projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[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 vxlan_rdst default_dst; /* default destination */
131 union vxlan_addr saddr; /* source address */
132 __be16 dst_port;
133 __u16 port_min; /* source port range */
134 __u16 port_max;
135 __u8 tos; /* TOS override */
136 __u8 ttl;
137 u32 flags; /* VXLAN_F_* below */
138
139 struct work_struct sock_work;
140 struct work_struct igmp_join;
141 struct work_struct igmp_leave;
142
143 unsigned long age_interval;
144 struct timer_list age_timer;
145 spinlock_t hash_lock;
146 unsigned int addrcnt;
147 unsigned int addrmax;
148
149 struct hlist_head fdb_head[FDB_HASH_SIZE];
150 };
151
152 #define VXLAN_F_LEARN 0x01
153 #define VXLAN_F_PROXY 0x02
154 #define VXLAN_F_RSC 0x04
155 #define VXLAN_F_L2MISS 0x08
156 #define VXLAN_F_L3MISS 0x10
157 #define VXLAN_F_IPV6 0x20 /* internal flag */
158
159 /* salt for hash table */
160 static u32 vxlan_salt __read_mostly;
161 static struct workqueue_struct *vxlan_wq;
162
163 static void vxlan_sock_work(struct work_struct *work);
164
165 #if IS_ENABLED(CONFIG_IPV6)
166 static inline
167 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
168 {
169 if (a->sa.sa_family != b->sa.sa_family)
170 return false;
171 if (a->sa.sa_family == AF_INET6)
172 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
173 else
174 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
175 }
176
177 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
178 {
179 if (ipa->sa.sa_family == AF_INET6)
180 return ipv6_addr_any(&ipa->sin6.sin6_addr);
181 else
182 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
183 }
184
185 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
186 {
187 if (ipa->sa.sa_family == AF_INET6)
188 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
189 else
190 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
191 }
192
193 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
194 {
195 if (nla_len(nla) >= sizeof(struct in6_addr)) {
196 nla_memcpy(&ip->sin6.sin6_addr, nla, sizeof(struct in6_addr));
197 ip->sa.sa_family = AF_INET6;
198 return 0;
199 } else if (nla_len(nla) >= sizeof(__be32)) {
200 ip->sin.sin_addr.s_addr = nla_get_be32(nla);
201 ip->sa.sa_family = AF_INET;
202 return 0;
203 } else {
204 return -EAFNOSUPPORT;
205 }
206 }
207
208 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
209 const union vxlan_addr *ip)
210 {
211 if (ip->sa.sa_family == AF_INET6)
212 return nla_put(skb, attr, sizeof(struct in6_addr), &ip->sin6.sin6_addr);
213 else
214 return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
215 }
216
217 #else /* !CONFIG_IPV6 */
218
219 static inline
220 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
221 {
222 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
223 }
224
225 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
226 {
227 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
228 }
229
230 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
231 {
232 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
233 }
234
235 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
236 {
237 if (nla_len(nla) >= sizeof(struct in6_addr)) {
238 return -EAFNOSUPPORT;
239 } else if (nla_len(nla) >= sizeof(__be32)) {
240 ip->sin.sin_addr.s_addr = nla_get_be32(nla);
241 ip->sa.sa_family = AF_INET;
242 return 0;
243 } else {
244 return -EAFNOSUPPORT;
245 }
246 }
247
248 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
249 const union vxlan_addr *ip)
250 {
251 return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
252 }
253 #endif
254
255 /* Virtual Network hash table head */
256 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
257 {
258 return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
259 }
260
261 /* Socket hash table head */
262 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
263 {
264 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
265
266 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
267 }
268
269 /* First remote destination for a forwarding entry.
270 * Guaranteed to be non-NULL because remotes are never deleted.
271 */
272 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
273 {
274 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
275 }
276
277 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
278 {
279 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
280 }
281
282 /* Find VXLAN socket based on network namespace and UDP port */
283 static struct vxlan_sock *vxlan_find_sock(struct net *net, __be16 port)
284 {
285 struct vxlan_sock *vs;
286
287 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
288 if (inet_sk(vs->sock->sk)->inet_sport == port)
289 return vs;
290 }
291 return NULL;
292 }
293
294 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
295 {
296 struct vxlan_dev *vxlan;
297
298 hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
299 if (vxlan->default_dst.remote_vni == id)
300 return vxlan;
301 }
302
303 return NULL;
304 }
305
306 /* Look up VNI in a per net namespace table */
307 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id, __be16 port)
308 {
309 struct vxlan_sock *vs;
310
311 vs = vxlan_find_sock(net, port);
312 if (!vs)
313 return NULL;
314
315 return vxlan_vs_find_vni(vs, id);
316 }
317
318 /* Fill in neighbour message in skbuff. */
319 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
320 const struct vxlan_fdb *fdb,
321 u32 portid, u32 seq, int type, unsigned int flags,
322 const struct vxlan_rdst *rdst)
323 {
324 unsigned long now = jiffies;
325 struct nda_cacheinfo ci;
326 struct nlmsghdr *nlh;
327 struct ndmsg *ndm;
328 bool send_ip, send_eth;
329
330 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
331 if (nlh == NULL)
332 return -EMSGSIZE;
333
334 ndm = nlmsg_data(nlh);
335 memset(ndm, 0, sizeof(*ndm));
336
337 send_eth = send_ip = true;
338
339 if (type == RTM_GETNEIGH) {
340 ndm->ndm_family = AF_INET;
341 send_ip = !vxlan_addr_any(&rdst->remote_ip);
342 send_eth = !is_zero_ether_addr(fdb->eth_addr);
343 } else
344 ndm->ndm_family = AF_BRIDGE;
345 ndm->ndm_state = fdb->state;
346 ndm->ndm_ifindex = vxlan->dev->ifindex;
347 ndm->ndm_flags = fdb->flags;
348 ndm->ndm_type = NDA_DST;
349
350 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
351 goto nla_put_failure;
352
353 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
354 goto nla_put_failure;
355
356 if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
357 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
358 goto nla_put_failure;
359 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
360 nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
361 goto nla_put_failure;
362 if (rdst->remote_ifindex &&
363 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
364 goto nla_put_failure;
365
366 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
367 ci.ndm_confirmed = 0;
368 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
369 ci.ndm_refcnt = 0;
370
371 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
372 goto nla_put_failure;
373
374 return nlmsg_end(skb, nlh);
375
376 nla_put_failure:
377 nlmsg_cancel(skb, nlh);
378 return -EMSGSIZE;
379 }
380
381 static inline size_t vxlan_nlmsg_size(void)
382 {
383 return NLMSG_ALIGN(sizeof(struct ndmsg))
384 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
385 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
386 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
387 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
388 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
389 + nla_total_size(sizeof(struct nda_cacheinfo));
390 }
391
392 static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
393 struct vxlan_fdb *fdb, int type)
394 {
395 struct net *net = dev_net(vxlan->dev);
396 struct sk_buff *skb;
397 int err = -ENOBUFS;
398
399 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
400 if (skb == NULL)
401 goto errout;
402
403 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0,
404 first_remote_rtnl(fdb));
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 INIT_LIST_HEAD(&f.remotes);
431 list_add_rcu(&remote.list, &f.remotes);
432
433 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
434 }
435
436 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
437 {
438 struct vxlan_fdb f = {
439 .state = NUD_STALE,
440 };
441
442 INIT_LIST_HEAD(&f.remotes);
443 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
444
445 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
446 }
447
448 /* Hash Ethernet address */
449 static u32 eth_hash(const unsigned char *addr)
450 {
451 u64 value = get_unaligned((u64 *)addr);
452
453 /* only want 6 bytes */
454 #ifdef __BIG_ENDIAN
455 value >>= 16;
456 #else
457 value <<= 16;
458 #endif
459 return hash_64(value, FDB_HASH_BITS);
460 }
461
462 /* Hash chain to use given mac address */
463 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
464 const u8 *mac)
465 {
466 return &vxlan->fdb_head[eth_hash(mac)];
467 }
468
469 /* Look up Ethernet address in forwarding table */
470 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
471 const u8 *mac)
472 {
473 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
474 struct vxlan_fdb *f;
475
476 hlist_for_each_entry_rcu(f, head, hlist) {
477 if (ether_addr_equal(mac, f->eth_addr))
478 return f;
479 }
480
481 return NULL;
482 }
483
484 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
485 const u8 *mac)
486 {
487 struct vxlan_fdb *f;
488
489 f = __vxlan_find_mac(vxlan, mac);
490 if (f)
491 f->used = jiffies;
492
493 return f;
494 }
495
496 /* caller should hold vxlan->hash_lock */
497 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
498 union vxlan_addr *ip, __be16 port,
499 __u32 vni, __u32 ifindex)
500 {
501 struct vxlan_rdst *rd;
502
503 list_for_each_entry(rd, &f->remotes, list) {
504 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
505 rd->remote_port == port &&
506 rd->remote_vni == vni &&
507 rd->remote_ifindex == ifindex)
508 return rd;
509 }
510
511 return NULL;
512 }
513
514 /* Replace destination of unicast mac */
515 static int vxlan_fdb_replace(struct vxlan_fdb *f,
516 union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
517 {
518 struct vxlan_rdst *rd;
519
520 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
521 if (rd)
522 return 0;
523
524 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
525 if (!rd)
526 return 0;
527 rd->remote_ip = *ip;
528 rd->remote_port = port;
529 rd->remote_vni = vni;
530 rd->remote_ifindex = ifindex;
531 return 1;
532 }
533
534 /* Add/update destinations for multicast */
535 static int vxlan_fdb_append(struct vxlan_fdb *f,
536 union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
537 {
538 struct vxlan_rdst *rd;
539
540 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
541 if (rd)
542 return 0;
543
544 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
545 if (rd == NULL)
546 return -ENOBUFS;
547 rd->remote_ip = *ip;
548 rd->remote_port = port;
549 rd->remote_vni = vni;
550 rd->remote_ifindex = ifindex;
551
552 list_add_tail_rcu(&rd->list, &f->remotes);
553
554 return 1;
555 }
556
557 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head, struct sk_buff *skb)
558 {
559 struct sk_buff *p, **pp = NULL;
560 struct vxlanhdr *vh, *vh2;
561 struct ethhdr *eh, *eh2;
562 unsigned int hlen, off_vx, off_eth;
563 const struct packet_offload *ptype;
564 __be16 type;
565 int flush = 1;
566
567 off_vx = skb_gro_offset(skb);
568 hlen = off_vx + sizeof(*vh);
569 vh = skb_gro_header_fast(skb, off_vx);
570 if (skb_gro_header_hard(skb, hlen)) {
571 vh = skb_gro_header_slow(skb, hlen, off_vx);
572 if (unlikely(!vh))
573 goto out;
574 }
575 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
576
577 off_eth = skb_gro_offset(skb);
578 hlen = off_eth + sizeof(*eh);
579 eh = skb_gro_header_fast(skb, off_eth);
580 if (skb_gro_header_hard(skb, hlen)) {
581 eh = skb_gro_header_slow(skb, hlen, off_eth);
582 if (unlikely(!eh))
583 goto out;
584 }
585
586 flush = 0;
587
588 for (p = *head; p; p = p->next) {
589 if (!NAPI_GRO_CB(p)->same_flow)
590 continue;
591
592 vh2 = (struct vxlanhdr *)(p->data + off_vx);
593 eh2 = (struct ethhdr *)(p->data + off_eth);
594 if (vh->vx_vni != vh2->vx_vni || compare_ether_header(eh, eh2)) {
595 NAPI_GRO_CB(p)->same_flow = 0;
596 continue;
597 }
598 }
599
600 type = eh->h_proto;
601
602 rcu_read_lock();
603 ptype = gro_find_receive_by_type(type);
604 if (ptype == NULL) {
605 flush = 1;
606 goto out_unlock;
607 }
608
609 skb_gro_pull(skb, sizeof(*eh)); /* pull inner eth header */
610 pp = ptype->callbacks.gro_receive(head, skb);
611
612 out_unlock:
613 rcu_read_unlock();
614 out:
615 NAPI_GRO_CB(skb)->flush |= flush;
616
617 return pp;
618 }
619
620 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff)
621 {
622 struct ethhdr *eh;
623 struct packet_offload *ptype;
624 __be16 type;
625 int vxlan_len = sizeof(struct vxlanhdr) + sizeof(struct ethhdr);
626 int err = -ENOSYS;
627
628 eh = (struct ethhdr *)(skb->data + nhoff + sizeof(struct vxlanhdr));
629 type = eh->h_proto;
630
631 rcu_read_lock();
632 ptype = gro_find_complete_by_type(type);
633 if (ptype != NULL)
634 err = ptype->callbacks.gro_complete(skb, nhoff + vxlan_len);
635
636 rcu_read_unlock();
637 return err;
638 }
639
640 /* Notify netdevs that UDP port started listening */
641 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
642 {
643 struct net_device *dev;
644 struct sock *sk = vs->sock->sk;
645 struct net *net = sock_net(sk);
646 sa_family_t sa_family = sk->sk_family;
647 __be16 port = inet_sk(sk)->inet_sport;
648 int err;
649
650 if (sa_family == AF_INET) {
651 err = udp_add_offload(&vs->udp_offloads);
652 if (err)
653 pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
654 }
655
656 rcu_read_lock();
657 for_each_netdev_rcu(net, dev) {
658 if (dev->netdev_ops->ndo_add_vxlan_port)
659 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
660 port);
661 }
662 rcu_read_unlock();
663 }
664
665 /* Notify netdevs that UDP port is no more listening */
666 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
667 {
668 struct net_device *dev;
669 struct sock *sk = vs->sock->sk;
670 struct net *net = sock_net(sk);
671 sa_family_t sa_family = sk->sk_family;
672 __be16 port = inet_sk(sk)->inet_sport;
673
674 rcu_read_lock();
675 for_each_netdev_rcu(net, dev) {
676 if (dev->netdev_ops->ndo_del_vxlan_port)
677 dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
678 port);
679 }
680 rcu_read_unlock();
681
682 if (sa_family == AF_INET)
683 udp_del_offload(&vs->udp_offloads);
684 }
685
686 /* Add new entry to forwarding table -- assumes lock held */
687 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
688 const u8 *mac, union vxlan_addr *ip,
689 __u16 state, __u16 flags,
690 __be16 port, __u32 vni, __u32 ifindex,
691 __u8 ndm_flags)
692 {
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
731 if (rc < 0)
732 return rc;
733 notify |= rc;
734 }
735 } else {
736 if (!(flags & NLM_F_CREATE))
737 return -ENOENT;
738
739 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
740 return -ENOSPC;
741
742 /* Disallow replace to add a multicast entry */
743 if ((flags & NLM_F_REPLACE) &&
744 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
745 return -EOPNOTSUPP;
746
747 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
748 f = kmalloc(sizeof(*f), GFP_ATOMIC);
749 if (!f)
750 return -ENOMEM;
751
752 notify = 1;
753 f->state = state;
754 f->flags = ndm_flags;
755 f->updated = f->used = jiffies;
756 INIT_LIST_HEAD(&f->remotes);
757 memcpy(f->eth_addr, mac, ETH_ALEN);
758
759 vxlan_fdb_append(f, ip, port, vni, ifindex);
760
761 ++vxlan->addrcnt;
762 hlist_add_head_rcu(&f->hlist,
763 vxlan_fdb_head(vxlan, mac));
764 }
765
766 if (notify)
767 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
768
769 return 0;
770 }
771
772 static void vxlan_fdb_free(struct rcu_head *head)
773 {
774 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
775 struct vxlan_rdst *rd, *nd;
776
777 list_for_each_entry_safe(rd, nd, &f->remotes, list)
778 kfree(rd);
779 kfree(f);
780 }
781
782 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
783 {
784 netdev_dbg(vxlan->dev,
785 "delete %pM\n", f->eth_addr);
786
787 --vxlan->addrcnt;
788 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
789
790 hlist_del_rcu(&f->hlist);
791 call_rcu(&f->rcu, vxlan_fdb_free);
792 }
793
794 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
795 union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
796 {
797 struct net *net = dev_net(vxlan->dev);
798 int err;
799
800 if (tb[NDA_DST]) {
801 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
802 if (err)
803 return err;
804 } else {
805 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
806 if (remote->sa.sa_family == AF_INET) {
807 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
808 ip->sa.sa_family = AF_INET;
809 #if IS_ENABLED(CONFIG_IPV6)
810 } else {
811 ip->sin6.sin6_addr = in6addr_any;
812 ip->sa.sa_family = AF_INET6;
813 #endif
814 }
815 }
816
817 if (tb[NDA_PORT]) {
818 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
819 return -EINVAL;
820 *port = nla_get_be16(tb[NDA_PORT]);
821 } else {
822 *port = vxlan->dst_port;
823 }
824
825 if (tb[NDA_VNI]) {
826 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
827 return -EINVAL;
828 *vni = nla_get_u32(tb[NDA_VNI]);
829 } else {
830 *vni = vxlan->default_dst.remote_vni;
831 }
832
833 if (tb[NDA_IFINDEX]) {
834 struct net_device *tdev;
835
836 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
837 return -EINVAL;
838 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
839 tdev = __dev_get_by_index(net, *ifindex);
840 if (!tdev)
841 return -EADDRNOTAVAIL;
842 } else {
843 *ifindex = 0;
844 }
845
846 return 0;
847 }
848
849 /* Add static entry (via netlink) */
850 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
851 struct net_device *dev,
852 const unsigned char *addr, u16 flags)
853 {
854 struct vxlan_dev *vxlan = netdev_priv(dev);
855 /* struct net *net = dev_net(vxlan->dev); */
856 union vxlan_addr ip;
857 __be16 port;
858 u32 vni, ifindex;
859 int err;
860
861 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
862 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
863 ndm->ndm_state);
864 return -EINVAL;
865 }
866
867 if (tb[NDA_DST] == NULL)
868 return -EINVAL;
869
870 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
871 if (err)
872 return err;
873
874 spin_lock_bh(&vxlan->hash_lock);
875 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
876 port, vni, ifindex, ndm->ndm_flags);
877 spin_unlock_bh(&vxlan->hash_lock);
878
879 return err;
880 }
881
882 /* Delete entry (via netlink) */
883 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
884 struct net_device *dev,
885 const unsigned char *addr)
886 {
887 struct vxlan_dev *vxlan = netdev_priv(dev);
888 struct vxlan_fdb *f;
889 struct vxlan_rdst *rd = NULL;
890 union vxlan_addr ip;
891 __be16 port;
892 u32 vni, ifindex;
893 int err;
894
895 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
896 if (err)
897 return err;
898
899 err = -ENOENT;
900
901 spin_lock_bh(&vxlan->hash_lock);
902 f = vxlan_find_mac(vxlan, addr);
903 if (!f)
904 goto out;
905
906 if (!vxlan_addr_any(&ip)) {
907 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
908 if (!rd)
909 goto out;
910 }
911
912 err = 0;
913
914 /* remove a destination if it's not the only one on the list,
915 * otherwise destroy the fdb entry
916 */
917 if (rd && !list_is_singular(&f->remotes)) {
918 list_del_rcu(&rd->list);
919 kfree_rcu(rd, rcu);
920 goto out;
921 }
922
923 vxlan_fdb_destroy(vxlan, f);
924
925 out:
926 spin_unlock_bh(&vxlan->hash_lock);
927
928 return err;
929 }
930
931 /* Dump forwarding table */
932 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
933 struct net_device *dev, int idx)
934 {
935 struct vxlan_dev *vxlan = netdev_priv(dev);
936 unsigned int h;
937
938 for (h = 0; h < FDB_HASH_SIZE; ++h) {
939 struct vxlan_fdb *f;
940 int err;
941
942 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
943 struct vxlan_rdst *rd;
944
945 if (idx < cb->args[0])
946 goto skip;
947
948 list_for_each_entry_rcu(rd, &f->remotes, list) {
949 err = vxlan_fdb_info(skb, vxlan, f,
950 NETLINK_CB(cb->skb).portid,
951 cb->nlh->nlmsg_seq,
952 RTM_NEWNEIGH,
953 NLM_F_MULTI, rd);
954 if (err < 0)
955 goto out;
956 }
957 skip:
958 ++idx;
959 }
960 }
961 out:
962 return idx;
963 }
964
965 /* Watch incoming packets to learn mapping between Ethernet address
966 * and Tunnel endpoint.
967 * Return true if packet is bogus and should be droppped.
968 */
969 static bool vxlan_snoop(struct net_device *dev,
970 union vxlan_addr *src_ip, const u8 *src_mac)
971 {
972 struct vxlan_dev *vxlan = netdev_priv(dev);
973 struct vxlan_fdb *f;
974
975 f = vxlan_find_mac(vxlan, src_mac);
976 if (likely(f)) {
977 struct vxlan_rdst *rdst = first_remote_rcu(f);
978
979 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
980 return false;
981
982 /* Don't migrate static entries, drop packets */
983 if (f->state & NUD_NOARP)
984 return true;
985
986 if (net_ratelimit())
987 netdev_info(dev,
988 "%pM migrated from %pIS to %pIS\n",
989 src_mac, &rdst->remote_ip, &src_ip);
990
991 rdst->remote_ip = *src_ip;
992 f->updated = jiffies;
993 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
994 } else {
995 /* learned new entry */
996 spin_lock(&vxlan->hash_lock);
997
998 /* close off race between vxlan_flush and incoming packets */
999 if (netif_running(dev))
1000 vxlan_fdb_create(vxlan, src_mac, src_ip,
1001 NUD_REACHABLE,
1002 NLM_F_EXCL|NLM_F_CREATE,
1003 vxlan->dst_port,
1004 vxlan->default_dst.remote_vni,
1005 0, NTF_SELF);
1006 spin_unlock(&vxlan->hash_lock);
1007 }
1008
1009 return false;
1010 }
1011
1012 /* See if multicast group is already in use by other ID */
1013 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1014 {
1015 struct vxlan_dev *vxlan;
1016
1017 /* The vxlan_sock is only used by dev, leaving group has
1018 * no effect on other vxlan devices.
1019 */
1020 if (atomic_read(&dev->vn_sock->refcnt) == 1)
1021 return false;
1022
1023 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1024 if (!netif_running(vxlan->dev) || vxlan == dev)
1025 continue;
1026
1027 if (vxlan->vn_sock != dev->vn_sock)
1028 continue;
1029
1030 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1031 &dev->default_dst.remote_ip))
1032 continue;
1033
1034 if (vxlan->default_dst.remote_ifindex !=
1035 dev->default_dst.remote_ifindex)
1036 continue;
1037
1038 return true;
1039 }
1040
1041 return false;
1042 }
1043
1044 static void vxlan_sock_hold(struct vxlan_sock *vs)
1045 {
1046 atomic_inc(&vs->refcnt);
1047 }
1048
1049 void vxlan_sock_release(struct vxlan_sock *vs)
1050 {
1051 struct sock *sk = vs->sock->sk;
1052 struct net *net = sock_net(sk);
1053 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1054
1055 if (!atomic_dec_and_test(&vs->refcnt))
1056 return;
1057
1058 spin_lock(&vn->sock_lock);
1059 hlist_del_rcu(&vs->hlist);
1060 rcu_assign_sk_user_data(vs->sock->sk, NULL);
1061 vxlan_notify_del_rx_port(vs);
1062 spin_unlock(&vn->sock_lock);
1063
1064 queue_work(vxlan_wq, &vs->del_work);
1065 }
1066 EXPORT_SYMBOL_GPL(vxlan_sock_release);
1067
1068 /* Callback to update multicast group membership when first VNI on
1069 * multicast asddress is brought up
1070 * Done as workqueue because ip_mc_join_group acquires RTNL.
1071 */
1072 static void vxlan_igmp_join(struct work_struct *work)
1073 {
1074 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join);
1075 struct vxlan_sock *vs = vxlan->vn_sock;
1076 struct sock *sk = vs->sock->sk;
1077 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1078 int ifindex = vxlan->default_dst.remote_ifindex;
1079
1080 lock_sock(sk);
1081 if (ip->sa.sa_family == AF_INET) {
1082 struct ip_mreqn mreq = {
1083 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1084 .imr_ifindex = ifindex,
1085 };
1086
1087 ip_mc_join_group(sk, &mreq);
1088 #if IS_ENABLED(CONFIG_IPV6)
1089 } else {
1090 ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1091 &ip->sin6.sin6_addr);
1092 #endif
1093 }
1094 release_sock(sk);
1095
1096 vxlan_sock_release(vs);
1097 dev_put(vxlan->dev);
1098 }
1099
1100 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1101 static void vxlan_igmp_leave(struct work_struct *work)
1102 {
1103 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave);
1104 struct vxlan_sock *vs = vxlan->vn_sock;
1105 struct sock *sk = vs->sock->sk;
1106 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1107 int ifindex = vxlan->default_dst.remote_ifindex;
1108
1109 lock_sock(sk);
1110 if (ip->sa.sa_family == AF_INET) {
1111 struct ip_mreqn mreq = {
1112 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1113 .imr_ifindex = ifindex,
1114 };
1115
1116 ip_mc_leave_group(sk, &mreq);
1117 #if IS_ENABLED(CONFIG_IPV6)
1118 } else {
1119 ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1120 &ip->sin6.sin6_addr);
1121 #endif
1122 }
1123
1124 release_sock(sk);
1125
1126 vxlan_sock_release(vs);
1127 dev_put(vxlan->dev);
1128 }
1129
1130 /* Callback from net/ipv4/udp.c to receive packets */
1131 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1132 {
1133 struct vxlan_sock *vs;
1134 struct vxlanhdr *vxh;
1135
1136 /* Need Vxlan and inner Ethernet header to be present */
1137 if (!pskb_may_pull(skb, VXLAN_HLEN))
1138 goto error;
1139
1140 /* Return packets with reserved bits set */
1141 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1142 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
1143 (vxh->vx_vni & htonl(0xff))) {
1144 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1145 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1146 goto error;
1147 }
1148
1149 if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1150 goto drop;
1151
1152 vs = rcu_dereference_sk_user_data(sk);
1153 if (!vs)
1154 goto drop;
1155
1156 /* If the NIC driver gave us an encapsulated packet
1157 * with the encapsulation mark, the device checksummed it
1158 * for us. Otherwise force the upper layers to verify it.
1159 */
1160 if ((skb->ip_summed != CHECKSUM_UNNECESSARY && skb->ip_summed != CHECKSUM_PARTIAL) ||
1161 !skb->encapsulation)
1162 skb->ip_summed = CHECKSUM_NONE;
1163
1164 skb->encapsulation = 0;
1165
1166 vs->rcv(vs, skb, vxh->vx_vni);
1167 return 0;
1168
1169 drop:
1170 /* Consume bad packet */
1171 kfree_skb(skb);
1172 return 0;
1173
1174 error:
1175 /* Return non vxlan pkt */
1176 return 1;
1177 }
1178
1179 static void vxlan_rcv(struct vxlan_sock *vs,
1180 struct sk_buff *skb, __be32 vx_vni)
1181 {
1182 struct iphdr *oip = NULL;
1183 struct ipv6hdr *oip6 = NULL;
1184 struct vxlan_dev *vxlan;
1185 struct pcpu_sw_netstats *stats;
1186 union vxlan_addr saddr;
1187 __u32 vni;
1188 int err = 0;
1189 union vxlan_addr *remote_ip;
1190
1191 vni = ntohl(vx_vni) >> 8;
1192 /* Is this VNI defined? */
1193 vxlan = vxlan_vs_find_vni(vs, vni);
1194 if (!vxlan)
1195 goto drop;
1196
1197 remote_ip = &vxlan->default_dst.remote_ip;
1198 skb_reset_mac_header(skb);
1199 skb->protocol = eth_type_trans(skb, vxlan->dev);
1200
1201 /* Ignore packet loops (and multicast echo) */
1202 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1203 goto drop;
1204
1205 /* Re-examine inner Ethernet packet */
1206 if (remote_ip->sa.sa_family == AF_INET) {
1207 oip = ip_hdr(skb);
1208 saddr.sin.sin_addr.s_addr = oip->saddr;
1209 saddr.sa.sa_family = AF_INET;
1210 #if IS_ENABLED(CONFIG_IPV6)
1211 } else {
1212 oip6 = ipv6_hdr(skb);
1213 saddr.sin6.sin6_addr = oip6->saddr;
1214 saddr.sa.sa_family = AF_INET6;
1215 #endif
1216 }
1217
1218 if ((vxlan->flags & VXLAN_F_LEARN) &&
1219 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1220 goto drop;
1221
1222 skb_reset_network_header(skb);
1223
1224 if (oip6)
1225 err = IP6_ECN_decapsulate(oip6, skb);
1226 if (oip)
1227 err = IP_ECN_decapsulate(oip, skb);
1228
1229 if (unlikely(err)) {
1230 if (log_ecn_error) {
1231 if (oip6)
1232 net_info_ratelimited("non-ECT from %pI6\n",
1233 &oip6->saddr);
1234 if (oip)
1235 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1236 &oip->saddr, oip->tos);
1237 }
1238 if (err > 1) {
1239 ++vxlan->dev->stats.rx_frame_errors;
1240 ++vxlan->dev->stats.rx_errors;
1241 goto drop;
1242 }
1243 }
1244
1245 stats = this_cpu_ptr(vxlan->dev->tstats);
1246 u64_stats_update_begin(&stats->syncp);
1247 stats->rx_packets++;
1248 stats->rx_bytes += skb->len;
1249 u64_stats_update_end(&stats->syncp);
1250
1251 netif_rx(skb);
1252
1253 return;
1254 drop:
1255 /* Consume bad packet */
1256 kfree_skb(skb);
1257 }
1258
1259 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1260 {
1261 struct vxlan_dev *vxlan = netdev_priv(dev);
1262 struct arphdr *parp;
1263 u8 *arpptr, *sha;
1264 __be32 sip, tip;
1265 struct neighbour *n;
1266
1267 if (dev->flags & IFF_NOARP)
1268 goto out;
1269
1270 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1271 dev->stats.tx_dropped++;
1272 goto out;
1273 }
1274 parp = arp_hdr(skb);
1275
1276 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1277 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1278 parp->ar_pro != htons(ETH_P_IP) ||
1279 parp->ar_op != htons(ARPOP_REQUEST) ||
1280 parp->ar_hln != dev->addr_len ||
1281 parp->ar_pln != 4)
1282 goto out;
1283 arpptr = (u8 *)parp + sizeof(struct arphdr);
1284 sha = arpptr;
1285 arpptr += dev->addr_len; /* sha */
1286 memcpy(&sip, arpptr, sizeof(sip));
1287 arpptr += sizeof(sip);
1288 arpptr += dev->addr_len; /* tha */
1289 memcpy(&tip, arpptr, sizeof(tip));
1290
1291 if (ipv4_is_loopback(tip) ||
1292 ipv4_is_multicast(tip))
1293 goto out;
1294
1295 n = neigh_lookup(&arp_tbl, &tip, dev);
1296
1297 if (n) {
1298 struct vxlan_fdb *f;
1299 struct sk_buff *reply;
1300
1301 if (!(n->nud_state & NUD_CONNECTED)) {
1302 neigh_release(n);
1303 goto out;
1304 }
1305
1306 f = vxlan_find_mac(vxlan, n->ha);
1307 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1308 /* bridge-local neighbor */
1309 neigh_release(n);
1310 goto out;
1311 }
1312
1313 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1314 n->ha, sha);
1315
1316 neigh_release(n);
1317
1318 if (reply == NULL)
1319 goto out;
1320
1321 skb_reset_mac_header(reply);
1322 __skb_pull(reply, skb_network_offset(reply));
1323 reply->ip_summed = CHECKSUM_UNNECESSARY;
1324 reply->pkt_type = PACKET_HOST;
1325
1326 if (netif_rx_ni(reply) == NET_RX_DROP)
1327 dev->stats.rx_dropped++;
1328 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1329 union vxlan_addr ipa = {
1330 .sin.sin_addr.s_addr = tip,
1331 .sa.sa_family = AF_INET,
1332 };
1333
1334 vxlan_ip_miss(dev, &ipa);
1335 }
1336 out:
1337 consume_skb(skb);
1338 return NETDEV_TX_OK;
1339 }
1340
1341 #if IS_ENABLED(CONFIG_IPV6)
1342
1343 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1344 struct neighbour *n, bool isrouter)
1345 {
1346 struct net_device *dev = request->dev;
1347 struct sk_buff *reply;
1348 struct nd_msg *ns, *na;
1349 struct ipv6hdr *pip6;
1350 u8 *daddr;
1351 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1352 int ns_olen;
1353 int i, len;
1354
1355 if (dev == NULL)
1356 return NULL;
1357
1358 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1359 sizeof(*na) + na_olen + dev->needed_tailroom;
1360 reply = alloc_skb(len, GFP_ATOMIC);
1361 if (reply == NULL)
1362 return NULL;
1363
1364 reply->protocol = htons(ETH_P_IPV6);
1365 reply->dev = dev;
1366 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1367 skb_push(reply, sizeof(struct ethhdr));
1368 skb_set_mac_header(reply, 0);
1369
1370 ns = (struct nd_msg *)skb_transport_header(request);
1371
1372 daddr = eth_hdr(request)->h_source;
1373 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1374 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1375 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1376 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1377 break;
1378 }
1379 }
1380
1381 /* Ethernet header */
1382 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1383 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1384 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1385 reply->protocol = htons(ETH_P_IPV6);
1386
1387 skb_pull(reply, sizeof(struct ethhdr));
1388 skb_set_network_header(reply, 0);
1389 skb_put(reply, sizeof(struct ipv6hdr));
1390
1391 /* IPv6 header */
1392
1393 pip6 = ipv6_hdr(reply);
1394 memset(pip6, 0, sizeof(struct ipv6hdr));
1395 pip6->version = 6;
1396 pip6->priority = ipv6_hdr(request)->priority;
1397 pip6->nexthdr = IPPROTO_ICMPV6;
1398 pip6->hop_limit = 255;
1399 pip6->daddr = ipv6_hdr(request)->saddr;
1400 pip6->saddr = *(struct in6_addr *)n->primary_key;
1401
1402 skb_pull(reply, sizeof(struct ipv6hdr));
1403 skb_set_transport_header(reply, 0);
1404
1405 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1406
1407 /* Neighbor Advertisement */
1408 memset(na, 0, sizeof(*na)+na_olen);
1409 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1410 na->icmph.icmp6_router = isrouter;
1411 na->icmph.icmp6_override = 1;
1412 na->icmph.icmp6_solicited = 1;
1413 na->target = ns->target;
1414 ether_addr_copy(&na->opt[2], n->ha);
1415 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1416 na->opt[1] = na_olen >> 3;
1417
1418 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1419 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1420 csum_partial(na, sizeof(*na)+na_olen, 0));
1421
1422 pip6->payload_len = htons(sizeof(*na)+na_olen);
1423
1424 skb_push(reply, sizeof(struct ipv6hdr));
1425
1426 reply->ip_summed = CHECKSUM_UNNECESSARY;
1427
1428 return reply;
1429 }
1430
1431 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1432 {
1433 struct vxlan_dev *vxlan = netdev_priv(dev);
1434 struct nd_msg *msg;
1435 const struct ipv6hdr *iphdr;
1436 const struct in6_addr *saddr, *daddr;
1437 struct neighbour *n;
1438 struct inet6_dev *in6_dev;
1439
1440 in6_dev = __in6_dev_get(dev);
1441 if (!in6_dev)
1442 goto out;
1443
1444 if (!pskb_may_pull(skb, skb->len))
1445 goto out;
1446
1447 iphdr = ipv6_hdr(skb);
1448 saddr = &iphdr->saddr;
1449 daddr = &iphdr->daddr;
1450
1451 msg = (struct nd_msg *)skb_transport_header(skb);
1452 if (msg->icmph.icmp6_code != 0 ||
1453 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1454 goto out;
1455
1456 if (ipv6_addr_loopback(daddr) ||
1457 ipv6_addr_is_multicast(&msg->target))
1458 goto out;
1459
1460 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1461
1462 if (n) {
1463 struct vxlan_fdb *f;
1464 struct sk_buff *reply;
1465
1466 if (!(n->nud_state & NUD_CONNECTED)) {
1467 neigh_release(n);
1468 goto out;
1469 }
1470
1471 f = vxlan_find_mac(vxlan, n->ha);
1472 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1473 /* bridge-local neighbor */
1474 neigh_release(n);
1475 goto out;
1476 }
1477
1478 reply = vxlan_na_create(skb, n,
1479 !!(f ? f->flags & NTF_ROUTER : 0));
1480
1481 neigh_release(n);
1482
1483 if (reply == NULL)
1484 goto out;
1485
1486 if (netif_rx_ni(reply) == NET_RX_DROP)
1487 dev->stats.rx_dropped++;
1488
1489 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1490 union vxlan_addr ipa = {
1491 .sin6.sin6_addr = msg->target,
1492 .sa.sa_family = AF_INET6,
1493 };
1494
1495 vxlan_ip_miss(dev, &ipa);
1496 }
1497
1498 out:
1499 consume_skb(skb);
1500 return NETDEV_TX_OK;
1501 }
1502 #endif
1503
1504 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1505 {
1506 struct vxlan_dev *vxlan = netdev_priv(dev);
1507 struct neighbour *n;
1508
1509 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1510 return false;
1511
1512 n = NULL;
1513 switch (ntohs(eth_hdr(skb)->h_proto)) {
1514 case ETH_P_IP:
1515 {
1516 struct iphdr *pip;
1517
1518 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1519 return false;
1520 pip = ip_hdr(skb);
1521 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1522 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1523 union vxlan_addr ipa = {
1524 .sin.sin_addr.s_addr = pip->daddr,
1525 .sa.sa_family = AF_INET,
1526 };
1527
1528 vxlan_ip_miss(dev, &ipa);
1529 return false;
1530 }
1531
1532 break;
1533 }
1534 #if IS_ENABLED(CONFIG_IPV6)
1535 case ETH_P_IPV6:
1536 {
1537 struct ipv6hdr *pip6;
1538
1539 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1540 return false;
1541 pip6 = ipv6_hdr(skb);
1542 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1543 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1544 union vxlan_addr ipa = {
1545 .sin6.sin6_addr = pip6->daddr,
1546 .sa.sa_family = AF_INET6,
1547 };
1548
1549 vxlan_ip_miss(dev, &ipa);
1550 return false;
1551 }
1552
1553 break;
1554 }
1555 #endif
1556 default:
1557 return false;
1558 }
1559
1560 if (n) {
1561 bool diff;
1562
1563 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1564 if (diff) {
1565 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1566 dev->addr_len);
1567 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1568 }
1569 neigh_release(n);
1570 return diff;
1571 }
1572
1573 return false;
1574 }
1575
1576 /* Compute source port for outgoing packet
1577 * first choice to use L4 flow hash since it will spread
1578 * better and maybe available from hardware
1579 * secondary choice is to use jhash on the Ethernet header
1580 */
1581 __be16 vxlan_src_port(__u16 port_min, __u16 port_max, struct sk_buff *skb)
1582 {
1583 unsigned int range = (port_max - port_min) + 1;
1584 u32 hash;
1585
1586 hash = skb_get_hash(skb);
1587 if (!hash)
1588 hash = jhash(skb->data, 2 * ETH_ALEN,
1589 (__force u32) skb->protocol);
1590
1591 return htons((((u64) hash * range) >> 32) + port_min);
1592 }
1593 EXPORT_SYMBOL_GPL(vxlan_src_port);
1594
1595 static int handle_offloads(struct sk_buff *skb)
1596 {
1597 if (skb_is_gso(skb)) {
1598 int err = skb_unclone(skb, GFP_ATOMIC);
1599 if (unlikely(err))
1600 return err;
1601
1602 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
1603 } else if (skb->ip_summed != CHECKSUM_PARTIAL)
1604 skb->ip_summed = CHECKSUM_NONE;
1605
1606 return 0;
1607 }
1608
1609 #if IS_ENABLED(CONFIG_IPV6)
1610 static int vxlan6_xmit_skb(struct vxlan_sock *vs,
1611 struct dst_entry *dst, struct sk_buff *skb,
1612 struct net_device *dev, struct in6_addr *saddr,
1613 struct in6_addr *daddr, __u8 prio, __u8 ttl,
1614 __be16 src_port, __be16 dst_port, __be32 vni)
1615 {
1616 struct ipv6hdr *ip6h;
1617 struct vxlanhdr *vxh;
1618 struct udphdr *uh;
1619 int min_headroom;
1620 int err;
1621
1622 if (!skb->encapsulation) {
1623 skb_reset_inner_headers(skb);
1624 skb->encapsulation = 1;
1625 }
1626
1627 skb_scrub_packet(skb, false);
1628
1629 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1630 + VXLAN_HLEN + sizeof(struct ipv6hdr)
1631 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1632
1633 /* Need space for new headers (invalidates iph ptr) */
1634 err = skb_cow_head(skb, min_headroom);
1635 if (unlikely(err))
1636 return err;
1637
1638 if (vlan_tx_tag_present(skb)) {
1639 if (WARN_ON(!__vlan_put_tag(skb,
1640 skb->vlan_proto,
1641 vlan_tx_tag_get(skb))))
1642 return -ENOMEM;
1643
1644 skb->vlan_tci = 0;
1645 }
1646
1647 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1648 vxh->vx_flags = htonl(VXLAN_FLAGS);
1649 vxh->vx_vni = vni;
1650
1651 __skb_push(skb, sizeof(*uh));
1652 skb_reset_transport_header(skb);
1653 uh = udp_hdr(skb);
1654
1655 uh->dest = dst_port;
1656 uh->source = src_port;
1657
1658 uh->len = htons(skb->len);
1659 uh->check = 0;
1660
1661 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1662 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1663 IPSKB_REROUTED);
1664 skb_dst_set(skb, dst);
1665
1666 if (!skb_is_gso(skb) && !(dst->dev->features & NETIF_F_IPV6_CSUM)) {
1667 __wsum csum = skb_checksum(skb, 0, skb->len, 0);
1668 skb->ip_summed = CHECKSUM_UNNECESSARY;
1669 uh->check = csum_ipv6_magic(saddr, daddr, skb->len,
1670 IPPROTO_UDP, csum);
1671 if (uh->check == 0)
1672 uh->check = CSUM_MANGLED_0;
1673 } else {
1674 skb->ip_summed = CHECKSUM_PARTIAL;
1675 skb->csum_start = skb_transport_header(skb) - skb->head;
1676 skb->csum_offset = offsetof(struct udphdr, check);
1677 uh->check = ~csum_ipv6_magic(saddr, daddr,
1678 skb->len, IPPROTO_UDP, 0);
1679 }
1680
1681 __skb_push(skb, sizeof(*ip6h));
1682 skb_reset_network_header(skb);
1683 ip6h = ipv6_hdr(skb);
1684 ip6h->version = 6;
1685 ip6h->priority = prio;
1686 ip6h->flow_lbl[0] = 0;
1687 ip6h->flow_lbl[1] = 0;
1688 ip6h->flow_lbl[2] = 0;
1689 ip6h->payload_len = htons(skb->len);
1690 ip6h->nexthdr = IPPROTO_UDP;
1691 ip6h->hop_limit = ttl;
1692 ip6h->daddr = *daddr;
1693 ip6h->saddr = *saddr;
1694
1695 err = handle_offloads(skb);
1696 if (err)
1697 return err;
1698
1699 ip6tunnel_xmit(skb, dev);
1700 return 0;
1701 }
1702 #endif
1703
1704 int vxlan_xmit_skb(struct vxlan_sock *vs,
1705 struct rtable *rt, struct sk_buff *skb,
1706 __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1707 __be16 src_port, __be16 dst_port, __be32 vni)
1708 {
1709 struct vxlanhdr *vxh;
1710 struct udphdr *uh;
1711 int min_headroom;
1712 int err;
1713
1714 if (!skb->encapsulation) {
1715 skb_reset_inner_headers(skb);
1716 skb->encapsulation = 1;
1717 }
1718
1719 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1720 + VXLAN_HLEN + sizeof(struct iphdr)
1721 + (vlan_tx_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 return err;
1727
1728 if (vlan_tx_tag_present(skb)) {
1729 if (WARN_ON(!__vlan_put_tag(skb,
1730 skb->vlan_proto,
1731 vlan_tx_tag_get(skb))))
1732 return -ENOMEM;
1733
1734 skb->vlan_tci = 0;
1735 }
1736
1737 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1738 vxh->vx_flags = htonl(VXLAN_FLAGS);
1739 vxh->vx_vni = vni;
1740
1741 __skb_push(skb, sizeof(*uh));
1742 skb_reset_transport_header(skb);
1743 uh = udp_hdr(skb);
1744
1745 uh->dest = dst_port;
1746 uh->source = src_port;
1747
1748 uh->len = htons(skb->len);
1749 uh->check = 0;
1750
1751 err = handle_offloads(skb);
1752 if (err)
1753 return err;
1754
1755 return iptunnel_xmit(rt, skb, src, dst, IPPROTO_UDP, tos, ttl, df,
1756 false);
1757 }
1758 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1759
1760 /* Bypass encapsulation if the destination is local */
1761 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1762 struct vxlan_dev *dst_vxlan)
1763 {
1764 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1765 union vxlan_addr loopback;
1766 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1767
1768 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1769 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1770 skb->pkt_type = PACKET_HOST;
1771 skb->encapsulation = 0;
1772 skb->dev = dst_vxlan->dev;
1773 __skb_pull(skb, skb_network_offset(skb));
1774
1775 if (remote_ip->sa.sa_family == AF_INET) {
1776 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1777 loopback.sa.sa_family = AF_INET;
1778 #if IS_ENABLED(CONFIG_IPV6)
1779 } else {
1780 loopback.sin6.sin6_addr = in6addr_loopback;
1781 loopback.sa.sa_family = AF_INET6;
1782 #endif
1783 }
1784
1785 if (dst_vxlan->flags & VXLAN_F_LEARN)
1786 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1787
1788 u64_stats_update_begin(&tx_stats->syncp);
1789 tx_stats->tx_packets++;
1790 tx_stats->tx_bytes += skb->len;
1791 u64_stats_update_end(&tx_stats->syncp);
1792
1793 if (netif_rx(skb) == NET_RX_SUCCESS) {
1794 u64_stats_update_begin(&rx_stats->syncp);
1795 rx_stats->rx_packets++;
1796 rx_stats->rx_bytes += skb->len;
1797 u64_stats_update_end(&rx_stats->syncp);
1798 } else {
1799 skb->dev->stats.rx_dropped++;
1800 }
1801 }
1802
1803 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1804 struct vxlan_rdst *rdst, bool did_rsc)
1805 {
1806 struct vxlan_dev *vxlan = netdev_priv(dev);
1807 struct rtable *rt = NULL;
1808 const struct iphdr *old_iph;
1809 struct flowi4 fl4;
1810 union vxlan_addr *dst;
1811 __be16 src_port = 0, dst_port;
1812 u32 vni;
1813 __be16 df = 0;
1814 __u8 tos, ttl;
1815 int err;
1816
1817 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1818 vni = rdst->remote_vni;
1819 dst = &rdst->remote_ip;
1820
1821 if (vxlan_addr_any(dst)) {
1822 if (did_rsc) {
1823 /* short-circuited back to local bridge */
1824 vxlan_encap_bypass(skb, vxlan, vxlan);
1825 return;
1826 }
1827 goto drop;
1828 }
1829
1830 old_iph = ip_hdr(skb);
1831
1832 ttl = vxlan->ttl;
1833 if (!ttl && vxlan_addr_multicast(dst))
1834 ttl = 1;
1835
1836 tos = vxlan->tos;
1837 if (tos == 1)
1838 tos = ip_tunnel_get_dsfield(old_iph, skb);
1839
1840 src_port = vxlan_src_port(vxlan->port_min, vxlan->port_max, skb);
1841
1842 if (dst->sa.sa_family == AF_INET) {
1843 memset(&fl4, 0, sizeof(fl4));
1844 fl4.flowi4_oif = rdst->remote_ifindex;
1845 fl4.flowi4_tos = RT_TOS(tos);
1846 fl4.daddr = dst->sin.sin_addr.s_addr;
1847 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1848
1849 rt = ip_route_output_key(dev_net(dev), &fl4);
1850 if (IS_ERR(rt)) {
1851 netdev_dbg(dev, "no route to %pI4\n",
1852 &dst->sin.sin_addr.s_addr);
1853 dev->stats.tx_carrier_errors++;
1854 goto tx_error;
1855 }
1856
1857 if (rt->dst.dev == dev) {
1858 netdev_dbg(dev, "circular route to %pI4\n",
1859 &dst->sin.sin_addr.s_addr);
1860 dev->stats.collisions++;
1861 goto rt_tx_error;
1862 }
1863
1864 /* Bypass encapsulation if the destination is local */
1865 if (rt->rt_flags & RTCF_LOCAL &&
1866 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1867 struct vxlan_dev *dst_vxlan;
1868
1869 ip_rt_put(rt);
1870 dst_vxlan = vxlan_find_vni(dev_net(dev), vni, dst_port);
1871 if (!dst_vxlan)
1872 goto tx_error;
1873 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1874 return;
1875 }
1876
1877 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1878 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1879
1880 err = vxlan_xmit_skb(vxlan->vn_sock, rt, skb,
1881 fl4.saddr, dst->sin.sin_addr.s_addr,
1882 tos, ttl, df, src_port, dst_port,
1883 htonl(vni << 8));
1884
1885 if (err < 0)
1886 goto rt_tx_error;
1887 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1888 #if IS_ENABLED(CONFIG_IPV6)
1889 } else {
1890 struct sock *sk = vxlan->vn_sock->sock->sk;
1891 struct dst_entry *ndst;
1892 struct flowi6 fl6;
1893 u32 flags;
1894
1895 memset(&fl6, 0, sizeof(fl6));
1896 fl6.flowi6_oif = rdst->remote_ifindex;
1897 fl6.daddr = dst->sin6.sin6_addr;
1898 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
1899 fl6.flowi6_proto = IPPROTO_UDP;
1900
1901 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
1902 netdev_dbg(dev, "no route to %pI6\n",
1903 &dst->sin6.sin6_addr);
1904 dev->stats.tx_carrier_errors++;
1905 goto tx_error;
1906 }
1907
1908 if (ndst->dev == dev) {
1909 netdev_dbg(dev, "circular route to %pI6\n",
1910 &dst->sin6.sin6_addr);
1911 dst_release(ndst);
1912 dev->stats.collisions++;
1913 goto tx_error;
1914 }
1915
1916 /* Bypass encapsulation if the destination is local */
1917 flags = ((struct rt6_info *)ndst)->rt6i_flags;
1918 if (flags & RTF_LOCAL &&
1919 !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1920 struct vxlan_dev *dst_vxlan;
1921
1922 dst_release(ndst);
1923 dst_vxlan = vxlan_find_vni(dev_net(dev), vni, dst_port);
1924 if (!dst_vxlan)
1925 goto tx_error;
1926 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1927 return;
1928 }
1929
1930 ttl = ttl ? : ip6_dst_hoplimit(ndst);
1931
1932 err = vxlan6_xmit_skb(vxlan->vn_sock, ndst, skb,
1933 dev, &fl6.saddr, &fl6.daddr, 0, ttl,
1934 src_port, dst_port, htonl(vni << 8));
1935 #endif
1936 }
1937
1938 return;
1939
1940 drop:
1941 dev->stats.tx_dropped++;
1942 goto tx_free;
1943
1944 rt_tx_error:
1945 ip_rt_put(rt);
1946 tx_error:
1947 dev->stats.tx_errors++;
1948 tx_free:
1949 dev_kfree_skb(skb);
1950 }
1951
1952 /* Transmit local packets over Vxlan
1953 *
1954 * Outer IP header inherits ECN and DF from inner header.
1955 * Outer UDP destination is the VXLAN assigned port.
1956 * source port is based on hash of flow
1957 */
1958 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1959 {
1960 struct vxlan_dev *vxlan = netdev_priv(dev);
1961 struct ethhdr *eth;
1962 bool did_rsc = false;
1963 struct vxlan_rdst *rdst, *fdst = NULL;
1964 struct vxlan_fdb *f;
1965
1966 skb_reset_mac_header(skb);
1967 eth = eth_hdr(skb);
1968
1969 if ((vxlan->flags & VXLAN_F_PROXY)) {
1970 if (ntohs(eth->h_proto) == ETH_P_ARP)
1971 return arp_reduce(dev, skb);
1972 #if IS_ENABLED(CONFIG_IPV6)
1973 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
1974 skb->len >= sizeof(struct ipv6hdr) + sizeof(struct nd_msg) &&
1975 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
1976 struct nd_msg *msg;
1977
1978 msg = (struct nd_msg *)skb_transport_header(skb);
1979 if (msg->icmph.icmp6_code == 0 &&
1980 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
1981 return neigh_reduce(dev, skb);
1982 }
1983 #endif
1984 }
1985
1986 f = vxlan_find_mac(vxlan, eth->h_dest);
1987 did_rsc = false;
1988
1989 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1990 (ntohs(eth->h_proto) == ETH_P_IP ||
1991 ntohs(eth->h_proto) == ETH_P_IPV6)) {
1992 did_rsc = route_shortcircuit(dev, skb);
1993 if (did_rsc)
1994 f = vxlan_find_mac(vxlan, eth->h_dest);
1995 }
1996
1997 if (f == NULL) {
1998 f = vxlan_find_mac(vxlan, all_zeros_mac);
1999 if (f == NULL) {
2000 if ((vxlan->flags & VXLAN_F_L2MISS) &&
2001 !is_multicast_ether_addr(eth->h_dest))
2002 vxlan_fdb_miss(vxlan, eth->h_dest);
2003
2004 dev->stats.tx_dropped++;
2005 kfree_skb(skb);
2006 return NETDEV_TX_OK;
2007 }
2008 }
2009
2010 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2011 struct sk_buff *skb1;
2012
2013 if (!fdst) {
2014 fdst = rdst;
2015 continue;
2016 }
2017 skb1 = skb_clone(skb, GFP_ATOMIC);
2018 if (skb1)
2019 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2020 }
2021
2022 if (fdst)
2023 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2024 else
2025 kfree_skb(skb);
2026 return NETDEV_TX_OK;
2027 }
2028
2029 /* Walk the forwarding table and purge stale entries */
2030 static void vxlan_cleanup(unsigned long arg)
2031 {
2032 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2033 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2034 unsigned int h;
2035
2036 if (!netif_running(vxlan->dev))
2037 return;
2038
2039 spin_lock_bh(&vxlan->hash_lock);
2040 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2041 struct hlist_node *p, *n;
2042 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2043 struct vxlan_fdb *f
2044 = container_of(p, struct vxlan_fdb, hlist);
2045 unsigned long timeout;
2046
2047 if (f->state & NUD_PERMANENT)
2048 continue;
2049
2050 timeout = f->used + vxlan->age_interval * HZ;
2051 if (time_before_eq(timeout, jiffies)) {
2052 netdev_dbg(vxlan->dev,
2053 "garbage collect %pM\n",
2054 f->eth_addr);
2055 f->state = NUD_STALE;
2056 vxlan_fdb_destroy(vxlan, f);
2057 } else if (time_before(timeout, next_timer))
2058 next_timer = timeout;
2059 }
2060 }
2061 spin_unlock_bh(&vxlan->hash_lock);
2062
2063 mod_timer(&vxlan->age_timer, next_timer);
2064 }
2065
2066 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2067 {
2068 __u32 vni = vxlan->default_dst.remote_vni;
2069
2070 vxlan->vn_sock = vs;
2071 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2072 }
2073
2074 /* Setup stats when device is created */
2075 static int vxlan_init(struct net_device *dev)
2076 {
2077 struct vxlan_dev *vxlan = netdev_priv(dev);
2078 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2079 struct vxlan_sock *vs;
2080
2081 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2082 if (!dev->tstats)
2083 return -ENOMEM;
2084
2085 spin_lock(&vn->sock_lock);
2086 vs = vxlan_find_sock(dev_net(dev), vxlan->dst_port);
2087 if (vs) {
2088 /* If we have a socket with same port already, reuse it */
2089 atomic_inc(&vs->refcnt);
2090 vxlan_vs_add_dev(vs, vxlan);
2091 } else {
2092 /* otherwise make new socket outside of RTNL */
2093 dev_hold(dev);
2094 queue_work(vxlan_wq, &vxlan->sock_work);
2095 }
2096 spin_unlock(&vn->sock_lock);
2097
2098 return 0;
2099 }
2100
2101 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2102 {
2103 struct vxlan_fdb *f;
2104
2105 spin_lock_bh(&vxlan->hash_lock);
2106 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2107 if (f)
2108 vxlan_fdb_destroy(vxlan, f);
2109 spin_unlock_bh(&vxlan->hash_lock);
2110 }
2111
2112 static void vxlan_uninit(struct net_device *dev)
2113 {
2114 struct vxlan_dev *vxlan = netdev_priv(dev);
2115 struct vxlan_sock *vs = vxlan->vn_sock;
2116
2117 vxlan_fdb_delete_default(vxlan);
2118
2119 if (vs)
2120 vxlan_sock_release(vs);
2121 free_percpu(dev->tstats);
2122 }
2123
2124 /* Start ageing timer and join group when device is brought up */
2125 static int vxlan_open(struct net_device *dev)
2126 {
2127 struct vxlan_dev *vxlan = netdev_priv(dev);
2128 struct vxlan_sock *vs = vxlan->vn_sock;
2129
2130 /* socket hasn't been created */
2131 if (!vs)
2132 return -ENOTCONN;
2133
2134 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2135 vxlan_sock_hold(vs);
2136 dev_hold(dev);
2137 queue_work(vxlan_wq, &vxlan->igmp_join);
2138 }
2139
2140 if (vxlan->age_interval)
2141 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2142
2143 return 0;
2144 }
2145
2146 /* Purge the forwarding table */
2147 static void vxlan_flush(struct vxlan_dev *vxlan)
2148 {
2149 unsigned int h;
2150
2151 spin_lock_bh(&vxlan->hash_lock);
2152 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2153 struct hlist_node *p, *n;
2154 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2155 struct vxlan_fdb *f
2156 = container_of(p, struct vxlan_fdb, hlist);
2157 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2158 if (!is_zero_ether_addr(f->eth_addr))
2159 vxlan_fdb_destroy(vxlan, f);
2160 }
2161 }
2162 spin_unlock_bh(&vxlan->hash_lock);
2163 }
2164
2165 /* Cleanup timer and forwarding table on shutdown */
2166 static int vxlan_stop(struct net_device *dev)
2167 {
2168 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2169 struct vxlan_dev *vxlan = netdev_priv(dev);
2170 struct vxlan_sock *vs = vxlan->vn_sock;
2171
2172 if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2173 !vxlan_group_used(vn, vxlan)) {
2174 vxlan_sock_hold(vs);
2175 dev_hold(dev);
2176 queue_work(vxlan_wq, &vxlan->igmp_leave);
2177 }
2178
2179 del_timer_sync(&vxlan->age_timer);
2180
2181 vxlan_flush(vxlan);
2182
2183 return 0;
2184 }
2185
2186 /* Stub, nothing needs to be done. */
2187 static void vxlan_set_multicast_list(struct net_device *dev)
2188 {
2189 }
2190
2191 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2192 {
2193 struct vxlan_dev *vxlan = netdev_priv(dev);
2194 struct vxlan_rdst *dst = &vxlan->default_dst;
2195 struct net_device *lowerdev;
2196 int max_mtu;
2197
2198 lowerdev = __dev_get_by_index(dev_net(dev), dst->remote_ifindex);
2199 if (lowerdev == NULL)
2200 return eth_change_mtu(dev, new_mtu);
2201
2202 if (dst->remote_ip.sa.sa_family == AF_INET6)
2203 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2204 else
2205 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2206
2207 if (new_mtu < 68 || new_mtu > max_mtu)
2208 return -EINVAL;
2209
2210 dev->mtu = new_mtu;
2211 return 0;
2212 }
2213
2214 static const struct net_device_ops vxlan_netdev_ops = {
2215 .ndo_init = vxlan_init,
2216 .ndo_uninit = vxlan_uninit,
2217 .ndo_open = vxlan_open,
2218 .ndo_stop = vxlan_stop,
2219 .ndo_start_xmit = vxlan_xmit,
2220 .ndo_get_stats64 = ip_tunnel_get_stats64,
2221 .ndo_set_rx_mode = vxlan_set_multicast_list,
2222 .ndo_change_mtu = vxlan_change_mtu,
2223 .ndo_validate_addr = eth_validate_addr,
2224 .ndo_set_mac_address = eth_mac_addr,
2225 .ndo_fdb_add = vxlan_fdb_add,
2226 .ndo_fdb_del = vxlan_fdb_delete,
2227 .ndo_fdb_dump = vxlan_fdb_dump,
2228 };
2229
2230 /* Info for udev, that this is a virtual tunnel endpoint */
2231 static struct device_type vxlan_type = {
2232 .name = "vxlan",
2233 };
2234
2235 /* Calls the ndo_add_vxlan_port of the caller in order to
2236 * supply the listening VXLAN udp ports. Callers are expected
2237 * to implement the ndo_add_vxlan_port.
2238 */
2239 void vxlan_get_rx_port(struct net_device *dev)
2240 {
2241 struct vxlan_sock *vs;
2242 struct net *net = dev_net(dev);
2243 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2244 sa_family_t sa_family;
2245 __be16 port;
2246 unsigned int i;
2247
2248 spin_lock(&vn->sock_lock);
2249 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2250 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2251 port = inet_sk(vs->sock->sk)->inet_sport;
2252 sa_family = vs->sock->sk->sk_family;
2253 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2254 port);
2255 }
2256 }
2257 spin_unlock(&vn->sock_lock);
2258 }
2259 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2260
2261 /* Initialize the device structure. */
2262 static void vxlan_setup(struct net_device *dev)
2263 {
2264 struct vxlan_dev *vxlan = netdev_priv(dev);
2265 unsigned int h;
2266 int low, high;
2267
2268 eth_hw_addr_random(dev);
2269 ether_setup(dev);
2270 if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2271 dev->hard_header_len = ETH_HLEN + VXLAN6_HEADROOM;
2272 else
2273 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
2274
2275 dev->netdev_ops = &vxlan_netdev_ops;
2276 dev->destructor = free_netdev;
2277 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2278
2279 dev->tx_queue_len = 0;
2280 dev->features |= NETIF_F_LLTX;
2281 dev->features |= NETIF_F_NETNS_LOCAL;
2282 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2283 dev->features |= NETIF_F_RXCSUM;
2284 dev->features |= NETIF_F_GSO_SOFTWARE;
2285
2286 dev->vlan_features = dev->features;
2287 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2288 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2289 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2290 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2291 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
2292 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2293
2294 INIT_LIST_HEAD(&vxlan->next);
2295 spin_lock_init(&vxlan->hash_lock);
2296 INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join);
2297 INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave);
2298 INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
2299
2300 init_timer_deferrable(&vxlan->age_timer);
2301 vxlan->age_timer.function = vxlan_cleanup;
2302 vxlan->age_timer.data = (unsigned long) vxlan;
2303
2304 inet_get_local_port_range(dev_net(dev), &low, &high);
2305 vxlan->port_min = low;
2306 vxlan->port_max = high;
2307 vxlan->dst_port = htons(vxlan_port);
2308
2309 vxlan->dev = dev;
2310
2311 for (h = 0; h < FDB_HASH_SIZE; ++h)
2312 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2313 }
2314
2315 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2316 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2317 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2318 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2319 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2320 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2321 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2322 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2323 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2324 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2325 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2326 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2327 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2328 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2329 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2330 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2331 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2332 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2333 };
2334
2335 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2336 {
2337 if (tb[IFLA_ADDRESS]) {
2338 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2339 pr_debug("invalid link address (not ethernet)\n");
2340 return -EINVAL;
2341 }
2342
2343 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2344 pr_debug("invalid all zero ethernet address\n");
2345 return -EADDRNOTAVAIL;
2346 }
2347 }
2348
2349 if (!data)
2350 return -EINVAL;
2351
2352 if (data[IFLA_VXLAN_ID]) {
2353 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2354 if (id >= VXLAN_VID_MASK)
2355 return -ERANGE;
2356 }
2357
2358 if (data[IFLA_VXLAN_PORT_RANGE]) {
2359 const struct ifla_vxlan_port_range *p
2360 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2361
2362 if (ntohs(p->high) < ntohs(p->low)) {
2363 pr_debug("port range %u .. %u not valid\n",
2364 ntohs(p->low), ntohs(p->high));
2365 return -EINVAL;
2366 }
2367 }
2368
2369 return 0;
2370 }
2371
2372 static void vxlan_get_drvinfo(struct net_device *netdev,
2373 struct ethtool_drvinfo *drvinfo)
2374 {
2375 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2376 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2377 }
2378
2379 static const struct ethtool_ops vxlan_ethtool_ops = {
2380 .get_drvinfo = vxlan_get_drvinfo,
2381 .get_link = ethtool_op_get_link,
2382 };
2383
2384 static void vxlan_del_work(struct work_struct *work)
2385 {
2386 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2387
2388 sk_release_kernel(vs->sock->sk);
2389 kfree_rcu(vs, rcu);
2390 }
2391
2392 #if IS_ENABLED(CONFIG_IPV6)
2393 /* Create UDP socket for encapsulation receive. AF_INET6 socket
2394 * could be used for both IPv4 and IPv6 communications, but
2395 * users may set bindv6only=1.
2396 */
2397 static struct socket *create_v6_sock(struct net *net, __be16 port)
2398 {
2399 struct sock *sk;
2400 struct socket *sock;
2401 struct sockaddr_in6 vxlan_addr = {
2402 .sin6_family = AF_INET6,
2403 .sin6_port = port,
2404 };
2405 int rc, val = 1;
2406
2407 rc = sock_create_kern(AF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
2408 if (rc < 0) {
2409 pr_debug("UDPv6 socket create failed\n");
2410 return ERR_PTR(rc);
2411 }
2412
2413 /* Put in proper namespace */
2414 sk = sock->sk;
2415 sk_change_net(sk, net);
2416
2417 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
2418 (char *)&val, sizeof(val));
2419 rc = kernel_bind(sock, (struct sockaddr *)&vxlan_addr,
2420 sizeof(struct sockaddr_in6));
2421 if (rc < 0) {
2422 pr_debug("bind for UDPv6 socket %pI6:%u (%d)\n",
2423 &vxlan_addr.sin6_addr, ntohs(vxlan_addr.sin6_port), rc);
2424 sk_release_kernel(sk);
2425 return ERR_PTR(rc);
2426 }
2427 /* At this point, IPv6 module should have been loaded in
2428 * sock_create_kern().
2429 */
2430 BUG_ON(!ipv6_stub);
2431
2432 /* Disable multicast loopback */
2433 inet_sk(sk)->mc_loop = 0;
2434 return sock;
2435 }
2436
2437 #else
2438
2439 static struct socket *create_v6_sock(struct net *net, __be16 port)
2440 {
2441 return ERR_PTR(-EPFNOSUPPORT);
2442 }
2443 #endif
2444
2445 static struct socket *create_v4_sock(struct net *net, __be16 port)
2446 {
2447 struct sock *sk;
2448 struct socket *sock;
2449 struct sockaddr_in vxlan_addr = {
2450 .sin_family = AF_INET,
2451 .sin_addr.s_addr = htonl(INADDR_ANY),
2452 .sin_port = port,
2453 };
2454 int rc;
2455
2456 /* Create UDP socket for encapsulation receive. */
2457 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
2458 if (rc < 0) {
2459 pr_debug("UDP socket create failed\n");
2460 return ERR_PTR(rc);
2461 }
2462
2463 /* Put in proper namespace */
2464 sk = sock->sk;
2465 sk_change_net(sk, net);
2466
2467 rc = kernel_bind(sock, (struct sockaddr *) &vxlan_addr,
2468 sizeof(vxlan_addr));
2469 if (rc < 0) {
2470 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
2471 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
2472 sk_release_kernel(sk);
2473 return ERR_PTR(rc);
2474 }
2475
2476 /* Disable multicast loopback */
2477 inet_sk(sk)->mc_loop = 0;
2478 return sock;
2479 }
2480
2481 /* Create new listen socket if needed */
2482 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2483 vxlan_rcv_t *rcv, void *data, bool ipv6)
2484 {
2485 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2486 struct vxlan_sock *vs;
2487 struct socket *sock;
2488 struct sock *sk;
2489 unsigned int h;
2490
2491 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2492 if (!vs)
2493 return ERR_PTR(-ENOMEM);
2494
2495 for (h = 0; h < VNI_HASH_SIZE; ++h)
2496 INIT_HLIST_HEAD(&vs->vni_list[h]);
2497
2498 INIT_WORK(&vs->del_work, vxlan_del_work);
2499
2500 if (ipv6)
2501 sock = create_v6_sock(net, port);
2502 else
2503 sock = create_v4_sock(net, port);
2504 if (IS_ERR(sock)) {
2505 kfree(vs);
2506 return ERR_CAST(sock);
2507 }
2508
2509 vs->sock = sock;
2510 sk = sock->sk;
2511 atomic_set(&vs->refcnt, 1);
2512 vs->rcv = rcv;
2513 vs->data = data;
2514 rcu_assign_sk_user_data(vs->sock->sk, vs);
2515
2516 /* Initialize the vxlan udp offloads structure */
2517 vs->udp_offloads.port = port;
2518 vs->udp_offloads.callbacks.gro_receive = vxlan_gro_receive;
2519 vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2520
2521 spin_lock(&vn->sock_lock);
2522 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2523 vxlan_notify_add_rx_port(vs);
2524 spin_unlock(&vn->sock_lock);
2525
2526 /* Mark socket as an encapsulation socket. */
2527 udp_sk(sk)->encap_type = 1;
2528 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
2529 #if IS_ENABLED(CONFIG_IPV6)
2530 if (ipv6)
2531 ipv6_stub->udpv6_encap_enable();
2532 else
2533 #endif
2534 udp_encap_enable();
2535
2536 return vs;
2537 }
2538
2539 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2540 vxlan_rcv_t *rcv, void *data,
2541 bool no_share, bool ipv6)
2542 {
2543 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2544 struct vxlan_sock *vs;
2545
2546 vs = vxlan_socket_create(net, port, rcv, data, ipv6);
2547 if (!IS_ERR(vs))
2548 return vs;
2549
2550 if (no_share) /* Return error if sharing is not allowed. */
2551 return vs;
2552
2553 spin_lock(&vn->sock_lock);
2554 vs = vxlan_find_sock(net, port);
2555 if (vs) {
2556 if (vs->rcv == rcv)
2557 atomic_inc(&vs->refcnt);
2558 else
2559 vs = ERR_PTR(-EBUSY);
2560 }
2561 spin_unlock(&vn->sock_lock);
2562
2563 if (!vs)
2564 vs = ERR_PTR(-EINVAL);
2565
2566 return vs;
2567 }
2568 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2569
2570 /* Scheduled at device creation to bind to a socket */
2571 static void vxlan_sock_work(struct work_struct *work)
2572 {
2573 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2574 struct net *net = dev_net(vxlan->dev);
2575 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2576 __be16 port = vxlan->dst_port;
2577 struct vxlan_sock *nvs;
2578
2579 nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags & VXLAN_F_IPV6);
2580 spin_lock(&vn->sock_lock);
2581 if (!IS_ERR(nvs))
2582 vxlan_vs_add_dev(nvs, vxlan);
2583 spin_unlock(&vn->sock_lock);
2584
2585 dev_put(vxlan->dev);
2586 }
2587
2588 static int vxlan_newlink(struct net *net, struct net_device *dev,
2589 struct nlattr *tb[], struct nlattr *data[])
2590 {
2591 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2592 struct vxlan_dev *vxlan = netdev_priv(dev);
2593 struct vxlan_rdst *dst = &vxlan->default_dst;
2594 __u32 vni;
2595 int err;
2596 bool use_ipv6 = false;
2597
2598 if (!data[IFLA_VXLAN_ID])
2599 return -EINVAL;
2600
2601 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2602 dst->remote_vni = vni;
2603
2604 if (data[IFLA_VXLAN_GROUP]) {
2605 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2606 dst->remote_ip.sa.sa_family = AF_INET;
2607 } else if (data[IFLA_VXLAN_GROUP6]) {
2608 if (!IS_ENABLED(CONFIG_IPV6))
2609 return -EPFNOSUPPORT;
2610
2611 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2612 sizeof(struct in6_addr));
2613 dst->remote_ip.sa.sa_family = AF_INET6;
2614 use_ipv6 = true;
2615 }
2616
2617 if (data[IFLA_VXLAN_LOCAL]) {
2618 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2619 vxlan->saddr.sa.sa_family = AF_INET;
2620 } else if (data[IFLA_VXLAN_LOCAL6]) {
2621 if (!IS_ENABLED(CONFIG_IPV6))
2622 return -EPFNOSUPPORT;
2623
2624 /* TODO: respect scope id */
2625 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2626 sizeof(struct in6_addr));
2627 vxlan->saddr.sa.sa_family = AF_INET6;
2628 use_ipv6 = true;
2629 }
2630
2631 if (data[IFLA_VXLAN_LINK] &&
2632 (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2633 struct net_device *lowerdev
2634 = __dev_get_by_index(net, dst->remote_ifindex);
2635
2636 if (!lowerdev) {
2637 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2638 return -ENODEV;
2639 }
2640
2641 #if IS_ENABLED(CONFIG_IPV6)
2642 if (use_ipv6) {
2643 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2644 if (idev && idev->cnf.disable_ipv6) {
2645 pr_info("IPv6 is disabled via sysctl\n");
2646 return -EPERM;
2647 }
2648 vxlan->flags |= VXLAN_F_IPV6;
2649 }
2650 #endif
2651
2652 if (!tb[IFLA_MTU])
2653 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2654
2655 /* update header length based on lower device */
2656 dev->hard_header_len = lowerdev->hard_header_len +
2657 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2658 } else if (use_ipv6)
2659 vxlan->flags |= VXLAN_F_IPV6;
2660
2661 if (data[IFLA_VXLAN_TOS])
2662 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2663
2664 if (data[IFLA_VXLAN_TTL])
2665 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2666
2667 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2668 vxlan->flags |= VXLAN_F_LEARN;
2669
2670 if (data[IFLA_VXLAN_AGEING])
2671 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2672 else
2673 vxlan->age_interval = FDB_AGE_DEFAULT;
2674
2675 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2676 vxlan->flags |= VXLAN_F_PROXY;
2677
2678 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2679 vxlan->flags |= VXLAN_F_RSC;
2680
2681 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2682 vxlan->flags |= VXLAN_F_L2MISS;
2683
2684 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2685 vxlan->flags |= VXLAN_F_L3MISS;
2686
2687 if (data[IFLA_VXLAN_LIMIT])
2688 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2689
2690 if (data[IFLA_VXLAN_PORT_RANGE]) {
2691 const struct ifla_vxlan_port_range *p
2692 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2693 vxlan->port_min = ntohs(p->low);
2694 vxlan->port_max = ntohs(p->high);
2695 }
2696
2697 if (data[IFLA_VXLAN_PORT])
2698 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2699
2700 if (vxlan_find_vni(net, vni, vxlan->dst_port)) {
2701 pr_info("duplicate VNI %u\n", vni);
2702 return -EEXIST;
2703 }
2704
2705 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
2706
2707 /* create an fdb entry for a valid default destination */
2708 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2709 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2710 &vxlan->default_dst.remote_ip,
2711 NUD_REACHABLE|NUD_PERMANENT,
2712 NLM_F_EXCL|NLM_F_CREATE,
2713 vxlan->dst_port,
2714 vxlan->default_dst.remote_vni,
2715 vxlan->default_dst.remote_ifindex,
2716 NTF_SELF);
2717 if (err)
2718 return err;
2719 }
2720
2721 err = register_netdevice(dev);
2722 if (err) {
2723 vxlan_fdb_delete_default(vxlan);
2724 return err;
2725 }
2726
2727 list_add(&vxlan->next, &vn->vxlan_list);
2728
2729 return 0;
2730 }
2731
2732 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2733 {
2734 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2735 struct vxlan_dev *vxlan = netdev_priv(dev);
2736
2737 spin_lock(&vn->sock_lock);
2738 if (!hlist_unhashed(&vxlan->hlist))
2739 hlist_del_rcu(&vxlan->hlist);
2740 spin_unlock(&vn->sock_lock);
2741
2742 list_del(&vxlan->next);
2743 unregister_netdevice_queue(dev, head);
2744 }
2745
2746 static size_t vxlan_get_size(const struct net_device *dev)
2747 {
2748
2749 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
2750 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2751 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2752 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2753 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
2754 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
2755 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
2756 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
2757 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
2758 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
2759 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
2760 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2761 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2762 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2763 nla_total_size(sizeof(__be16))+ /* IFLA_VXLAN_PORT */
2764 0;
2765 }
2766
2767 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2768 {
2769 const struct vxlan_dev *vxlan = netdev_priv(dev);
2770 const struct vxlan_rdst *dst = &vxlan->default_dst;
2771 struct ifla_vxlan_port_range ports = {
2772 .low = htons(vxlan->port_min),
2773 .high = htons(vxlan->port_max),
2774 };
2775
2776 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2777 goto nla_put_failure;
2778
2779 if (!vxlan_addr_any(&dst->remote_ip)) {
2780 if (dst->remote_ip.sa.sa_family == AF_INET) {
2781 if (nla_put_be32(skb, IFLA_VXLAN_GROUP,
2782 dst->remote_ip.sin.sin_addr.s_addr))
2783 goto nla_put_failure;
2784 #if IS_ENABLED(CONFIG_IPV6)
2785 } else {
2786 if (nla_put(skb, IFLA_VXLAN_GROUP6, sizeof(struct in6_addr),
2787 &dst->remote_ip.sin6.sin6_addr))
2788 goto nla_put_failure;
2789 #endif
2790 }
2791 }
2792
2793 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2794 goto nla_put_failure;
2795
2796 if (!vxlan_addr_any(&vxlan->saddr)) {
2797 if (vxlan->saddr.sa.sa_family == AF_INET) {
2798 if (nla_put_be32(skb, IFLA_VXLAN_LOCAL,
2799 vxlan->saddr.sin.sin_addr.s_addr))
2800 goto nla_put_failure;
2801 #if IS_ENABLED(CONFIG_IPV6)
2802 } else {
2803 if (nla_put(skb, IFLA_VXLAN_LOCAL6, sizeof(struct in6_addr),
2804 &vxlan->saddr.sin6.sin6_addr))
2805 goto nla_put_failure;
2806 #endif
2807 }
2808 }
2809
2810 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2811 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2812 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2813 !!(vxlan->flags & VXLAN_F_LEARN)) ||
2814 nla_put_u8(skb, IFLA_VXLAN_PROXY,
2815 !!(vxlan->flags & VXLAN_F_PROXY)) ||
2816 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2817 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2818 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2819 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2820 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2821 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2822 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2823 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port))
2824 goto nla_put_failure;
2825
2826 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2827 goto nla_put_failure;
2828
2829 return 0;
2830
2831 nla_put_failure:
2832 return -EMSGSIZE;
2833 }
2834
2835 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2836 .kind = "vxlan",
2837 .maxtype = IFLA_VXLAN_MAX,
2838 .policy = vxlan_policy,
2839 .priv_size = sizeof(struct vxlan_dev),
2840 .setup = vxlan_setup,
2841 .validate = vxlan_validate,
2842 .newlink = vxlan_newlink,
2843 .dellink = vxlan_dellink,
2844 .get_size = vxlan_get_size,
2845 .fill_info = vxlan_fill_info,
2846 };
2847
2848 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
2849 struct net_device *dev)
2850 {
2851 struct vxlan_dev *vxlan, *next;
2852 LIST_HEAD(list_kill);
2853
2854 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2855 struct vxlan_rdst *dst = &vxlan->default_dst;
2856
2857 /* In case we created vxlan device with carrier
2858 * and we loose the carrier due to module unload
2859 * we also need to remove vxlan device. In other
2860 * cases, it's not necessary and remote_ifindex
2861 * is 0 here, so no matches.
2862 */
2863 if (dst->remote_ifindex == dev->ifindex)
2864 vxlan_dellink(vxlan->dev, &list_kill);
2865 }
2866
2867 unregister_netdevice_many(&list_kill);
2868 }
2869
2870 static int vxlan_lowerdev_event(struct notifier_block *unused,
2871 unsigned long event, void *ptr)
2872 {
2873 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2874 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2875
2876 if (event == NETDEV_UNREGISTER)
2877 vxlan_handle_lowerdev_unregister(vn, dev);
2878
2879 return NOTIFY_DONE;
2880 }
2881
2882 static struct notifier_block vxlan_notifier_block __read_mostly = {
2883 .notifier_call = vxlan_lowerdev_event,
2884 };
2885
2886 static __net_init int vxlan_init_net(struct net *net)
2887 {
2888 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2889 unsigned int h;
2890
2891 INIT_LIST_HEAD(&vn->vxlan_list);
2892 spin_lock_init(&vn->sock_lock);
2893
2894 for (h = 0; h < PORT_HASH_SIZE; ++h)
2895 INIT_HLIST_HEAD(&vn->sock_list[h]);
2896
2897 return 0;
2898 }
2899
2900 static struct pernet_operations vxlan_net_ops = {
2901 .init = vxlan_init_net,
2902 .id = &vxlan_net_id,
2903 .size = sizeof(struct vxlan_net),
2904 };
2905
2906 static int __init vxlan_init_module(void)
2907 {
2908 int rc;
2909
2910 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
2911 if (!vxlan_wq)
2912 return -ENOMEM;
2913
2914 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
2915
2916 rc = register_pernet_subsys(&vxlan_net_ops);
2917 if (rc)
2918 goto out1;
2919
2920 rc = register_netdevice_notifier(&vxlan_notifier_block);
2921 if (rc)
2922 goto out2;
2923
2924 rc = rtnl_link_register(&vxlan_link_ops);
2925 if (rc)
2926 goto out3;
2927
2928 return 0;
2929 out3:
2930 unregister_netdevice_notifier(&vxlan_notifier_block);
2931 out2:
2932 unregister_pernet_subsys(&vxlan_net_ops);
2933 out1:
2934 destroy_workqueue(vxlan_wq);
2935 return rc;
2936 }
2937 late_initcall(vxlan_init_module);
2938
2939 static void __exit vxlan_cleanup_module(void)
2940 {
2941 rtnl_link_unregister(&vxlan_link_ops);
2942 unregister_netdevice_notifier(&vxlan_notifier_block);
2943 destroy_workqueue(vxlan_wq);
2944 unregister_pernet_subsys(&vxlan_net_ops);
2945 /* rcu_barrier() is called by netns */
2946 }
2947 module_exit(vxlan_cleanup_module);
2948
2949 MODULE_LICENSE("GPL");
2950 MODULE_VERSION(VXLAN_VERSION);
2951 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
2952 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
2953 MODULE_ALIAS_RTNL_LINK("vxlan");
Linux kernel - Deliverables
This page took 0.125734 seconds and 6 git commands to generate.