skbuff. */
320 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
321 const struct vxlan_fdb *fdb,
322 u32 portid, u32 seq, int type, unsigned int flags,
323 const struct vxlan_rdst *rdst)
324 {
325 unsigned long now = jiffies;
326 struct nda_cacheinfo ci;
327 struct nlmsghdr *nlh;
328 struct ndmsg *ndm;
329 bool send_ip, send_eth;
330
331 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
332 if (nlh == NULL)
333 return -EMSGSIZE;
334
335 ndm = nlmsg_data(nlh);
336 memset(ndm, 0, sizeof(*ndm));
337
338 send_eth = send_ip = true;
339
340 if (type == RTM_GETNEIGH) {
341 ndm->ndm_family = AF_INET;
342 send_ip = !vxlan_addr_any(&rdst->remote_ip);
343 send_eth = !is_zero_ether_addr(fdb->eth_addr);
344 } else
345 ndm->ndm_family = AF_BRIDGE;
346 ndm->ndm_state = fdb->state;
347 ndm->ndm_ifindex = vxlan->dev->ifindex;
348 ndm->ndm_flags = fdb->flags;
349 ndm->ndm_type = NDA_DST;
350
351 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
352 goto nla_put_failure;
353
354 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
355 goto nla_put_failure;
356
357 if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
358 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
359 goto nla_put_failure;
360 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
361 nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
362 goto nla_put_failure;
363 if (rdst->remote_ifindex &&
364 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
365 goto nla_put_failure;
366
367 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
368 ci.ndm_confirmed = 0;
369 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
370 ci.ndm_refcnt = 0;
371
372 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
373 goto nla_put_failure;
374
375 return nlmsg_end(skb, nlh);
376
377 nla_put_failure:
378 nlmsg_cancel(skb, nlh);
379 return -EMSGSIZE;
380 }
381
382 static inline size_t vxlan_nlmsg_size(void)
383 {
384 return NLMSG_ALIGN(sizeof(struct ndmsg))
385 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
386 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
387 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
388 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
389 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
390 + nla_total_size(sizeof(struct nda_cacheinfo));
391 }
392
393 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
394 struct vxlan_rdst *rd, int type)
395 {
396 struct net *net = dev_net(vxlan->dev);
397 struct sk_buff *skb;
398 int err = -ENOBUFS;
399
400 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
401 if (skb == NULL)
402 goto errout;
403
404 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
405 if (err < 0) {
406 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
407 WARN_ON(err == -EMSGSIZE);
408 kfree_skb(skb);
409 goto errout;
410 }
411
412 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
413 return;
414 errout:
415 if (err < 0)
416 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
417 }
418
419 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
420 {
421 struct vxlan_dev *vxlan = netdev_priv(dev);
422 struct vxlan_fdb f = {
423 .state = NUD_STALE,
424 };
425 struct vxlan_rdst remote = {
426 .remote_ip = *ipa, /* goes to NDA_DST */
427 .remote_vni = VXLAN_N_VID,
428 };
429
430 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
431 }
432
433 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
434 {
435 struct vxlan_fdb f = {
436 .state = NUD_STALE,
437 };
438 struct vxlan_rdst remote = { };
439
440 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
441
442 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
443 }
444
445 /* Hash Ethernet address */
446 static u32 eth_hash(const unsigned char *addr)
447 {
448 u64 value = get_unaligned((u64 *)addr);
449
450 /* only want 6 bytes */
451 #ifdef __BIG_ENDIAN
452 value >>= 16;
453 #else
454 value <<= 16;
455 #endif
456 return hash_64(value, FDB_HASH_BITS);
457 }
458
459 /* Hash chain to use given mac address */
460 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
461 const u8 *mac)
462 {
463 return &vxlan->fdb_head[eth_hash(mac)];
464 }
465
466 /* Look up Ethernet address in forwarding table */
467 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
468 const u8 *mac)
469 {
470 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
471 struct vxlan_fdb *f;
472
473 hlist_for_each_entry_rcu(f, head, hlist) {
474 if (ether_addr_equal(mac, f->eth_addr))
475 return f;
476 }
477
478 return NULL;
479 }
480
481 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
482 const u8 *mac)
483 {
484 struct vxlan_fdb *f;
485
486 f = __vxlan_find_mac(vxlan, mac);
487 if (f)
488 f->used = jiffies;
489
490 return f;
491 }
492
493 /* caller should hold vxlan->hash_lock */
494 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
495 union vxlan_addr *ip, __be16 port,
496 __u32 vni, __u32 ifindex)
497 {
498 struct vxlan_rdst *rd;
499
500 list_for_each_entry(rd, &f->remotes, list) {
501 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
502 rd->remote_port == port &&
503 rd->remote_vni == vni &&
504 rd->remote_ifindex == ifindex)
505 return rd;
506 }
507
508 return NULL;
509 }
510
511 /* Replace destination of unicast mac */
512 static int vxlan_fdb_replace(struct vxlan_fdb *f,
513 union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
514 {
515 struct vxlan_rdst *rd;
516
517 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
518 if (rd)
519 return 0;
520
521 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
522 if (!rd)
523 return 0;
524 rd->remote_ip = *ip;
525 rd->remote_port = port;
526 rd->remote_vni = vni;
527 rd->remote_ifindex = ifindex;
528 return 1;
529 }
530
531 /* Add/update destinations for multicast */
532 static int vxlan_fdb_append(struct vxlan_fdb *f,
533 union vxlan_addr *ip, __be16 port, __u32 vni,
534 __u32 ifindex, struct vxlan_rdst **rdp)
535 {
536 struct vxlan_rdst *rd;
537
538 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
539 if (rd)
540 return 0;
541
542 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
543 if (rd == NULL)
544 return -ENOBUFS;
545 rd->remote_ip = *ip;
546 rd->remote_port = port;
547 rd->remote_vni = vni;
548 rd->remote_ifindex = ifindex;
549
550 list_add_tail_rcu(&rd->list, &f->remotes);
551
552 *rdp = rd;
553 return 1;
554 }
555
556 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head, struct sk_buff *skb)
557 {
558 struct sk_buff *p, **pp = NULL;
559 struct vxlanhdr *vh, *vh2;
560 struct ethhdr *eh, *eh2;
561 unsigned int hlen, off_vx, off_eth;
562 const struct packet_offload *ptype;
563 __be16 type;
564 int flush = 1;
565
566 off_vx = skb_gro_offset(skb);
567 hlen = off_vx + sizeof(*vh);
568 vh = skb_gro_header_fast(skb, off_vx);
569 if (skb_gro_header_hard(skb, hlen)) {
570 vh = skb_gro_header_slow(skb, hlen, off_vx);
571 if (unlikely(!vh))
572 goto out;
573 }
574 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
575
576 off_eth = skb_gro_offset(skb);
577 hlen = off_eth + sizeof(*eh);
578 eh = skb_gro_header_fast(skb, off_eth);
579 if (skb_gro_header_hard(skb, hlen)) {
580 eh = skb_gro_header_slow(skb, hlen, off_eth);
581 if (unlikely(!eh))
582 goto out;
583 }
584
585 flush = 0;
586
587 for (p = *head; p; p = p->next) {
588 if (!NAPI_GRO_CB(p)->same_flow)
589 continue;
590
591 vh2 = (struct vxlanhdr *)(p->data + off_vx);
592 eh2 = (struct ethhdr *)(p->data + off_eth);
593 if (vh->vx_vni != vh2->vx_vni || compare_ether_header(eh, eh2)) {
594 NAPI_GRO_CB(p)->same_flow = 0;
595 continue;
596 }
597 }
598
599 type = eh->h_proto;
600
601 rcu_read_lock();
602 ptype = gro_find_receive_by_type(type);
603 if (ptype == NULL) {
604 flush = 1;
605 goto out_unlock;
606 }
607
608 skb_gro_pull(skb, sizeof(*eh)); /* pull inner eth header */
609 pp = ptype->callbacks.gro_receive(head, skb);
610
611 out_unlock:
612 rcu_read_unlock();
613 out:
614 NAPI_GRO_CB(skb)->flush |= flush;
615
616 return pp;
617 }
618
619 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff)
620 {
621 struct ethhdr *eh;
622 struct packet_offload *ptype;
623 __be16 type;
624 int vxlan_len = sizeof(struct vxlanhdr) + sizeof(struct ethhdr);
625 int err = -ENOSYS;
626
627 eh = (struct ethhdr *)(skb->data + nhoff + sizeof(struct vxlanhdr));
628 type = eh->h_proto;
629
630 rcu_read_lock();
631 ptype = gro_find_complete_by_type(type);
632 if (ptype != NULL)
633 err = ptype->callbacks.gro_complete(skb, nhoff + vxlan_len);
634
635 rcu_read_unlock();
636 return err;
637 }
638
639 /* Notify netdevs that UDP port started listening */
640 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
641 {
642 struct net_device *dev;
643 struct sock *sk = vs->sock->sk;
644 struct net *net = sock_net(sk);
645 sa_family_t sa_family = sk->sk_family;
646 __be16 port = inet_sk(sk)->inet_sport;
647 int err;
648
649 if (sa_family == AF_INET) {
650 err = udp_add_offload(&vs->udp_offloads);
651 if (err)
652 pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
653 }
654
655 rcu_read_lock();
656 for_each_netdev_rcu(net, dev) {
657 if (dev->netdev_ops->ndo_add_vxlan_port)
658 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
659 port);
660 }
661 rcu_read_unlock();
662 }
663
664 /* Notify netdevs that UDP port is no more listening */
665 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
666 {
667 struct net_device *dev;
668 struct sock *sk = vs->sock->sk;
669 struct net *net = sock_net(sk);
670 sa_family_t sa_family = sk->sk_family;
671 __be16 port = inet_sk(sk)->inet_sport;
672
673 rcu_read_lock();
674 for_each_netdev_rcu(net, dev) {
675 if (dev->netdev_ops->ndo_del_vxlan_port)
676 dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
677 port);
678 }
679 rcu_read_unlock();
680
681 if (sa_family == AF_INET)
682 udp_del_offload(&vs->udp_offloads);
683 }
684
685 /* Add new entry to forwarding table -- assumes lock held */
686 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
687 const u8 *mac, union vxlan_addr *ip,
688 __u16 state, __u16 flags,
689 __be16 port, __u32 vni, __u32 ifindex,
690 __u8 ndm_flags)
691 {
692 struct vxlan_rdst *rd = NULL;
693 struct vxlan_fdb *f;
694 int notify = 0;
695
696 f = __vxlan_find_mac(vxlan, mac);
697 if (f) {
698 if (flags & NLM_F_EXCL) {
699 netdev_dbg(vxlan->dev,
700 "lost race to create %pM\n", mac);
701 return -EEXIST;
702 }
703 if (f->state != state) {
704 f->state = state;
705 f->updated = jiffies;
706 notify = 1;
707 }
708 if (f->flags != ndm_flags) {
709 f->flags = ndm_flags;
710 f->updated = jiffies;
711 notify = 1;
712 }
713 if ((flags & NLM_F_REPLACE)) {
714 /* Only change unicasts */
715 if (!(is_multicast_ether_addr(f->eth_addr) ||
716 is_zero_ether_addr(f->eth_addr))) {
717 int rc = vxlan_fdb_replace(f, ip, port, vni,
718 ifindex);
719
720 if (rc < 0)
721 return rc;
722 notify |= rc;
723 } else
724 return -EOPNOTSUPP;
725 }
726 if ((flags & NLM_F_APPEND) &&
727 (is_multicast_ether_addr(f->eth_addr) ||
728 is_zero_ether_addr(f->eth_addr))) {
729 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
730 &rd);
731
732 if (rc < 0)
733 return rc;
734 notify |= rc;
735 }
736 } else {
737 if (!(flags & NLM_F_CREATE))
738 return -ENOENT;
739
740 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
741 return -ENOSPC;
742
743 /* Disallow replace to add a multicast entry */
744 if ((flags & NLM_F_REPLACE) &&
745 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
746 return -EOPNOTSUPP;
747
748 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
749 f = kmalloc(sizeof(*f), GFP_ATOMIC);
750 if (!f)
751 return -ENOMEM;
752
753 notify = 1;
754 f->state = state;
755 f->flags = ndm_flags;
756 f->updated = f->used = jiffies;
757 INIT_LIST_HEAD(&f->remotes);
758 memcpy(f->eth_addr, mac, ETH_ALEN);
759
760 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
761
762 ++vxlan->addrcnt;
763 hlist_add_head_rcu(&f->hlist,
764 vxlan_fdb_head(vxlan, mac));
765 }
766
767 if (notify) {
768 if (rd == NULL)
769 rd = first_remote_rtnl(f);
770 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
771 }
772
773 return 0;
774 }
775
776 static void vxlan_fdb_free(struct rcu_head *head)
777 {
778 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
779 struct vxlan_rdst *rd, *nd;
780
781 list_for_each_entry_safe(rd, nd, &f->remotes, list)
782 kfree(rd);
783 kfree(f);
784 }
785
786 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
787 {
788 netdev_dbg(vxlan->dev,
789 "delete %pM\n", f->eth_addr);
790
791 --vxlan->addrcnt;
792 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
793
794 hlist_del_rcu(&f->hlist);
795 call_rcu(&f->rcu, vxlan_fdb_free);
796 }
797
798 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
799 union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
800 {
801 struct net *net = dev_net(vxlan->dev);
802 int err;
803
804 if (tb[NDA_DST]) {
805 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
806 if (err)
807 return err;
808 } else {
809 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
810 if (remote->sa.sa_family == AF_INET) {
811 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
812 ip->sa.sa_family = AF_INET;
813 #if IS_ENABLED(CONFIG_IPV6)
814 } else {
815 ip->sin6.sin6_addr = in6addr_any;
816 ip->sa.sa_family = AF_INET6;
817 #endif
818 }
819 }
820
821 if (tb[NDA_PORT]) {
822 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
823 return -EINVAL;
824 *port = nla_get_be16(tb[NDA_PORT]);
825 } else {
826 *port = vxlan->dst_port;
827 }
828
829 if (tb[NDA_VNI]) {
830 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
831 return -EINVAL;
832 *vni = nla_get_u32(tb[NDA_VNI]);
833 } else {
834 *vni = vxlan->default_dst.remote_vni;
835 }
836
837 if (tb[NDA_IFINDEX]) {
838 struct net_device *tdev;
839
840 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
841 return -EINVAL;
842 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
843 tdev = __dev_get_by_index(net, *ifindex);
844 if (!tdev)
845 return -EADDRNOTAVAIL;
846 } else {
847 *ifindex = 0;
848 }
849
850 return 0;
851 }
852
853 /* Add static entry (via netlink) */
854 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
855 struct net_device *dev,
856 const unsigned char *addr, u16 flags)
857 {
858 struct vxlan_dev *vxlan = netdev_priv(dev);
859 /* struct net *net = dev_net(vxlan->dev); */
860 union vxlan_addr ip;
861 __be16 port;
862 u32 vni, ifindex;
863 int err;
864
865 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
866 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
867 ndm->ndm_state);
868 return -EINVAL;
869 }
870
871 if (tb[NDA_DST] == NULL)
872 return -EINVAL;
873
874 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
875 if (err)
876 return err;
877
878 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
879 return -EAFNOSUPPORT;
880
881 spin_lock_bh(&vxlan->hash_lock);
882 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
883 port, vni, ifindex, ndm->ndm_flags);
884 spin_unlock_bh(&vxlan->hash_lock);
885
886 return err;
887 }
888
889 /* Delete entry (via netlink) */
890 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
891 struct net_device *dev,
892 const unsigned char *addr)
893 {
894 struct vxlan_dev *vxlan = netdev_priv(dev);
895 struct vxlan_fdb *f;
896 struct vxlan_rdst *rd = NULL;
897 union vxlan_addr ip;
898 __be16 port;
899 u32 vni, ifindex;
900 int err;
901
902 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
903 if (err)
904 return err;
905
906 err = -ENOENT;
907
908 spin_lock_bh(&vxlan->hash_lock);
909 f = vxlan_find_mac(vxlan, addr);
910 if (!f)
911 goto out;
912
913 if (!vxlan_addr_any(&ip)) {
914 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
915 if (!rd)
916 goto out;
917 }
918
919 err = 0;
920
921 /* remove a destination if it's not the only one on the list,
922 * otherwise destroy the fdb entry
923 */
924 if (rd && !list_is_singular(&f->remotes)) {
925 list_del_rcu(&rd->list);
926 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
927 kfree_rcu(rd, rcu);
928 goto out;
929 }
930
931 vxlan_fdb_destroy(vxlan, f);
932
933 out:
934 spin_unlock_bh(&vxlan->hash_lock);
935
936 return err;
937 }
938
939 /* Dump forwarding table */
940 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
941 struct net_device *dev, int idx)
942 {
943 struct vxlan_dev *vxlan = netdev_priv(dev);
944 unsigned int h;
945
946 for (h = 0; h < FDB_HASH_SIZE; ++h) {
947 struct vxlan_fdb *f;
948 int err;
949
950 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
951 struct vxlan_rdst *rd;
952
953 if (idx < cb->args[0])
954 goto skip;
955
956 list_for_each_entry_rcu(rd, &f->remotes, list) {
957 err = vxlan_fdb_info(skb, vxlan, f,
958 NETLINK_CB(cb->skb).portid,
959 cb->nlh->nlmsg_seq,
960 RTM_NEWNEIGH,
961 NLM_F_MULTI, rd);
962 if (err < 0)
963 goto out;
964 }
965 skip:
966 ++idx;
967 }
968 }
969 out:
970 return idx;
971 }
972
973 /* Watch incoming packets to learn mapping between Ethernet address
974 * and Tunnel endpoint.
975 * Return true if packet is bogus and should be droppped.
976 */
977 static bool vxlan_snoop(struct net_device *dev,
978 union vxlan_addr *src_ip, const u8 *src_mac)
979 {
980 struct vxlan_dev *vxlan = netdev_priv(dev);
981 struct vxlan_fdb *f;
982
983 f = vxlan_find_mac(vxlan, src_mac);
984 if (likely(f)) {
985 struct vxlan_rdst *rdst = first_remote_rcu(f);
986
987 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
988 return false;
989
990 /* Don't migrate static entries, drop packets */
991 if (f->state & NUD_NOARP)
992 return true;
993
994 if (net_ratelimit())
995 netdev_info(dev,
996 "%pM migrated from %pIS to %pIS\n",
997 src_mac, &rdst->remote_ip, &src_ip);
998
999 rdst->remote_ip = *src_ip;
1000 f->updated = jiffies;
1001 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1002 } else {
1003 /* learned new entry */
1004 spin_lock(&vxlan->hash_lock);
1005
1006 /* close off race between vxlan_flush and incoming packets */
1007 if (netif_running(dev))
1008 vxlan_fdb_create(vxlan, src_mac, src_ip,
1009 NUD_REACHABLE,
1010 NLM_F_EXCL|NLM_F_CREATE,
1011 vxlan->dst_port,
1012 vxlan->default_dst.remote_vni,
1013 0, NTF_SELF);
1014 spin_unlock(&vxlan->hash_lock);
1015 }
1016
1017 return false;
1018 }
1019
1020 /* See if multicast group is already in use by other ID */
1021 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1022 {
1023 struct vxlan_dev *vxlan;
1024
1025 /* The vxlan_sock is only used by dev, leaving group has
1026 * no effect on other vxlan devices.
1027 */
1028 if (atomic_read(&dev->vn_sock->refcnt) == 1)
1029 return false;
1030
1031 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1032 if (!netif_running(vxlan->dev) || vxlan == dev)
1033 continue;
1034
1035 if (vxlan->vn_sock != dev->vn_sock)
1036 continue;
1037
1038 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1039 &dev->default_dst.remote_ip))
1040 continue;
1041
1042 if (vxlan->default_dst.remote_ifindex !=
1043 dev->default_dst.remote_ifindex)
1044 continue;
1045
1046 return true;
1047 }
1048
1049 return false;
1050 }
1051
1052 static void vxlan_sock_hold(struct vxlan_sock *vs)
1053 {
1054 atomic_inc(&vs->refcnt);
1055 }
1056
1057 void vxlan_sock_release(struct vxlan_sock *vs)
1058 {
1059 struct sock *sk = vs->sock->sk;
1060 struct net *net = sock_net(sk);
1061 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1062
1063 if (!atomic_dec_and_test(&vs->refcnt))
1064 return;
1065
1066 spin_lock(&vn->sock_lock);
1067 hlist_del_rcu(&vs->hlist);
1068 rcu_assign_sk_user_data(vs->sock->sk, NULL);
1069 vxlan_notify_del_rx_port(vs);
1070 spin_unlock(&vn->sock_lock);
1071
1072 queue_work(vxlan_wq, &vs->del_work);
1073 }
1074 EXPORT_SYMBOL_GPL(vxlan_sock_release);
1075
1076 /* Callback to update multicast group membership when first VNI on
1077 * multicast asddress is brought up
1078 * Done as workqueue because ip_mc_join_group acquires RTNL.
1079 */
1080 static void vxlan_igmp_join(struct work_struct *work)
1081 {
1082 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join);
1083 struct vxlan_sock *vs = vxlan->vn_sock;
1084 struct sock *sk = vs->sock->sk;
1085 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1086 int ifindex = vxlan->default_dst.remote_ifindex;
1087
1088 lock_sock(sk);
1089 if (ip->sa.sa_family == AF_INET) {
1090 struct ip_mreqn mreq = {
1091 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1092 .imr_ifindex = ifindex,
1093 };
1094
1095 ip_mc_join_group(sk, &mreq);
1096 #if IS_ENABLED(CONFIG_IPV6)
1097 } else {
1098 ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1099 &ip->sin6.sin6_addr);
1100 #endif
1101 }
1102 release_sock(sk);
1103
1104 vxlan_sock_release(vs);
1105 dev_put(vxlan->dev);
1106 }
1107
1108 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1109 static void vxlan_igmp_leave(struct work_struct *work)
1110 {
1111 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave);
1112 struct vxlan_sock *vs = vxlan->vn_sock;
1113 struct sock *sk = vs->sock->sk;
1114 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1115 int ifindex = vxlan->default_dst.remote_ifindex;
1116
1117 lock_sock(sk);
1118 if (ip->sa.sa_family == AF_INET) {
1119 struct ip_mreqn mreq = {
1120 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1121 .imr_ifindex = ifindex,
1122 };
1123
1124 ip_mc_leave_group(sk, &mreq);
1125 #if IS_ENABLED(CONFIG_IPV6)
1126 } else {
1127 ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1128 &ip->sin6.sin6_addr);
1129 #endif
1130 }
1131
1132 release_sock(sk);
1133
1134 vxlan_sock_release(vs);
1135 dev_put(vxlan->dev);
1136 }
1137
1138 /* Callback from net/ipv4/udp.c to receive packets */
1139 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1140 {
1141 struct vxlan_sock *vs;
1142 struct vxlanhdr *vxh;
1143
1144 /* Need Vxlan and inner Ethernet header to be present */
1145 if (!pskb_may_pull(skb, VXLAN_HLEN))
1146 goto error;
1147
1148 /* Return packets with reserved bits set */
1149 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1150 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
1151 (vxh->vx_vni & htonl(0xff))) {
1152 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1153 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1154 goto error;
1155 }
1156
1157 if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1158 goto drop;
1159
1160 vs = rcu_dereference_sk_user_data(sk);
1161 if (!vs)
1162 goto drop;
1163
1164 /* If the NIC driver gave us an encapsulated packet
1165 * with the encapsulation mark, the device checksummed it
1166 * for us. Otherwise force the upper layers to verify it.
1167 */
1168 if ((skb->ip_summed != CHECKSUM_UNNECESSARY && skb->ip_summed != CHECKSUM_PARTIAL) ||
1169 !skb->encapsulation)
1170 skb->ip_summed = CHECKSUM_NONE;
1171
1172 skb->encapsulation = 0;
1173
1174 vs->rcv(vs, skb, vxh->vx_vni);
1175 return 0;
1176
1177 drop:
1178 /* Consume bad packet */
1179 kfree_skb(skb);
1180 return 0;
1181
1182 error:
1183 /* Return non vxlan pkt */
1184 return 1;
1185 }
1186
1187 static void vxlan_rcv(struct vxlan_sock *vs,
1188 struct sk_buff *skb, __be32 vx_vni)
1189 {
1190 struct iphdr *oip = NULL;
1191 struct ipv6hdr *oip6 = NULL;
1192 struct vxlan_dev *vxlan;
1193 struct pcpu_sw_netstats *stats;
1194 union vxlan_addr saddr;
1195 __u32 vni;
1196 int err = 0;
1197 union vxlan_addr *remote_ip;
1198
1199 vni = ntohl(vx_vni) >> 8;
1200 /* Is this VNI defined? */
1201 vxlan = vxlan_vs_find_vni(vs, vni);
1202 if (!vxlan)
1203 goto drop;
1204
1205 remote_ip = &vxlan->default_dst.remote_ip;
1206 skb_reset_mac_header(skb);
1207 skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1208 skb->protocol = eth_type_trans(skb, vxlan->dev);
1209
1210 /* Ignore packet loops (and multicast echo) */
1211 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1212 goto drop;
1213
1214 /* Re-examine inner Ethernet packet */
1215 if (remote_ip->sa.sa_family == AF_INET) {
1216 oip = ip_hdr(skb);
1217 saddr.sin.sin_addr.s_addr = oip->saddr;
1218 saddr.sa.sa_family = AF_INET;
1219 #if IS_ENABLED(CONFIG_IPV6)
1220 } else {
1221 oip6 = ipv6_hdr(skb);
1222 saddr.sin6.sin6_addr = oip6->saddr;
1223 saddr.sa.sa_family = AF_INET6;
1224 #endif
1225 }
1226
1227 if ((vxlan->flags & VXLAN_F_LEARN) &&
1228 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1229 goto drop;
1230
1231 skb_reset_network_header(skb);
1232
1233 if (oip6)
1234 err = IP6_ECN_decapsulate(oip6, skb);
1235 if (oip)
1236 err = IP_ECN_decapsulate(oip, skb);
1237
1238 if (unlikely(err)) {
1239 if (log_ecn_error) {
1240 if (oip6)
1241 net_info_ratelimited("non-ECT from %pI6\n",
1242 &oip6->saddr);
1243 if (oip)
1244 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1245 &oip->saddr, oip->tos);
1246 }
1247 if (err > 1) {
1248 ++vxlan->dev->stats.rx_frame_errors;
1249 ++vxlan->dev->stats.rx_errors;
1250 goto drop;
1251 }
1252 }
1253
1254 stats = this_cpu_ptr(vxlan->dev->tstats);
1255 u64_stats_update_begin(&stats->syncp);
1256 stats->rx_packets++;
1257 stats->rx_bytes += skb->len;
1258 u64_stats_update_end(&stats->syncp);
1259
1260 netif_rx(skb);
1261
1262 return;
1263 drop:
1264 /* Consume bad packet */
1265 kfree_skb(skb);
1266 }
1267
1268 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1269 {
1270 struct vxlan_dev *vxlan = netdev_priv(dev);
1271 struct arphdr *parp;
1272 u8 *arpptr, *sha;
1273 __be32 sip, tip;
1274 struct neighbour *n;
1275
1276 if (dev->flags & IFF_NOARP)
1277 goto out;
1278
1279 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1280 dev->stats.tx_dropped++;
1281 goto out;
1282 }
1283 parp = arp_hdr(skb);
1284
1285 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1286 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1287 parp->ar_pro != htons(ETH_P_IP) ||
1288 parp->ar_op != htons(ARPOP_REQUEST) ||
1289 parp->ar_hln != dev->addr_len ||
1290 parp->ar_pln != 4)
1291 goto out;
1292 arpptr = (u8 *)parp + sizeof(struct arphdr);
1293 sha = arpptr;
1294 arpptr += dev->addr_len; /* sha */
1295 memcpy(&sip, arpptr, sizeof(sip));
1296 arpptr += sizeof(sip);
1297 arpptr += dev->addr_len; /* tha */
1298 memcpy(&tip, arpptr, sizeof(tip));
1299
1300 if (ipv4_is_loopback(tip) ||
1301 ipv4_is_multicast(tip))
1302 goto out;
1303
1304 n = neigh_lookup(&arp_tbl, &tip, dev);
1305
1306 if (n) {
1307 struct vxlan_fdb *f;
1308 struct sk_buff *reply;
1309
1310 if (!(n->nud_state & NUD_CONNECTED)) {
1311 neigh_release(n);
1312 goto out;
1313 }
1314
1315 f = vxlan_find_mac(vxlan, n->ha);
1316 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1317 /* bridge-local neighbor */
1318 neigh_release(n);
1319 goto out;
1320 }
1321
1322 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1323 n->ha, sha);
1324
1325 neigh_release(n);
1326
1327 if (reply == NULL)
1328 goto out;
1329
1330 skb_reset_mac_header(reply);
1331 __skb_pull(reply, skb_network_offset(reply));
1332 reply->ip_summed = CHECKSUM_UNNECESSARY;
1333 reply->pkt_type = PACKET_HOST;
1334
1335 if (netif_rx_ni(reply) == NET_RX_DROP)
1336 dev->stats.rx_dropped++;
1337 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1338 union vxlan_addr ipa = {
1339 .sin.sin_addr.s_addr = tip,
1340 .sa.sa_family = AF_INET,
1341 };
1342
1343 vxlan_ip_miss(dev, &ipa);
1344 }
1345 out:
1346 consume_skb(skb);
1347 return NETDEV_TX_OK;
1348 }
1349
1350 #if IS_ENABLED(CONFIG_IPV6)
1351
1352 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1353 struct neighbour *n, bool isrouter)
1354 {
1355 struct net_device *dev = request->dev;
1356 struct sk_buff *reply;
1357 struct nd_msg *ns, *na;
1358 struct ipv6hdr *pip6;
1359 u8 *daddr;
1360 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1361 int ns_olen;
1362 int i, len;
1363
1364 if (dev == NULL)
1365 return NULL;
1366
1367 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1368 sizeof(*na) + na_olen + dev->needed_tailroom;
1369 reply = alloc_skb(len, GFP_ATOMIC);
1370 if (reply == NULL)
1371 return NULL;
1372
1373 reply->protocol = htons(ETH_P_IPV6);
1374 reply->dev = dev;
1375 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1376 skb_push(reply, sizeof(struct ethhdr));
1377 skb_set_mac_header(reply, 0);
1378
1379 ns = (struct nd_msg *)skb_transport_header(request);
1380
1381 daddr = eth_hdr(request)->h_source;
1382 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1383 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1384 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1385 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1386 break;
1387 }
1388 }
1389
1390 /* Ethernet header */
1391 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1392 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1393 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1394 reply->protocol = htons(ETH_P_IPV6);
1395
1396 skb_pull(reply, sizeof(struct ethhdr));
1397 skb_set_network_header(reply, 0);
1398 skb_put(reply, sizeof(struct ipv6hdr));
1399
1400 /* IPv6 header */
1401
1402 pip6 = ipv6_hdr(reply);
1403 memset(pip6, 0, sizeof(struct ipv6hdr));
1404 pip6->version = 6;
1405 pip6->priority = ipv6_hdr(request)->priority;
1406 pip6->nexthdr = IPPROTO_ICMPV6;
1407 pip6->hop_limit = 255;
1408 pip6->daddr = ipv6_hdr(request)->saddr;
1409 pip6->saddr = *(struct in6_addr *)n->primary_key;
1410
1411 skb_pull(reply, sizeof(struct ipv6hdr));
1412 skb_set_transport_header(reply, 0);
1413
1414 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1415
1416 /* Neighbor Advertisement */
1417 memset(na, 0, sizeof(*na)+na_olen);
1418 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1419 na->icmph.icmp6_router = isrouter;
1420 na->icmph.icmp6_override = 1;
1421 na->icmph.icmp6_solicited = 1;
1422 na->target = ns->target;
1423 ether_addr_copy(&na->opt[2], n->ha);
1424 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1425 na->opt[1] = na_olen >> 3;
1426
1427 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1428 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1429 csum_partial(na, sizeof(*na)+na_olen, 0));
1430
1431 pip6->payload_len = htons(sizeof(*na)+na_olen);
1432
1433 skb_push(reply, sizeof(struct ipv6hdr));
1434
1435 reply->ip_summed = CHECKSUM_UNNECESSARY;
1436
1437 return reply;
1438 }
1439
1440 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1441 {
1442 struct vxlan_dev *vxlan = netdev_priv(dev);
1443 struct nd_msg *msg;
1444 const struct ipv6hdr *iphdr;
1445 const struct in6_addr *saddr, *daddr;
1446 struct neighbour *n;
1447 struct inet6_dev *in6_dev;
1448
1449 in6_dev = __in6_dev_get(dev);
1450 if (!in6_dev)
1451 goto out;
1452
1453 if (!pskb_may_pull(skb, skb->len))
1454 goto out;
1455
1456 iphdr = ipv6_hdr(skb);
1457 saddr = &iphdr->saddr;
1458 daddr = &iphdr->daddr;
1459
1460 msg = (struct nd_msg *)skb_transport_header(skb);
1461 if (msg->icmph.icmp6_code != 0 ||
1462 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1463 goto out;
1464
1465 if (ipv6_addr_loopback(daddr) ||
1466 ipv6_addr_is_multicast(&msg->target))
1467 goto out;
1468
1469 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1470
1471 if (n) {
1472 struct vxlan_fdb *f;
1473 struct sk_buff *reply;
1474
1475 if (!(n->nud_state & NUD_CONNECTED)) {
1476 neigh_release(n);
1477 goto out;
1478 }
1479
1480 f = vxlan_find_mac(vxlan, n->ha);
1481 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1482 /* bridge-local neighbor */
1483 neigh_release(n);
1484 goto out;
1485 }
1486
1487 reply = vxlan_na_create(skb, n,
1488 !!(f ? f->flags & NTF_ROUTER : 0));
1489
1490 neigh_release(n);
1491
1492 if (reply == NULL)
1493 goto out;
1494
1495 if (netif_rx_ni(reply) == NET_RX_DROP)
1496 dev->stats.rx_dropped++;
1497
1498 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1499 union vxlan_addr ipa = {
1500 .sin6.sin6_addr = msg->target,
1501 .sa.sa_family = AF_INET6,
1502 };
1503
1504 vxlan_ip_miss(dev, &ipa);
1505 }
1506
1507 out:
1508 consume_skb(skb);
1509 return NETDEV_TX_OK;
1510 }
1511 #endif
1512
1513 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1514 {
1515 struct vxlan_dev *vxlan = netdev_priv(dev);
1516 struct neighbour *n;
1517
1518 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1519 return false;
1520
1521 n = NULL;
1522 switch (ntohs(eth_hdr(skb)->h_proto)) {
1523 case ETH_P_IP:
1524 {
1525 struct iphdr *pip;
1526
1527 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1528 return false;
1529 pip = ip_hdr(skb);
1530 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1531 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1532 union vxlan_addr ipa = {
1533 .sin.sin_addr.s_addr = pip->daddr,
1534 .sa.sa_family = AF_INET,
1535 };
1536
1537 vxlan_ip_miss(dev, &ipa);
1538 return false;
1539 }
1540
1541 break;
1542 }
1543 #if IS_ENABLED(CONFIG_IPV6)
1544 case ETH_P_IPV6:
1545 {
1546 struct ipv6hdr *pip6;
1547
1548 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1549 return false;
1550 pip6 = ipv6_hdr(skb);
1551 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1552 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1553 union vxlan_addr ipa = {
1554 .sin6.sin6_addr = pip6->daddr,
1555 .sa.sa_family = AF_INET6,
1556 };
1557
1558 vxlan_ip_miss(dev, &ipa);
1559 return false;
1560 }
1561
1562 break;
1563 }
1564 #endif
1565 default:
1566 return false;
1567 }
1568
1569 if (n) {
1570 bool diff;
1571
1572 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1573 if (diff) {
1574 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1575 dev->addr_len);
1576 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1577 }
1578 neigh_release(n);
1579 return diff;
1580 }
1581
1582 return false;
1583 }
1584
1585 /* Compute source port for outgoing packet
1586 * first choice to use L4 flow hash since it will spread
1587 * better and maybe available from hardware
1588 * secondary choice is to use jhash on the Ethernet header
1589 */
1590 __be16 vxlan_src_port(__u16 port_min, __u16 port_max, struct sk_buff *skb)
1591 {
1592 unsigned int range = (port_max - port_min) + 1;
1593 u32 hash;
1594
1595 hash = skb_get_hash(skb);
1596 if (!hash)
1597 hash = jhash(skb->data, 2 * ETH_ALEN,
1598 (__force u32) skb->protocol);
1599
1600 return htons((((u64) hash * range) >> 32) + port_min);
1601 }
1602 EXPORT_SYMBOL_GPL(vxlan_src_port);
1603
1604 static int handle_offloads(struct sk_buff *skb)
1605 {
1606 if (skb_is_gso(skb)) {
1607 int err = skb_unclone(skb, GFP_ATOMIC);
1608 if (unlikely(err))
1609 return err;
1610
1611 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
1612 } else if (skb->ip_summed != CHECKSUM_PARTIAL)
1613 skb->ip_summed = CHECKSUM_NONE;
1614
1615 return 0;
1616 }
1617
1618 #if IS_ENABLED(CONFIG_IPV6)
1619 static int vxlan6_xmit_skb(struct vxlan_sock *vs,
1620 struct dst_entry *dst, struct sk_buff *skb,
1621 struct net_device *dev, struct in6_addr *saddr,
1622 struct in6_addr *daddr, __u8 prio, __u8 ttl,
1623 __be16 src_port, __be16 dst_port, __be32 vni,
1624 bool xnet)
1625 {
1626 struct ipv6hdr *ip6h;
1627 struct vxlanhdr *vxh;
1628 struct udphdr *uh;
1629 int min_headroom;
1630 int err;
1631
1632 if (!skb->encapsulation) {
1633 skb_reset_inner_headers(skb);
1634 skb->encapsulation = 1;
1635 }
1636
1637 skb_scrub_packet(skb, xnet);
1638
1639 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1640 + VXLAN_HLEN + sizeof(struct ipv6hdr)
1641 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1642
1643 /* Need space for new headers (invalidates iph ptr) */
1644 err = skb_cow_head(skb, min_headroom);
1645 if (unlikely(err))
1646 return err;
1647
1648 if (vlan_tx_tag_present(skb)) {
1649 if (WARN_ON(!__vlan_put_tag(skb,
1650 skb->vlan_proto,
1651 vlan_tx_tag_get(skb))))
1652 return -ENOMEM;
1653
1654 skb->vlan_tci = 0;
1655 }
1656
1657 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1658 vxh->vx_flags = htonl(VXLAN_FLAGS);
1659 vxh->vx_vni = vni;
1660
1661 __skb_push(skb, sizeof(*uh));
1662 skb_reset_transport_header(skb);
1663 uh = udp_hdr(skb);
1664
1665 uh->dest = dst_port;
1666 uh->source = src_port;
1667
1668 uh->len = htons(skb->len);
1669 uh->check = 0;
1670
1671 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1672 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1673 IPSKB_REROUTED);
1674 skb_dst_set(skb, dst);
1675
1676 if (!skb_is_gso(skb) && !(dst->dev->features & NETIF_F_IPV6_CSUM)) {
1677 __wsum csum = skb_checksum(skb, 0, skb->len, 0);
1678 skb->ip_summed = CHECKSUM_UNNECESSARY;
1679 uh->check = csum_ipv6_magic(saddr, daddr, skb->len,
1680 IPPROTO_UDP, csum);
1681 if (uh->check == 0)
1682 uh->check = CSUM_MANGLED_0;
1683 } else {
1684 skb->ip_summed = CHECKSUM_PARTIAL;
1685 skb->csum_start = skb_transport_header(skb) - skb->head;
1686 skb->csum_offset = offsetof(struct udphdr, check);
1687 uh->check = ~csum_ipv6_magic(saddr, daddr,
1688 skb->len, IPPROTO_UDP, 0);
1689 }
1690
1691 __skb_push(skb, sizeof(*ip6h));
1692 skb_reset_network_header(skb);
1693 ip6h = ipv6_hdr(skb);
1694 ip6h->version = 6;
1695 ip6h->priority = prio;
1696 ip6h->flow_lbl[0] = 0;
1697 ip6h->flow_lbl[1] = 0;
1698 ip6h->flow_lbl[2] = 0;
1699 ip6h->payload_len = htons(skb->len);
1700 ip6h->nexthdr = IPPROTO_UDP;
1701 ip6h->hop_limit = ttl;
1702 ip6h->daddr = *daddr;
1703 ip6h->saddr = *saddr;
1704
1705 err = handle_offloads(skb);
1706 if (err)
1707 return err;
1708
1709 ip6tunnel_xmit(skb, dev);
1710 return 0;
1711 }
1712 #endif
1713
1714 int vxlan_xmit_skb(struct vxlan_sock *vs,
1715 struct rtable *rt, struct sk_buff *skb,
1716 __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1717 __be16 src_port, __be16 dst_port, __be32 vni, bool xnet)
1718 {
1719 struct vxlanhdr *vxh;
1720 struct udphdr *uh;
1721 int min_headroom;
1722 int err;
1723
1724 if (!skb->encapsulation) {
1725 skb_reset_inner_headers(skb);
1726 skb->encapsulation = 1;
1727 }
1728
1729 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1730 + VXLAN_HLEN + sizeof(struct iphdr)
1731 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1732
1733 /* Need space for new headers (invalidates iph ptr) */
1734 err = skb_cow_head(skb, min_headroom);
1735 if (unlikely(err))
1736 return err;
1737
1738 if (vlan_tx_tag_present(skb)) {
1739 if (WARN_ON(!__vlan_put_tag(skb,
1740 skb->vlan_proto,
1741 vlan_tx_tag_get(skb))))
1742 return -ENOMEM;
1743
1744 skb->vlan_tci = 0;
1745 }
1746
1747 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1748 vxh->vx_flags = htonl(VXLAN_FLAGS);
1749 vxh->vx_vni = vni;
1750
1751 __skb_push(skb, sizeof(*uh));
1752 skb_reset_transport_header(skb);
1753 uh = udp_hdr(skb);
1754
1755 uh->dest = dst_port;
1756 uh->source = src_port;
1757
1758 uh->len = htons(skb->len);
1759 uh->check = 0;
1760
1761 err = handle_offloads(skb);
1762 if (err)
1763 return err;
1764
1765 return iptunnel_xmit(vs->sock->sk, rt, skb, src, dst, IPPROTO_UDP,
1766 tos, ttl, df, xnet);
1767 }
1768 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1769
1770 /* Bypass encapsulation if the destination is local */
1771 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1772 struct vxlan_dev *dst_vxlan)
1773 {
1774 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1775 union vxlan_addr loopback;
1776 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1777
1778 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1779 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1780 skb->pkt_type = PACKET_HOST;
1781 skb->encapsulation = 0;
1782 skb->dev = dst_vxlan->dev;
1783 __skb_pull(skb, skb_network_offset(skb));
1784
1785 if (remote_ip->sa.sa_family == AF_INET) {
1786 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1787 loopback.sa.sa_family = AF_INET;
1788 #if IS_ENABLED(CONFIG_IPV6)
1789 } else {
1790 loopback.sin6.sin6_addr = in6addr_loopback;
1791 loopback.sa.sa_family = AF_INET6;
1792 #endif
1793 }
1794
1795 if (dst_vxlan->flags & VXLAN_F_LEARN)
1796 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1797
1798 u64_stats_update_begin(&tx_stats->syncp);
1799 tx_stats->tx_packets++;
1800 tx_stats->tx_bytes += skb->len;
1801 u64_stats_update_end(&tx_stats->syncp);
1802
1803 if (netif_rx(skb) == NET_RX_SUCCESS) {
1804 u64_stats_update_begin(&rx_stats->syncp);
1805 rx_stats->rx_packets++;
1806 rx_stats->rx_bytes += skb->len;
1807 u64_stats_update_end(&rx_stats->syncp);
1808 } else {
1809 skb->dev->stats.rx_dropped++;
1810 }
1811 }
1812
1813 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1814 struct vxlan_rdst *rdst, bool did_rsc)
1815 {
1816 struct vxlan_dev *vxlan = netdev_priv(dev);
1817 struct rtable *rt = NULL;
1818 const struct iphdr *old_iph;
1819 struct flowi4 fl4;
1820 union vxlan_addr *dst;
1821 __be16 src_port = 0, dst_port;
1822 u32 vni;
1823 __be16 df = 0;
1824 __u8 tos, ttl;
1825 int err;
1826
1827 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1828 vni = rdst->remote_vni;
1829 dst = &rdst->remote_ip;
1830
1831 if (vxlan_addr_any(dst)) {
1832 if (did_rsc) {
1833 /* short-circuited back to local bridge */
1834 vxlan_encap_bypass(skb, vxlan, vxlan);
1835 return;
1836 }
1837 goto drop;
1838 }
1839
1840 old_iph = ip_hdr(skb);
1841
1842 ttl = vxlan->ttl;
1843 if (!ttl && vxlan_addr_multicast(dst))
1844 ttl = 1;
1845
1846 tos = vxlan->tos;
1847 if (tos == 1)
1848 tos = ip_tunnel_get_dsfield(old_iph, skb);
1849
1850 src_port = vxlan_src_port(vxlan->port_min, vxlan->port_max, skb);
1851
1852 if (dst->sa.sa_family == AF_INET) {
1853 memset(&fl4, 0, sizeof(fl4));
1854 fl4.flowi4_oif = rdst->remote_ifindex;
1855 fl4.flowi4_tos = RT_TOS(tos);
1856 fl4.daddr = dst->sin.sin_addr.s_addr;
1857 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1858
1859 rt = ip_route_output_key(vxlan->net, &fl4);
1860 if (IS_ERR(rt)) {
1861 netdev_dbg(dev, "no route to %pI4\n",
1862 &dst->sin.sin_addr.s_addr);
1863 dev->stats.tx_carrier_errors++;
1864 goto tx_error;
1865 }
1866
1867 if (rt->dst.dev == dev) {
1868 netdev_dbg(dev, "circular route to %pI4\n",
1869 &dst->sin.sin_addr.s_addr);
1870 dev->stats.collisions++;
1871 goto rt_tx_error;
1872 }
1873
1874 /* Bypass encapsulation if the destination is local */
1875 if (rt->rt_flags & RTCF_LOCAL &&
1876 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1877 struct vxlan_dev *dst_vxlan;
1878
1879 ip_rt_put(rt);
1880 dst_vxlan = vxlan_find_vni(vxlan->net, vni, dst_port);
1881 if (!dst_vxlan)
1882 goto tx_error;
1883 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1884 return;
1885 }
1886
1887 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1888 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1889
1890 err = vxlan_xmit_skb(vxlan->vn_sock, rt, skb,
1891 fl4.saddr, dst->sin.sin_addr.s_addr,
1892 tos, ttl, df, src_port, dst_port,
1893 htonl(vni << 8),
1894 !net_eq(vxlan->net, dev_net(vxlan->dev)));
1895
1896 if (err < 0)
1897 goto rt_tx_error;
1898 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1899 #if IS_ENABLED(CONFIG_IPV6)
1900 } else {
1901 struct sock *sk = vxlan->vn_sock->sock->sk;
1902 struct dst_entry *ndst;
1903 struct flowi6 fl6;
1904 u32 flags;
1905
1906 memset(&fl6, 0, sizeof(fl6));
1907 fl6.flowi6_oif = rdst->remote_ifindex;
1908 fl6.daddr = dst->sin6.sin6_addr;
1909 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
1910 fl6.flowi6_proto = IPPROTO_UDP;
1911
1912 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
1913 netdev_dbg(dev, "no route to %pI6\n",
1914 &dst->sin6.sin6_addr);
1915 dev->stats.tx_carrier_errors++;
1916 goto tx_error;
1917 }
1918
1919 if (ndst->dev == dev) {
1920 netdev_dbg(dev, "circular route to %pI6\n",
1921 &dst->sin6.sin6_addr);
1922 dst_release(ndst);
1923 dev->stats.collisions++;
1924 goto tx_error;
1925 }
1926
1927 /* Bypass encapsulation if the destination is local */
1928 flags = ((struct rt6_info *)ndst)->rt6i_flags;
1929 if (flags & RTF_LOCAL &&
1930 !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1931 struct vxlan_dev *dst_vxlan;
1932
1933 dst_release(ndst);
1934 dst_vxlan = vxlan_find_vni(vxlan->net, vni, dst_port);
1935 if (!dst_vxlan)
1936 goto tx_error;
1937 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1938 return;
1939 }
1940
1941 ttl = ttl ? : ip6_dst_hoplimit(ndst);
1942
1943 err = vxlan6_xmit_skb(vxlan->vn_sock, ndst, skb,
1944 dev, &fl6.saddr, &fl6.daddr, 0, ttl,
1945 src_port, dst_port, htonl(vni << 8),
1946 !net_eq(vxlan->net, dev_net(vxlan->dev)));
1947 #endif
1948 }
1949
1950 return;
1951
1952 drop:
1953 dev->stats.tx_dropped++;
1954 goto tx_free;
1955
1956 rt_tx_error:
1957 ip_rt_put(rt);
1958 tx_error:
1959 dev->stats.tx_errors++;
1960 tx_free:
1961 dev_kfree_skb(skb);
1962 }
1963
1964 /* Transmit local packets over Vxlan
1965 *
1966 * Outer IP header inherits ECN and DF from inner header.
1967 * Outer UDP destination is the VXLAN assigned port.
1968 * source port is based on hash of flow
1969 */
1970 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1971 {
1972 struct vxlan_dev *vxlan = netdev_priv(dev);
1973 struct ethhdr *eth;
1974 bool did_rsc = false;
1975 struct vxlan_rdst *rdst, *fdst = NULL;
1976 struct vxlan_fdb *f;
1977
1978 skb_reset_mac_header(skb);
1979 eth = eth_hdr(skb);
1980
1981 if ((vxlan->flags & VXLAN_F_PROXY)) {
1982 if (ntohs(eth->h_proto) == ETH_P_ARP)
1983 return arp_reduce(dev, skb);
1984 #if IS_ENABLED(CONFIG_IPV6)
1985 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
1986 skb->len >= sizeof(struct ipv6hdr) + sizeof(struct nd_msg) &&
1987 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
1988 struct nd_msg *msg;
1989
1990 msg = (struct nd_msg *)skb_transport_header(skb);
1991 if (msg->icmph.icmp6_code == 0 &&
1992 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
1993 return neigh_reduce(dev, skb);
1994 }
1995 #endif
1996 }
1997
1998 f = vxlan_find_mac(vxlan, eth->h_dest);
1999 did_rsc = false;
2000
2001 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2002 (ntohs(eth->h_proto) == ETH_P_IP ||
2003 ntohs(eth->h_proto) == ETH_P_IPV6)) {
2004 did_rsc = route_shortcircuit(dev, skb);
2005 if (did_rsc)
2006 f = vxlan_find_mac(vxlan, eth->h_dest);
2007 }
2008
2009 if (f == NULL) {
2010 f = vxlan_find_mac(vxlan, all_zeros_mac);
2011 if (f == NULL) {
2012 if ((vxlan->flags & VXLAN_F_L2MISS) &&
2013 !is_multicast_ether_addr(eth->h_dest))
2014 vxlan_fdb_miss(vxlan, eth->h_dest);
2015
2016 dev->stats.tx_dropped++;
2017 kfree_skb(skb);
2018 return NETDEV_TX_OK;
2019 }
2020 }
2021
2022 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2023 struct sk_buff *skb1;
2024
2025 if (!fdst) {
2026 fdst = rdst;
2027 continue;
2028 }
2029 skb1 = skb_clone(skb, GFP_ATOMIC);
2030 if (skb1)
2031 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2032 }
2033
2034 if (fdst)
2035 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2036 else
2037 kfree_skb(skb);
2038 return NETDEV_TX_OK;
2039 }
2040
2041 /* Walk the forwarding table and purge stale entries */
2042 static void vxlan_cleanup(unsigned long arg)
2043 {
2044 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2045 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2046 unsigned int h;
2047
2048 if (!netif_running(vxlan->dev))
2049 return;
2050
2051 spin_lock_bh(&vxlan->hash_lock);
2052 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2053 struct hlist_node *p, *n;
2054 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2055 struct vxlan_fdb *f
2056 = container_of(p, struct vxlan_fdb, hlist);
2057 unsigned long timeout;
2058
2059 if (f->state & NUD_PERMANENT)
2060 continue;
2061
2062 timeout = f->used + vxlan->age_interval * HZ;
2063 if (time_before_eq(timeout, jiffies)) {
2064 netdev_dbg(vxlan->dev,
2065 "garbage collect %pM\n",
2066 f->eth_addr);
2067 f->state = NUD_STALE;
2068 vxlan_fdb_destroy(vxlan, f);
2069 } else if (time_before(timeout, next_timer))
2070 next_timer = timeout;
2071 }
2072 }
2073 spin_unlock_bh(&vxlan->hash_lock);
2074
2075 mod_timer(&vxlan->age_timer, next_timer);
2076 }
2077
2078 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2079 {
2080 __u32 vni = vxlan->default_dst.remote_vni;
2081
2082 vxlan->vn_sock = vs;
2083 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2084 }
2085
2086 /* Setup stats when device is created */
2087 static int vxlan_init(struct net_device *dev)
2088 {
2089 struct vxlan_dev *vxlan = netdev_priv(dev);
2090 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2091 struct vxlan_sock *vs;
2092
2093 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2094 if (!dev->tstats)
2095 return -ENOMEM;
2096
2097 spin_lock(&vn->sock_lock);
2098 vs = vxlan_find_sock(vxlan->net, vxlan->dst_port);
2099 if (vs) {
2100 /* If we have a socket with same port already, reuse it */
2101 atomic_inc(&vs->refcnt);
2102 vxlan_vs_add_dev(vs, vxlan);
2103 } else {
2104 /* otherwise make new socket outside of RTNL */
2105 dev_hold(dev);
2106 queue_work(vxlan_wq, &vxlan->sock_work);
2107 }
2108 spin_unlock(&vn->sock_lock);
2109
2110 return 0;
2111 }
2112
2113 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2114 {
2115 struct vxlan_fdb *f;
2116
2117 spin_lock_bh(&vxlan->hash_lock);
2118 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2119 if (f)
2120 vxlan_fdb_destroy(vxlan, f);
2121 spin_unlock_bh(&vxlan->hash_lock);
2122 }
2123
2124 static void vxlan_uninit(struct net_device *dev)
2125 {
2126 struct vxlan_dev *vxlan = netdev_priv(dev);
2127 struct vxlan_sock *vs = vxlan->vn_sock;
2128
2129 vxlan_fdb_delete_default(vxlan);
2130
2131 if (vs)
2132 vxlan_sock_release(vs);
2133 free_percpu(dev->tstats);
2134 }
2135
2136 /* Start ageing timer and join group when device is brought up */
2137 static int vxlan_open(struct net_device *dev)
2138 {
2139 struct vxlan_dev *vxlan = netdev_priv(dev);
2140 struct vxlan_sock *vs = vxlan->vn_sock;
2141
2142 /* socket hasn't been created */
2143 if (!vs)
2144 return -ENOTCONN;
2145
2146 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2147 vxlan_sock_hold(vs);
2148 dev_hold(dev);
2149 queue_work(vxlan_wq, &vxlan->igmp_join);
2150 }
2151
2152 if (vxlan->age_interval)
2153 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2154
2155 return 0;
2156 }
2157
2158 /* Purge the forwarding table */
2159 static void vxlan_flush(struct vxlan_dev *vxlan)
2160 {
2161 unsigned int h;
2162
2163 spin_lock_bh(&vxlan->hash_lock);
2164 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2165 struct hlist_node *p, *n;
2166 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2167 struct vxlan_fdb *f
2168 = container_of(p, struct vxlan_fdb, hlist);
2169 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2170 if (!is_zero_ether_addr(f->eth_addr))
2171 vxlan_fdb_destroy(vxlan, f);
2172 }
2173 }
2174 spin_unlock_bh(&vxlan->hash_lock);
2175 }
2176
2177 /* Cleanup timer and forwarding table on shutdown */
2178 static int vxlan_stop(struct net_device *dev)
2179 {
2180 struct vxlan_dev *vxlan = netdev_priv(dev);
2181 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2182 struct vxlan_sock *vs = vxlan->vn_sock;
2183
2184 if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2185 !vxlan_group_used(vn, vxlan)) {
2186 vxlan_sock_hold(vs);
2187 dev_hold(dev);
2188 queue_work(vxlan_wq, &vxlan->igmp_leave);
2189 }
2190
2191 del_timer_sync(&vxlan->age_timer);
2192
2193 vxlan_flush(vxlan);
2194
2195 return 0;
2196 }
2197
2198 /* Stub, nothing needs to be done. */
2199 static void vxlan_set_multicast_list(struct net_device *dev)
2200 {
2201 }
2202
2203 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2204 {
2205 struct vxlan_dev *vxlan = netdev_priv(dev);
2206 struct vxlan_rdst *dst = &vxlan->default_dst;
2207 struct net_device *lowerdev;
2208 int max_mtu;
2209
2210 lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2211 if (lowerdev == NULL)
2212 return eth_change_mtu(dev, new_mtu);
2213
2214 if (dst->remote_ip.sa.sa_family == AF_INET6)
2215 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2216 else
2217 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2218
2219 if (new_mtu < 68 || new_mtu > max_mtu)
2220 return -EINVAL;
2221
2222 dev->mtu = new_mtu;
2223 return 0;
2224 }
2225
2226 static const struct net_device_ops vxlan_netdev_ops = {
2227 .ndo_init = vxlan_init,
2228 .ndo_uninit = vxlan_uninit,
2229 .ndo_open = vxlan_open,
2230 .ndo_stop = vxlan_stop,
2231 .ndo_start_xmit = vxlan_xmit,
2232 .ndo_get_stats64 = ip_tunnel_get_stats64,
2233 .ndo_set_rx_mode = vxlan_set_multicast_list,
2234 .ndo_change_mtu = vxlan_change_mtu,
2235 .ndo_validate_addr = eth_validate_addr,
2236 .ndo_set_mac_address = eth_mac_addr,
2237 .ndo_fdb_add = vxlan_fdb_add,
2238 .ndo_fdb_del = vxlan_fdb_delete,
2239 .ndo_fdb_dump = vxlan_fdb_dump,
2240 };
2241
2242 /* Info for udev, that this is a virtual tunnel endpoint */
2243 static struct device_type vxlan_type = {
2244 .name = "vxlan",
2245 };
2246
2247 /* Calls the ndo_add_vxlan_port of the caller in order to
2248 * supply the listening VXLAN udp ports. Callers are expected
2249 * to implement the ndo_add_vxlan_port.
2250 */
2251 void vxlan_get_rx_port(struct net_device *dev)
2252 {
2253 struct vxlan_sock *vs;
2254 struct net *net = dev_net(dev);
2255 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2256 sa_family_t sa_family;
2257 __be16 port;
2258 unsigned int i;
2259
2260 spin_lock(&vn->sock_lock);
2261 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2262 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2263 port = inet_sk(vs->sock->sk)->inet_sport;
2264 sa_family = vs->sock->sk->sk_family;
2265 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2266 port);
2267 }
2268 }
2269 spin_unlock(&vn->sock_lock);
2270 }
2271 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2272
2273 /* Initialize the device structure. */
2274 static void vxlan_setup(struct net_device *dev)
2275 {
2276 struct vxlan_dev *vxlan = netdev_priv(dev);
2277 unsigned int h;
2278 int low, high;
2279
2280 eth_hw_addr_random(dev);
2281 ether_setup(dev);
2282 if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2283 dev->hard_header_len = ETH_HLEN + VXLAN6_HEADROOM;
2284 else
2285 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
2286
2287 dev->netdev_ops = &vxlan_netdev_ops;
2288 dev->destructor = free_netdev;
2289 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2290
2291 dev->tx_queue_len = 0;
2292 dev->features |= NETIF_F_LLTX;
2293 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2294 dev->features |= NETIF_F_RXCSUM;
2295 dev->features |= NETIF_F_GSO_SOFTWARE;
2296
2297 dev->vlan_features = dev->features;
2298 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2299 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2300 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2301 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2302 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
2303 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2304
2305 INIT_LIST_HEAD(&vxlan->next);
2306 spin_lock_init(&vxlan->hash_lock);
2307 INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join);
2308 INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave);
2309 INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
2310
2311 init_timer_deferrable(&vxlan->age_timer);
2312 vxlan->age_timer.function = vxlan_cleanup;
2313 vxlan->age_timer.data = (unsigned long) vxlan;
2314
2315 inet_get_local_port_range(dev_net(dev), &low, &high);
2316 vxlan->port_min = low;
2317 vxlan->port_max = high;
2318 vxlan->dst_port = htons(vxlan_port);
2319
2320 vxlan->dev = dev;
2321
2322 for (h = 0; h < FDB_HASH_SIZE; ++h)
2323 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2324 }
2325
2326 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2327 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2328 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2329 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2330 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2331 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2332 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2333 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2334 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2335 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2336 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2337 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2338 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2339 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2340 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2341 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2342 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2343 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2344 };
2345
2346 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2347 {
2348 if (tb[IFLA_ADDRESS]) {
2349 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2350 pr_debug("invalid link address (not ethernet)\n");
2351 return -EINVAL;
2352 }
2353
2354 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2355 pr_debug("invalid all zero ethernet address\n");
2356 return -EADDRNOTAVAIL;
2357 }
2358 }
2359
2360 if (!data)
2361 return -EINVAL;
2362
2363 if (data[IFLA_VXLAN_ID]) {
2364 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2365 if (id >= VXLAN_VID_MASK)
2366 return -ERANGE;
2367 }
2368
2369 if (data[IFLA_VXLAN_PORT_RANGE]) {
2370 const struct ifla_vxlan_port_range *p
2371 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2372
2373 if (ntohs(p->high) < ntohs(p->low)) {
2374 pr_debug("port range %u .. %u not valid\n",
2375 ntohs(p->low), ntohs(p->high));
2376 return -EINVAL;
2377 }
2378 }
2379
2380 return 0;
2381 }
2382
2383 static void vxlan_get_drvinfo(struct net_device *netdev,
2384 struct ethtool_drvinfo *drvinfo)
2385 {
2386 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2387 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2388 }
2389
2390 static const struct ethtool_ops vxlan_ethtool_ops = {
2391 .get_drvinfo = vxlan_get_drvinfo,
2392 .get_link = ethtool_op_get_link,
2393 };
2394
2395 static void vxlan_del_work(struct work_struct *work)
2396 {
2397 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2398
2399 sk_release_kernel(vs->sock->sk);
2400 kfree_rcu(vs, rcu);
2401 }
2402
2403 #if IS_ENABLED(CONFIG_IPV6)
2404 /* Create UDP socket for encapsulation receive. AF_INET6 socket
2405 * could be used for both IPv4 and IPv6 communications, but
2406 * users may set bindv6only=1.
2407 */
2408 static struct socket *create_v6_sock(struct net *net, __be16 port)
2409 {
2410 struct sock *sk;
2411 struct socket *sock;
2412 struct sockaddr_in6 vxlan_addr = {
2413 .sin6_family = AF_INET6,
2414 .sin6_port = port,
2415 };
2416 int rc, val = 1;
2417
2418 rc = sock_create_kern(AF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
2419 if (rc < 0) {
2420 pr_debug("UDPv6 socket create failed\n");
2421 return ERR_PTR(rc);
2422 }
2423
2424 /* Put in proper namespace */
2425 sk = sock->sk;
2426 sk_change_net(sk, net);
2427
2428 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
2429 (char *)&val, sizeof(val));
2430 rc = kernel_bind(sock, (struct sockaddr *)&vxlan_addr,
2431 sizeof(struct sockaddr_in6));
2432 if (rc < 0) {
2433 pr_debug("bind for UDPv6 socket %pI6:%u (%d)\n",
2434 &vxlan_addr.sin6_addr, ntohs(vxlan_addr.sin6_port), rc);
2435 sk_release_kernel(sk);
2436 return ERR_PTR(rc);
2437 }
2438 /* At this point, IPv6 module should have been loaded in
2439 * sock_create_kern().
2440 */
2441 BUG_ON(!ipv6_stub);
2442
2443 /* Disable multicast loopback */
2444 inet_sk(sk)->mc_loop = 0;
2445 return sock;
2446 }
2447
2448 #else
2449
2450 static struct socket *create_v6_sock(struct net *net, __be16 port)
2451 {
2452 return ERR_PTR(-EPFNOSUPPORT);
2453 }
2454 #endif
2455
2456 static struct socket *create_v4_sock(struct net *net, __be16 port)
2457 {
2458 struct sock *sk;
2459 struct socket *sock;
2460 struct sockaddr_in vxlan_addr = {
2461 .sin_family = AF_INET,
2462 .sin_addr.s_addr = htonl(INADDR_ANY),
2463 .sin_port = port,
2464 };
2465 int rc;
2466
2467 /* Create UDP socket for encapsulation receive. */
2468 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
2469 if (rc < 0) {
2470 pr_debug("UDP socket create failed\n");
2471 return ERR_PTR(rc);
2472 }
2473
2474 /* Put in proper namespace */
2475 sk = sock->sk;
2476 sk_change_net(sk, net);
2477
2478 rc = kernel_bind(sock, (struct sockaddr *) &vxlan_addr,
2479 sizeof(vxlan_addr));
2480 if (rc < 0) {
2481 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
2482 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
2483 sk_release_kernel(sk);
2484 return ERR_PTR(rc);
2485 }
2486
2487 /* Disable multicast loopback */
2488 inet_sk(sk)->mc_loop = 0;
2489 return sock;
2490 }
2491
2492 /* Create new listen socket if needed */
2493 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2494 vxlan_rcv_t *rcv, void *data, bool ipv6)
2495 {
2496 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2497 struct vxlan_sock *vs;
2498 struct socket *sock;
2499 struct sock *sk;
2500 unsigned int h;
2501
2502 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2503 if (!vs)
2504 return ERR_PTR(-ENOMEM);
2505
2506 for (h = 0; h < VNI_HASH_SIZE; ++h)
2507 INIT_HLIST_HEAD(&vs->vni_list[h]);
2508
2509 INIT_WORK(&vs->del_work, vxlan_del_work);
2510
2511 if (ipv6)
2512 sock = create_v6_sock(net, port);
2513 else
2514 sock = create_v4_sock(net, port);
2515 if (IS_ERR(sock)) {
2516 kfree(vs);
2517 return ERR_CAST(sock);
2518 }
2519
2520 vs->sock = sock;
2521 sk = sock->sk;
2522 atomic_set(&vs->refcnt, 1);
2523 vs->rcv = rcv;
2524 vs->data = data;
2525 rcu_assign_sk_user_data(vs->sock->sk, vs);
2526
2527 /* Initialize the vxlan udp offloads structure */
2528 vs->udp_offloads.port = port;
2529 vs->udp_offloads.callbacks.gro_receive = vxlan_gro_receive;
2530 vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2531
2532 spin_lock(&vn->sock_lock);
2533 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2534 vxlan_notify_add_rx_port(vs);
2535 spin_unlock(&vn->sock_lock);
2536
2537 /* Mark socket as an encapsulation socket. */
2538 udp_sk(sk)->encap_type = 1;
2539 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
2540 #if IS_ENABLED(CONFIG_IPV6)
2541 if (ipv6)
2542 ipv6_stub->udpv6_encap_enable();
2543 else
2544 #endif
2545 udp_encap_enable();
2546
2547 return vs;
2548 }
2549
2550 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2551 vxlan_rcv_t *rcv, void *data,
2552 bool no_share, bool ipv6)
2553 {
2554 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2555 struct vxlan_sock *vs;
2556
2557 vs = vxlan_socket_create(net, port, rcv, data, ipv6);
2558 if (!IS_ERR(vs))
2559 return vs;
2560
2561 if (no_share) /* Return error if sharing is not allowed. */
2562 return vs;
2563
2564 spin_lock(&vn->sock_lock);
2565 vs = vxlan_find_sock(net, port);
2566 if (vs) {
2567 if (vs->rcv == rcv)
2568 atomic_inc(&vs->refcnt);
2569 else
2570 vs = ERR_PTR(-EBUSY);
2571 }
2572 spin_unlock(&vn->sock_lock);
2573
2574 if (!vs)
2575 vs = ERR_PTR(-EINVAL);
2576
2577 return vs;
2578 }
2579 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2580
2581 /* Scheduled at device creation to bind to a socket */
2582 static void vxlan_sock_work(struct work_struct *work)
2583 {
2584 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2585 struct net *net = vxlan->net;
2586 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2587 __be16 port = vxlan->dst_port;
2588 struct vxlan_sock *nvs;
2589
2590 nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags & VXLAN_F_IPV6);
2591 spin_lock(&vn->sock_lock);
2592 if (!IS_ERR(nvs))
2593 vxlan_vs_add_dev(nvs, vxlan);
2594 spin_unlock(&vn->sock_lock);
2595
2596 dev_put(vxlan->dev);
2597 }
2598
2599 static int vxlan_newlink(struct net *net, struct net_device *dev,
2600 struct nlattr *tb[], struct nlattr *data[])
2601 {
2602 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2603 struct vxlan_dev *vxlan = netdev_priv(dev);
2604 struct vxlan_rdst *dst = &vxlan->default_dst;
2605 __u32 vni;
2606 int err;
2607 bool use_ipv6 = false;
2608
2609 if (!data[IFLA_VXLAN_ID])
2610 return -EINVAL;
2611
2612 vxlan->net = dev_net(dev);
2613
2614 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2615 dst->remote_vni = vni;
2616
2617 /* Unless IPv6 is explicitly requested, assume IPv4 */
2618 dst->remote_ip.sa.sa_family = AF_INET;
2619 if (data[IFLA_VXLAN_GROUP]) {
2620 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2621 } else if (data[IFLA_VXLAN_GROUP6]) {
2622 if (!IS_ENABLED(CONFIG_IPV6))
2623 return -EPFNOSUPPORT;
2624
2625 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2626 sizeof(struct in6_addr));
2627 dst->remote_ip.sa.sa_family = AF_INET6;
2628 use_ipv6 = true;
2629 }
2630
2631 if (data[IFLA_VXLAN_LOCAL]) {
2632 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2633 vxlan->saddr.sa.sa_family = AF_INET;
2634 } else if (data[IFLA_VXLAN_LOCAL6]) {
2635 if (!IS_ENABLED(CONFIG_IPV6))
2636 return -EPFNOSUPPORT;
2637
2638 /* TODO: respect scope id */
2639 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2640 sizeof(struct in6_addr));
2641 vxlan->saddr.sa.sa_family = AF_INET6;
2642 use_ipv6 = true;
2643 }
2644
2645 if (data[IFLA_VXLAN_LINK] &&
2646 (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2647 struct net_device *lowerdev
2648 = __dev_get_by_index(net, dst->remote_ifindex);
2649
2650 if (!lowerdev) {
2651 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2652 return -ENODEV;
2653 }
2654
2655 #if IS_ENABLED(CONFIG_IPV6)
2656 if (use_ipv6) {
2657 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2658 if (idev && idev->cnf.disable_ipv6) {
2659 pr_info("IPv6 is disabled via sysctl\n");
2660 return -EPERM;
2661 }
2662 vxlan->flags |= VXLAN_F_IPV6;
2663 }
2664 #endif
2665
2666 if (!tb[IFLA_MTU])
2667 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2668
2669 /* update header length based on lower device */
2670 dev->hard_header_len = lowerdev->hard_header_len +
2671 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2672 } else if (use_ipv6)
2673 vxlan->flags |= VXLAN_F_IPV6;
2674
2675 if (data[IFLA_VXLAN_TOS])
2676 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2677
2678 if (data[IFLA_VXLAN_TTL])
2679 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2680
2681 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2682 vxlan->flags |= VXLAN_F_LEARN;
2683
2684 if (data[IFLA_VXLAN_AGEING])
2685 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2686 else
2687 vxlan->age_interval = FDB_AGE_DEFAULT;
2688
2689 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2690 vxlan->flags |= VXLAN_F_PROXY;
2691
2692 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2693 vxlan->flags |= VXLAN_F_RSC;
2694
2695 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2696 vxlan->flags |= VXLAN_F_L2MISS;
2697
2698 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2699 vxlan->flags |= VXLAN_F_L3MISS;
2700
2701 if (data[IFLA_VXLAN_LIMIT])
2702 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2703
2704 if (data[IFLA_VXLAN_PORT_RANGE]) {
2705 const struct ifla_vxlan_port_range *p
2706 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2707 vxlan->port_min = ntohs(p->low);
2708 vxlan->port_max = ntohs(p->high);
2709 }
2710
2711 if (data[IFLA_VXLAN_PORT])
2712 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2713
2714 if (vxlan_find_vni(net, vni, vxlan->dst_port)) {
2715 pr_info("duplicate VNI %u\n", vni);
2716 return -EEXIST;
2717 }
2718
2719 dev->ethtool_ops = &vxlan_ethtool_ops;
2720
2721 /* create an fdb entry for a valid default destination */
2722 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2723 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2724 &vxlan->default_dst.remote_ip,
2725 NUD_REACHABLE|NUD_PERMANENT,
2726 NLM_F_EXCL|NLM_F_CREATE,
2727 vxlan->dst_port,
2728 vxlan->default_dst.remote_vni,
2729 vxlan->default_dst.remote_ifindex,
2730 NTF_SELF);
2731 if (err)
2732 return err;
2733 }
2734
2735 err = register_netdevice(dev);
2736 if (err) {
2737 vxlan_fdb_delete_default(vxlan);
2738 return err;
2739 }
2740
2741 list_add(&vxlan->next, &vn->vxlan_list);
2742
2743 return 0;
2744 }
2745
2746 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2747 {
2748 struct vxlan_dev *vxlan = netdev_priv(dev);
2749 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2750
2751 spin_lock(&vn->sock_lock);
2752 if (!hlist_unhashed(&vxlan->hlist))
2753 hlist_del_rcu(&vxlan->hlist);
2754 spin_unlock(&vn->sock_lock);
2755
2756 list_del(&vxlan->next);
2757 unregister_netdevice_queue(dev, head);
2758 }
2759
2760 static size_t vxlan_get_size(const struct net_device *dev)
2761 {
2762
2763 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
2764 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2765 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2766 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2767 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
2768 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
2769 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
2770 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
2771 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
2772 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
2773 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
2774 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2775 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2776 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2777 nla_total_size(sizeof(__be16))+ /* IFLA_VXLAN_PORT */
2778 0;
2779 }
2780
2781 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2782 {
2783 const struct vxlan_dev *vxlan = netdev_priv(dev);
2784 const struct vxlan_rdst *dst = &vxlan->default_dst;
2785 struct ifla_vxlan_port_range ports = {
2786 .low = htons(vxlan->port_min),
2787 .high = htons(vxlan->port_max),
2788 };
2789
2790 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2791 goto nla_put_failure;
2792
2793 if (!vxlan_addr_any(&dst->remote_ip)) {
2794 if (dst->remote_ip.sa.sa_family == AF_INET) {
2795 if (nla_put_be32(skb, IFLA_VXLAN_GROUP,
2796 dst->remote_ip.sin.sin_addr.s_addr))
2797 goto nla_put_failure;
2798 #if IS_ENABLED(CONFIG_IPV6)
2799 } else {
2800 if (nla_put(skb, IFLA_VXLAN_GROUP6, sizeof(struct in6_addr),
2801 &dst->remote_ip.sin6.sin6_addr))
2802 goto nla_put_failure;
2803 #endif
2804 }
2805 }
2806
2807 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2808 goto nla_put_failure;
2809
2810 if (!vxlan_addr_any(&vxlan->saddr)) {
2811 if (vxlan->saddr.sa.sa_family == AF_INET) {
2812 if (nla_put_be32(skb, IFLA_VXLAN_LOCAL,
2813 vxlan->saddr.sin.sin_addr.s_addr))
2814 goto nla_put_failure;
2815 #if IS_ENABLED(CONFIG_IPV6)
2816 } else {
2817 if (nla_put(skb, IFLA_VXLAN_LOCAL6, sizeof(struct in6_addr),
2818 &vxlan->saddr.sin6.sin6_addr))
2819 goto nla_put_failure;
2820 #endif
2821 }
2822 }
2823
2824 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2825 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2826 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2827 !!(vxlan->flags & VXLAN_F_LEARN)) ||
2828 nla_put_u8(skb, IFLA_VXLAN_PROXY,
2829 !!(vxlan->flags & VXLAN_F_PROXY)) ||
2830 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2831 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2832 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2833 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2834 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2835 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2836 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2837 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port))
2838 goto nla_put_failure;
2839
2840 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2841 goto nla_put_failure;
2842
2843 return 0;
2844
2845 nla_put_failure:
2846 return -EMSGSIZE;
2847 }
2848
2849 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2850 .kind = "vxlan",
2851 .maxtype = IFLA_VXLAN_MAX,
2852 .policy = vxlan_policy,
2853 .priv_size = sizeof(struct vxlan_dev),
2854 .setup = vxlan_setup,
2855 .validate = vxlan_validate,
2856 .newlink = vxlan_newlink,
2857 .dellink = vxlan_dellink,
2858 .get_size = vxlan_get_size,
2859 .fill_info = vxlan_fill_info,
2860 };
2861
2862 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
2863 struct net_device *dev)
2864 {
2865 struct vxlan_dev *vxlan, *next;
2866 LIST_HEAD(list_kill);
2867
2868 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2869 struct vxlan_rdst *dst = &vxlan->default_dst;
2870
2871 /* In case we created vxlan device with carrier
2872 * and we loose the carrier due to module unload
2873 * we also need to remove vxlan device. In other
2874 * cases, it's not necessary and remote_ifindex
2875 * is 0 here, so no matches.
2876 */
2877 if (dst->remote_ifindex == dev->ifindex)
2878 vxlan_dellink(vxlan->dev, &list_kill);
2879 }
2880
2881 unregister_netdevice_many(&list_kill);
2882 }
2883
2884 static int vxlan_lowerdev_event(struct notifier_block *unused,
2885 unsigned long event, void *ptr)
2886 {
2887 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2888 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2889
2890 if (event == NETDEV_UNREGISTER)
2891 vxlan_handle_lowerdev_unregister(vn, dev);
2892
2893 return NOTIFY_DONE;
2894 }
2895
2896 static struct notifier_block vxlan_notifier_block __read_mostly = {
2897 .notifier_call = vxlan_lowerdev_event,
2898 };
2899
2900 static __net_init int vxlan_init_net(struct net *net)
2901 {
2902 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2903 unsigned int h;
2904
2905 INIT_LIST_HEAD(&vn->vxlan_list);
2906 spin_lock_init(&vn->sock_lock);
2907
2908 for (h = 0; h < PORT_HASH_SIZE; ++h)
2909 INIT_HLIST_HEAD(&vn->sock_list[h]);
2910
2911 return 0;
2912 }
2913
2914 static void __net_exit vxlan_exit_net(struct net *net)
2915 {
2916 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2917 struct vxlan_dev *vxlan, *next;
2918 struct net_device *dev, *aux;
2919 LIST_HEAD(list);
2920
2921 rtnl_lock();
2922 for_each_netdev_safe(net, dev, aux)
2923 if (dev->rtnl_link_ops == &vxlan_link_ops)
2924 unregister_netdevice_queue(dev, &list);
2925
2926 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2927 /* If vxlan->dev is in the same netns, it has already been added
2928 * to the list by the previous loop.
2929 */
2930 if (!net_eq(dev_net(vxlan->dev), net))
2931 unregister_netdevice_queue(dev, &list);
2932 }
2933
2934 unregister_netdevice_many(&list);
2935 rtnl_unlock();
2936 }
2937
2938 static struct pernet_operations vxlan_net_ops = {
2939 .init = vxlan_init_net,
2940 .exit = vxlan_exit_net,
2941 .id = &vxlan_net_id,
2942 .size = sizeof(struct vxlan_net),
2943 };
2944
2945 static int __init vxlan_init_module(void)
2946 {
2947 int rc;
2948
2949 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
2950 if (!vxlan_wq)
2951 return -ENOMEM;
2952
2953 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
2954
2955 rc = register_pernet_subsys(&vxlan_net_ops);
2956 if (rc)
2957 goto out1;
2958
2959 rc = register_netdevice_notifier(&vxlan_notifier_block);
2960 if (rc)
2961 goto out2;
2962
2963 rc = rtnl_link_register(&vxlan_link_ops);
2964 if (rc)
2965 goto out3;
2966
2967 return 0;
2968 out3:
2969 unregister_netdevice_notifier(&vxlan_notifier_block);
2970 out2:
2971 unregister_pernet_subsys(&vxlan_net_ops);
2972 out1:
2973 destroy_workqueue(vxlan_wq);
2974 return rc;
2975 }
2976 late_initcall(vxlan_init_module);
2977
2978 static void __exit vxlan_cleanup_module(void)
2979 {
2980 rtnl_link_unregister(&vxlan_link_ops);
2981 unregister_netdevice_notifier(&vxlan_notifier_block);
2982 destroy_workqueue(vxlan_wq);
2983 unregister_pernet_subsys(&vxlan_net_ops);
2984 /* rcu_barrier() is called by netns */
2985 }
2986 module_exit(vxlan_cleanup_module);
2987
2988 MODULE_LICENSE("GPL");
2989 MODULE_VERSION(VXLAN_VERSION);
2990 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
2991 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
2992 MODULE_ALIAS_RTNL_LINK("vxlan");
Linux kernel - Deliverables
This page took 0.121164 seconds and 5 git commands to generate.