projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: introduce netdev_alloc_pcpu_stats() for drivers
[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 __be16 port;
1136
1137 /* Need Vxlan and inner Ethernet header to be present */
1138 if (!pskb_may_pull(skb, VXLAN_HLEN))
1139 goto error;
1140
1141 /* Return packets with reserved bits set */
1142 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1143 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
1144 (vxh->vx_vni & htonl(0xff))) {
1145 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1146 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1147 goto error;
1148 }
1149
1150 if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1151 goto drop;
1152
1153 port = inet_sk(sk)->inet_sport;
1154
1155 vs = rcu_dereference_sk_user_data(sk);
1156 if (!vs)
1157 goto drop;
1158
1159 /* If the NIC driver gave us an encapsulated packet
1160 * with the encapsulation mark, the device checksummed it
1161 * for us. Otherwise force the upper layers to verify it.
1162 */
1163 if ((skb->ip_summed != CHECKSUM_UNNECESSARY && skb->ip_summed != CHECKSUM_PARTIAL) ||
1164 !skb->encapsulation)
1165 skb->ip_summed = CHECKSUM_NONE;
1166
1167 skb->encapsulation = 0;
1168
1169 vs->rcv(vs, skb, vxh->vx_vni);
1170 return 0;
1171
1172 drop:
1173 /* Consume bad packet */
1174 kfree_skb(skb);
1175 return 0;
1176
1177 error:
1178 /* Return non vxlan pkt */
1179 return 1;
1180 }
1181
1182 static void vxlan_rcv(struct vxlan_sock *vs,
1183 struct sk_buff *skb, __be32 vx_vni)
1184 {
1185 struct iphdr *oip = NULL;
1186 struct ipv6hdr *oip6 = NULL;
1187 struct vxlan_dev *vxlan;
1188 struct pcpu_sw_netstats *stats;
1189 union vxlan_addr saddr;
1190 __u32 vni;
1191 int err = 0;
1192 union vxlan_addr *remote_ip;
1193
1194 vni = ntohl(vx_vni) >> 8;
1195 /* Is this VNI defined? */
1196 vxlan = vxlan_vs_find_vni(vs, vni);
1197 if (!vxlan)
1198 goto drop;
1199
1200 remote_ip = &vxlan->default_dst.remote_ip;
1201 skb_reset_mac_header(skb);
1202 skb->protocol = eth_type_trans(skb, vxlan->dev);
1203
1204 /* Ignore packet loops (and multicast echo) */
1205 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1206 goto drop;
1207
1208 /* Re-examine inner Ethernet packet */
1209 if (remote_ip->sa.sa_family == AF_INET) {
1210 oip = ip_hdr(skb);
1211 saddr.sin.sin_addr.s_addr = oip->saddr;
1212 saddr.sa.sa_family = AF_INET;
1213 #if IS_ENABLED(CONFIG_IPV6)
1214 } else {
1215 oip6 = ipv6_hdr(skb);
1216 saddr.sin6.sin6_addr = oip6->saddr;
1217 saddr.sa.sa_family = AF_INET6;
1218 #endif
1219 }
1220
1221 if ((vxlan->flags & VXLAN_F_LEARN) &&
1222 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1223 goto drop;
1224
1225 skb_reset_network_header(skb);
1226
1227 if (oip6)
1228 err = IP6_ECN_decapsulate(oip6, skb);
1229 if (oip)
1230 err = IP_ECN_decapsulate(oip, skb);
1231
1232 if (unlikely(err)) {
1233 if (log_ecn_error) {
1234 if (oip6)
1235 net_info_ratelimited("non-ECT from %pI6\n",
1236 &oip6->saddr);
1237 if (oip)
1238 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1239 &oip->saddr, oip->tos);
1240 }
1241 if (err > 1) {
1242 ++vxlan->dev->stats.rx_frame_errors;
1243 ++vxlan->dev->stats.rx_errors;
1244 goto drop;
1245 }
1246 }
1247
1248 stats = this_cpu_ptr(vxlan->dev->tstats);
1249 u64_stats_update_begin(&stats->syncp);
1250 stats->rx_packets++;
1251 stats->rx_bytes += skb->len;
1252 u64_stats_update_end(&stats->syncp);
1253
1254 netif_rx(skb);
1255
1256 return;
1257 drop:
1258 /* Consume bad packet */
1259 kfree_skb(skb);
1260 }
1261
1262 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1263 {
1264 struct vxlan_dev *vxlan = netdev_priv(dev);
1265 struct arphdr *parp;
1266 u8 *arpptr, *sha;
1267 __be32 sip, tip;
1268 struct neighbour *n;
1269
1270 if (dev->flags & IFF_NOARP)
1271 goto out;
1272
1273 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1274 dev->stats.tx_dropped++;
1275 goto out;
1276 }
1277 parp = arp_hdr(skb);
1278
1279 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1280 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1281 parp->ar_pro != htons(ETH_P_IP) ||
1282 parp->ar_op != htons(ARPOP_REQUEST) ||
1283 parp->ar_hln != dev->addr_len ||
1284 parp->ar_pln != 4)
1285 goto out;
1286 arpptr = (u8 *)parp + sizeof(struct arphdr);
1287 sha = arpptr;
1288 arpptr += dev->addr_len; /* sha */
1289 memcpy(&sip, arpptr, sizeof(sip));
1290 arpptr += sizeof(sip);
1291 arpptr += dev->addr_len; /* tha */
1292 memcpy(&tip, arpptr, sizeof(tip));
1293
1294 if (ipv4_is_loopback(tip) ||
1295 ipv4_is_multicast(tip))
1296 goto out;
1297
1298 n = neigh_lookup(&arp_tbl, &tip, dev);
1299
1300 if (n) {
1301 struct vxlan_fdb *f;
1302 struct sk_buff *reply;
1303
1304 if (!(n->nud_state & NUD_CONNECTED)) {
1305 neigh_release(n);
1306 goto out;
1307 }
1308
1309 f = vxlan_find_mac(vxlan, n->ha);
1310 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1311 /* bridge-local neighbor */
1312 neigh_release(n);
1313 goto out;
1314 }
1315
1316 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1317 n->ha, sha);
1318
1319 neigh_release(n);
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 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1343 {
1344 struct vxlan_dev *vxlan = netdev_priv(dev);
1345 struct neighbour *n;
1346 union vxlan_addr ipa;
1347 const struct ipv6hdr *iphdr;
1348 const struct in6_addr *saddr, *daddr;
1349 struct nd_msg *msg;
1350 struct inet6_dev *in6_dev = NULL;
1351
1352 in6_dev = __in6_dev_get(dev);
1353 if (!in6_dev)
1354 goto out;
1355
1356 if (!pskb_may_pull(skb, skb->len))
1357 goto out;
1358
1359 iphdr = ipv6_hdr(skb);
1360 saddr = &iphdr->saddr;
1361 daddr = &iphdr->daddr;
1362
1363 if (ipv6_addr_loopback(daddr) ||
1364 ipv6_addr_is_multicast(daddr))
1365 goto out;
1366
1367 msg = (struct nd_msg *)skb_transport_header(skb);
1368 if (msg->icmph.icmp6_code != 0 ||
1369 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1370 goto out;
1371
1372 n = neigh_lookup(ipv6_stub->nd_tbl, daddr, dev);
1373
1374 if (n) {
1375 struct vxlan_fdb *f;
1376
1377 if (!(n->nud_state & NUD_CONNECTED)) {
1378 neigh_release(n);
1379 goto out;
1380 }
1381
1382 f = vxlan_find_mac(vxlan, n->ha);
1383 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1384 /* bridge-local neighbor */
1385 neigh_release(n);
1386 goto out;
1387 }
1388
1389 ipv6_stub->ndisc_send_na(dev, n, saddr, &msg->target,
1390 !!in6_dev->cnf.forwarding,
1391 true, false, false);
1392 neigh_release(n);
1393 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1394 ipa.sin6.sin6_addr = *daddr;
1395 ipa.sa.sa_family = AF_INET6;
1396 vxlan_ip_miss(dev, &ipa);
1397 }
1398
1399 out:
1400 consume_skb(skb);
1401 return NETDEV_TX_OK;
1402 }
1403 #endif
1404
1405 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1406 {
1407 struct vxlan_dev *vxlan = netdev_priv(dev);
1408 struct neighbour *n;
1409
1410 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1411 return false;
1412
1413 n = NULL;
1414 switch (ntohs(eth_hdr(skb)->h_proto)) {
1415 case ETH_P_IP:
1416 {
1417 struct iphdr *pip;
1418
1419 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1420 return false;
1421 pip = ip_hdr(skb);
1422 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1423 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1424 union vxlan_addr ipa = {
1425 .sin.sin_addr.s_addr = pip->daddr,
1426 .sa.sa_family = AF_INET,
1427 };
1428
1429 vxlan_ip_miss(dev, &ipa);
1430 return false;
1431 }
1432
1433 break;
1434 }
1435 #if IS_ENABLED(CONFIG_IPV6)
1436 case ETH_P_IPV6:
1437 {
1438 struct ipv6hdr *pip6;
1439
1440 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1441 return false;
1442 pip6 = ipv6_hdr(skb);
1443 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1444 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1445 union vxlan_addr ipa = {
1446 .sin6.sin6_addr = pip6->daddr,
1447 .sa.sa_family = AF_INET6,
1448 };
1449
1450 vxlan_ip_miss(dev, &ipa);
1451 return false;
1452 }
1453
1454 break;
1455 }
1456 #endif
1457 default:
1458 return false;
1459 }
1460
1461 if (n) {
1462 bool diff;
1463
1464 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1465 if (diff) {
1466 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1467 dev->addr_len);
1468 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1469 }
1470 neigh_release(n);
1471 return diff;
1472 }
1473
1474 return false;
1475 }
1476
1477 /* Compute source port for outgoing packet
1478 * first choice to use L4 flow hash since it will spread
1479 * better and maybe available from hardware
1480 * secondary choice is to use jhash on the Ethernet header
1481 */
1482 __be16 vxlan_src_port(__u16 port_min, __u16 port_max, struct sk_buff *skb)
1483 {
1484 unsigned int range = (port_max - port_min) + 1;
1485 u32 hash;
1486
1487 hash = skb_get_hash(skb);
1488 if (!hash)
1489 hash = jhash(skb->data, 2 * ETH_ALEN,
1490 (__force u32) skb->protocol);
1491
1492 return htons((((u64) hash * range) >> 32) + port_min);
1493 }
1494 EXPORT_SYMBOL_GPL(vxlan_src_port);
1495
1496 static int handle_offloads(struct sk_buff *skb)
1497 {
1498 if (skb_is_gso(skb)) {
1499 int err = skb_unclone(skb, GFP_ATOMIC);
1500 if (unlikely(err))
1501 return err;
1502
1503 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
1504 } else if (skb->ip_summed != CHECKSUM_PARTIAL)
1505 skb->ip_summed = CHECKSUM_NONE;
1506
1507 return 0;
1508 }
1509
1510 #if IS_ENABLED(CONFIG_IPV6)
1511 static int vxlan6_xmit_skb(struct vxlan_sock *vs,
1512 struct dst_entry *dst, struct sk_buff *skb,
1513 struct net_device *dev, struct in6_addr *saddr,
1514 struct in6_addr *daddr, __u8 prio, __u8 ttl,
1515 __be16 src_port, __be16 dst_port, __be32 vni)
1516 {
1517 struct ipv6hdr *ip6h;
1518 struct vxlanhdr *vxh;
1519 struct udphdr *uh;
1520 int min_headroom;
1521 int err;
1522
1523 if (!skb->encapsulation) {
1524 skb_reset_inner_headers(skb);
1525 skb->encapsulation = 1;
1526 }
1527
1528 skb_scrub_packet(skb, false);
1529
1530 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1531 + VXLAN_HLEN + sizeof(struct ipv6hdr)
1532 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1533
1534 /* Need space for new headers (invalidates iph ptr) */
1535 err = skb_cow_head(skb, min_headroom);
1536 if (unlikely(err))
1537 return err;
1538
1539 if (vlan_tx_tag_present(skb)) {
1540 if (WARN_ON(!__vlan_put_tag(skb,
1541 skb->vlan_proto,
1542 vlan_tx_tag_get(skb))))
1543 return -ENOMEM;
1544
1545 skb->vlan_tci = 0;
1546 }
1547
1548 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1549 vxh->vx_flags = htonl(VXLAN_FLAGS);
1550 vxh->vx_vni = vni;
1551
1552 __skb_push(skb, sizeof(*uh));
1553 skb_reset_transport_header(skb);
1554 uh = udp_hdr(skb);
1555
1556 uh->dest = dst_port;
1557 uh->source = src_port;
1558
1559 uh->len = htons(skb->len);
1560 uh->check = 0;
1561
1562 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1563 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1564 IPSKB_REROUTED);
1565 skb_dst_set(skb, dst);
1566
1567 if (!skb_is_gso(skb) && !(dst->dev->features & NETIF_F_IPV6_CSUM)) {
1568 __wsum csum = skb_checksum(skb, 0, skb->len, 0);
1569 skb->ip_summed = CHECKSUM_UNNECESSARY;
1570 uh->check = csum_ipv6_magic(saddr, daddr, skb->len,
1571 IPPROTO_UDP, csum);
1572 if (uh->check == 0)
1573 uh->check = CSUM_MANGLED_0;
1574 } else {
1575 skb->ip_summed = CHECKSUM_PARTIAL;
1576 skb->csum_start = skb_transport_header(skb) - skb->head;
1577 skb->csum_offset = offsetof(struct udphdr, check);
1578 uh->check = ~csum_ipv6_magic(saddr, daddr,
1579 skb->len, IPPROTO_UDP, 0);
1580 }
1581
1582 __skb_push(skb, sizeof(*ip6h));
1583 skb_reset_network_header(skb);
1584 ip6h = ipv6_hdr(skb);
1585 ip6h->version = 6;
1586 ip6h->priority = prio;
1587 ip6h->flow_lbl[0] = 0;
1588 ip6h->flow_lbl[1] = 0;
1589 ip6h->flow_lbl[2] = 0;
1590 ip6h->payload_len = htons(skb->len);
1591 ip6h->nexthdr = IPPROTO_UDP;
1592 ip6h->hop_limit = ttl;
1593 ip6h->daddr = *daddr;
1594 ip6h->saddr = *saddr;
1595
1596 err = handle_offloads(skb);
1597 if (err)
1598 return err;
1599
1600 ip6tunnel_xmit(skb, dev);
1601 return 0;
1602 }
1603 #endif
1604
1605 int vxlan_xmit_skb(struct vxlan_sock *vs,
1606 struct rtable *rt, struct sk_buff *skb,
1607 __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1608 __be16 src_port, __be16 dst_port, __be32 vni)
1609 {
1610 struct vxlanhdr *vxh;
1611 struct udphdr *uh;
1612 int min_headroom;
1613 int err;
1614
1615 if (!skb->encapsulation) {
1616 skb_reset_inner_headers(skb);
1617 skb->encapsulation = 1;
1618 }
1619
1620 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1621 + VXLAN_HLEN + sizeof(struct iphdr)
1622 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1623
1624 /* Need space for new headers (invalidates iph ptr) */
1625 err = skb_cow_head(skb, min_headroom);
1626 if (unlikely(err))
1627 return err;
1628
1629 if (vlan_tx_tag_present(skb)) {
1630 if (WARN_ON(!__vlan_put_tag(skb,
1631 skb->vlan_proto,
1632 vlan_tx_tag_get(skb))))
1633 return -ENOMEM;
1634
1635 skb->vlan_tci = 0;
1636 }
1637
1638 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1639 vxh->vx_flags = htonl(VXLAN_FLAGS);
1640 vxh->vx_vni = vni;
1641
1642 __skb_push(skb, sizeof(*uh));
1643 skb_reset_transport_header(skb);
1644 uh = udp_hdr(skb);
1645
1646 uh->dest = dst_port;
1647 uh->source = src_port;
1648
1649 uh->len = htons(skb->len);
1650 uh->check = 0;
1651
1652 err = handle_offloads(skb);
1653 if (err)
1654 return err;
1655
1656 return iptunnel_xmit(rt, skb, src, dst, IPPROTO_UDP, tos, ttl, df,
1657 false);
1658 }
1659 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1660
1661 /* Bypass encapsulation if the destination is local */
1662 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1663 struct vxlan_dev *dst_vxlan)
1664 {
1665 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1666 union vxlan_addr loopback;
1667 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1668
1669 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1670 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1671 skb->pkt_type = PACKET_HOST;
1672 skb->encapsulation = 0;
1673 skb->dev = dst_vxlan->dev;
1674 __skb_pull(skb, skb_network_offset(skb));
1675
1676 if (remote_ip->sa.sa_family == AF_INET) {
1677 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1678 loopback.sa.sa_family = AF_INET;
1679 #if IS_ENABLED(CONFIG_IPV6)
1680 } else {
1681 loopback.sin6.sin6_addr = in6addr_loopback;
1682 loopback.sa.sa_family = AF_INET6;
1683 #endif
1684 }
1685
1686 if (dst_vxlan->flags & VXLAN_F_LEARN)
1687 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1688
1689 u64_stats_update_begin(&tx_stats->syncp);
1690 tx_stats->tx_packets++;
1691 tx_stats->tx_bytes += skb->len;
1692 u64_stats_update_end(&tx_stats->syncp);
1693
1694 if (netif_rx(skb) == NET_RX_SUCCESS) {
1695 u64_stats_update_begin(&rx_stats->syncp);
1696 rx_stats->rx_packets++;
1697 rx_stats->rx_bytes += skb->len;
1698 u64_stats_update_end(&rx_stats->syncp);
1699 } else {
1700 skb->dev->stats.rx_dropped++;
1701 }
1702 }
1703
1704 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1705 struct vxlan_rdst *rdst, bool did_rsc)
1706 {
1707 struct vxlan_dev *vxlan = netdev_priv(dev);
1708 struct rtable *rt = NULL;
1709 const struct iphdr *old_iph;
1710 struct flowi4 fl4;
1711 union vxlan_addr *dst;
1712 __be16 src_port = 0, dst_port;
1713 u32 vni;
1714 __be16 df = 0;
1715 __u8 tos, ttl;
1716 int err;
1717
1718 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1719 vni = rdst->remote_vni;
1720 dst = &rdst->remote_ip;
1721
1722 if (vxlan_addr_any(dst)) {
1723 if (did_rsc) {
1724 /* short-circuited back to local bridge */
1725 vxlan_encap_bypass(skb, vxlan, vxlan);
1726 return;
1727 }
1728 goto drop;
1729 }
1730
1731 old_iph = ip_hdr(skb);
1732
1733 ttl = vxlan->ttl;
1734 if (!ttl && vxlan_addr_multicast(dst))
1735 ttl = 1;
1736
1737 tos = vxlan->tos;
1738 if (tos == 1)
1739 tos = ip_tunnel_get_dsfield(old_iph, skb);
1740
1741 src_port = vxlan_src_port(vxlan->port_min, vxlan->port_max, skb);
1742
1743 if (dst->sa.sa_family == AF_INET) {
1744 memset(&fl4, 0, sizeof(fl4));
1745 fl4.flowi4_oif = rdst->remote_ifindex;
1746 fl4.flowi4_tos = RT_TOS(tos);
1747 fl4.daddr = dst->sin.sin_addr.s_addr;
1748 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1749
1750 rt = ip_route_output_key(dev_net(dev), &fl4);
1751 if (IS_ERR(rt)) {
1752 netdev_dbg(dev, "no route to %pI4\n",
1753 &dst->sin.sin_addr.s_addr);
1754 dev->stats.tx_carrier_errors++;
1755 goto tx_error;
1756 }
1757
1758 if (rt->dst.dev == dev) {
1759 netdev_dbg(dev, "circular route to %pI4\n",
1760 &dst->sin.sin_addr.s_addr);
1761 dev->stats.collisions++;
1762 goto rt_tx_error;
1763 }
1764
1765 /* Bypass encapsulation if the destination is local */
1766 if (rt->rt_flags & RTCF_LOCAL &&
1767 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1768 struct vxlan_dev *dst_vxlan;
1769
1770 ip_rt_put(rt);
1771 dst_vxlan = vxlan_find_vni(dev_net(dev), vni, dst_port);
1772 if (!dst_vxlan)
1773 goto tx_error;
1774 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1775 return;
1776 }
1777
1778 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1779 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1780
1781 err = vxlan_xmit_skb(vxlan->vn_sock, rt, skb,
1782 fl4.saddr, dst->sin.sin_addr.s_addr,
1783 tos, ttl, df, src_port, dst_port,
1784 htonl(vni << 8));
1785
1786 if (err < 0)
1787 goto rt_tx_error;
1788 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1789 #if IS_ENABLED(CONFIG_IPV6)
1790 } else {
1791 struct sock *sk = vxlan->vn_sock->sock->sk;
1792 struct dst_entry *ndst;
1793 struct flowi6 fl6;
1794 u32 flags;
1795
1796 memset(&fl6, 0, sizeof(fl6));
1797 fl6.flowi6_oif = rdst->remote_ifindex;
1798 fl6.daddr = dst->sin6.sin6_addr;
1799 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
1800 fl6.flowi6_proto = IPPROTO_UDP;
1801
1802 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
1803 netdev_dbg(dev, "no route to %pI6\n",
1804 &dst->sin6.sin6_addr);
1805 dev->stats.tx_carrier_errors++;
1806 goto tx_error;
1807 }
1808
1809 if (ndst->dev == dev) {
1810 netdev_dbg(dev, "circular route to %pI6\n",
1811 &dst->sin6.sin6_addr);
1812 dst_release(ndst);
1813 dev->stats.collisions++;
1814 goto tx_error;
1815 }
1816
1817 /* Bypass encapsulation if the destination is local */
1818 flags = ((struct rt6_info *)ndst)->rt6i_flags;
1819 if (flags & RTF_LOCAL &&
1820 !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1821 struct vxlan_dev *dst_vxlan;
1822
1823 dst_release(ndst);
1824 dst_vxlan = vxlan_find_vni(dev_net(dev), vni, dst_port);
1825 if (!dst_vxlan)
1826 goto tx_error;
1827 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1828 return;
1829 }
1830
1831 ttl = ttl ? : ip6_dst_hoplimit(ndst);
1832
1833 err = vxlan6_xmit_skb(vxlan->vn_sock, ndst, skb,
1834 dev, &fl6.saddr, &fl6.daddr, 0, ttl,
1835 src_port, dst_port, htonl(vni << 8));
1836 #endif
1837 }
1838
1839 return;
1840
1841 drop:
1842 dev->stats.tx_dropped++;
1843 goto tx_free;
1844
1845 rt_tx_error:
1846 ip_rt_put(rt);
1847 tx_error:
1848 dev->stats.tx_errors++;
1849 tx_free:
1850 dev_kfree_skb(skb);
1851 }
1852
1853 /* Transmit local packets over Vxlan
1854 *
1855 * Outer IP header inherits ECN and DF from inner header.
1856 * Outer UDP destination is the VXLAN assigned port.
1857 * source port is based on hash of flow
1858 */
1859 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1860 {
1861 struct vxlan_dev *vxlan = netdev_priv(dev);
1862 struct ethhdr *eth;
1863 bool did_rsc = false;
1864 struct vxlan_rdst *rdst, *fdst = NULL;
1865 struct vxlan_fdb *f;
1866
1867 skb_reset_mac_header(skb);
1868 eth = eth_hdr(skb);
1869
1870 if ((vxlan->flags & VXLAN_F_PROXY)) {
1871 if (ntohs(eth->h_proto) == ETH_P_ARP)
1872 return arp_reduce(dev, skb);
1873 #if IS_ENABLED(CONFIG_IPV6)
1874 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
1875 skb->len >= sizeof(struct ipv6hdr) + sizeof(struct nd_msg) &&
1876 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
1877 struct nd_msg *msg;
1878
1879 msg = (struct nd_msg *)skb_transport_header(skb);
1880 if (msg->icmph.icmp6_code == 0 &&
1881 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
1882 return neigh_reduce(dev, skb);
1883 }
1884 #endif
1885 }
1886
1887 f = vxlan_find_mac(vxlan, eth->h_dest);
1888 did_rsc = false;
1889
1890 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1891 (ntohs(eth->h_proto) == ETH_P_IP ||
1892 ntohs(eth->h_proto) == ETH_P_IPV6)) {
1893 did_rsc = route_shortcircuit(dev, skb);
1894 if (did_rsc)
1895 f = vxlan_find_mac(vxlan, eth->h_dest);
1896 }
1897
1898 if (f == NULL) {
1899 f = vxlan_find_mac(vxlan, all_zeros_mac);
1900 if (f == NULL) {
1901 if ((vxlan->flags & VXLAN_F_L2MISS) &&
1902 !is_multicast_ether_addr(eth->h_dest))
1903 vxlan_fdb_miss(vxlan, eth->h_dest);
1904
1905 dev->stats.tx_dropped++;
1906 kfree_skb(skb);
1907 return NETDEV_TX_OK;
1908 }
1909 }
1910
1911 list_for_each_entry_rcu(rdst, &f->remotes, list) {
1912 struct sk_buff *skb1;
1913
1914 if (!fdst) {
1915 fdst = rdst;
1916 continue;
1917 }
1918 skb1 = skb_clone(skb, GFP_ATOMIC);
1919 if (skb1)
1920 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1921 }
1922
1923 if (fdst)
1924 vxlan_xmit_one(skb, dev, fdst, did_rsc);
1925 else
1926 kfree_skb(skb);
1927 return NETDEV_TX_OK;
1928 }
1929
1930 /* Walk the forwarding table and purge stale entries */
1931 static void vxlan_cleanup(unsigned long arg)
1932 {
1933 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1934 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1935 unsigned int h;
1936
1937 if (!netif_running(vxlan->dev))
1938 return;
1939
1940 spin_lock_bh(&vxlan->hash_lock);
1941 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1942 struct hlist_node *p, *n;
1943 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1944 struct vxlan_fdb *f
1945 = container_of(p, struct vxlan_fdb, hlist);
1946 unsigned long timeout;
1947
1948 if (f->state & NUD_PERMANENT)
1949 continue;
1950
1951 timeout = f->used + vxlan->age_interval * HZ;
1952 if (time_before_eq(timeout, jiffies)) {
1953 netdev_dbg(vxlan->dev,
1954 "garbage collect %pM\n",
1955 f->eth_addr);
1956 f->state = NUD_STALE;
1957 vxlan_fdb_destroy(vxlan, f);
1958 } else if (time_before(timeout, next_timer))
1959 next_timer = timeout;
1960 }
1961 }
1962 spin_unlock_bh(&vxlan->hash_lock);
1963
1964 mod_timer(&vxlan->age_timer, next_timer);
1965 }
1966
1967 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
1968 {
1969 __u32 vni = vxlan->default_dst.remote_vni;
1970
1971 vxlan->vn_sock = vs;
1972 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
1973 }
1974
1975 /* Setup stats when device is created */
1976 static int vxlan_init(struct net_device *dev)
1977 {
1978 struct vxlan_dev *vxlan = netdev_priv(dev);
1979 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
1980 struct vxlan_sock *vs;
1981
1982 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1983 if (!dev->tstats)
1984 return -ENOMEM;
1985
1986 spin_lock(&vn->sock_lock);
1987 vs = vxlan_find_sock(dev_net(dev), vxlan->dst_port);
1988 if (vs) {
1989 /* If we have a socket with same port already, reuse it */
1990 atomic_inc(&vs->refcnt);
1991 vxlan_vs_add_dev(vs, vxlan);
1992 } else {
1993 /* otherwise make new socket outside of RTNL */
1994 dev_hold(dev);
1995 queue_work(vxlan_wq, &vxlan->sock_work);
1996 }
1997 spin_unlock(&vn->sock_lock);
1998
1999 return 0;
2000 }
2001
2002 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2003 {
2004 struct vxlan_fdb *f;
2005
2006 spin_lock_bh(&vxlan->hash_lock);
2007 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2008 if (f)
2009 vxlan_fdb_destroy(vxlan, f);
2010 spin_unlock_bh(&vxlan->hash_lock);
2011 }
2012
2013 static void vxlan_uninit(struct net_device *dev)
2014 {
2015 struct vxlan_dev *vxlan = netdev_priv(dev);
2016 struct vxlan_sock *vs = vxlan->vn_sock;
2017
2018 vxlan_fdb_delete_default(vxlan);
2019
2020 if (vs)
2021 vxlan_sock_release(vs);
2022 free_percpu(dev->tstats);
2023 }
2024
2025 /* Start ageing timer and join group when device is brought up */
2026 static int vxlan_open(struct net_device *dev)
2027 {
2028 struct vxlan_dev *vxlan = netdev_priv(dev);
2029 struct vxlan_sock *vs = vxlan->vn_sock;
2030
2031 /* socket hasn't been created */
2032 if (!vs)
2033 return -ENOTCONN;
2034
2035 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2036 vxlan_sock_hold(vs);
2037 dev_hold(dev);
2038 queue_work(vxlan_wq, &vxlan->igmp_join);
2039 }
2040
2041 if (vxlan->age_interval)
2042 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2043
2044 return 0;
2045 }
2046
2047 /* Purge the forwarding table */
2048 static void vxlan_flush(struct vxlan_dev *vxlan)
2049 {
2050 unsigned int h;
2051
2052 spin_lock_bh(&vxlan->hash_lock);
2053 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2054 struct hlist_node *p, *n;
2055 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2056 struct vxlan_fdb *f
2057 = container_of(p, struct vxlan_fdb, hlist);
2058 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2059 if (!is_zero_ether_addr(f->eth_addr))
2060 vxlan_fdb_destroy(vxlan, f);
2061 }
2062 }
2063 spin_unlock_bh(&vxlan->hash_lock);
2064 }
2065
2066 /* Cleanup timer and forwarding table on shutdown */
2067 static int vxlan_stop(struct net_device *dev)
2068 {
2069 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2070 struct vxlan_dev *vxlan = netdev_priv(dev);
2071 struct vxlan_sock *vs = vxlan->vn_sock;
2072
2073 if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2074 !vxlan_group_used(vn, vxlan)) {
2075 vxlan_sock_hold(vs);
2076 dev_hold(dev);
2077 queue_work(vxlan_wq, &vxlan->igmp_leave);
2078 }
2079
2080 del_timer_sync(&vxlan->age_timer);
2081
2082 vxlan_flush(vxlan);
2083
2084 return 0;
2085 }
2086
2087 /* Stub, nothing needs to be done. */
2088 static void vxlan_set_multicast_list(struct net_device *dev)
2089 {
2090 }
2091
2092 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2093 {
2094 struct vxlan_dev *vxlan = netdev_priv(dev);
2095 struct vxlan_rdst *dst = &vxlan->default_dst;
2096 struct net_device *lowerdev;
2097 int max_mtu;
2098
2099 lowerdev = __dev_get_by_index(dev_net(dev), dst->remote_ifindex);
2100 if (lowerdev == NULL)
2101 return eth_change_mtu(dev, new_mtu);
2102
2103 if (dst->remote_ip.sa.sa_family == AF_INET6)
2104 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2105 else
2106 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2107
2108 if (new_mtu < 68 || new_mtu > max_mtu)
2109 return -EINVAL;
2110
2111 dev->mtu = new_mtu;
2112 return 0;
2113 }
2114
2115 static const struct net_device_ops vxlan_netdev_ops = {
2116 .ndo_init = vxlan_init,
2117 .ndo_uninit = vxlan_uninit,
2118 .ndo_open = vxlan_open,
2119 .ndo_stop = vxlan_stop,
2120 .ndo_start_xmit = vxlan_xmit,
2121 .ndo_get_stats64 = ip_tunnel_get_stats64,
2122 .ndo_set_rx_mode = vxlan_set_multicast_list,
2123 .ndo_change_mtu = vxlan_change_mtu,
2124 .ndo_validate_addr = eth_validate_addr,
2125 .ndo_set_mac_address = eth_mac_addr,
2126 .ndo_fdb_add = vxlan_fdb_add,
2127 .ndo_fdb_del = vxlan_fdb_delete,
2128 .ndo_fdb_dump = vxlan_fdb_dump,
2129 };
2130
2131 /* Info for udev, that this is a virtual tunnel endpoint */
2132 static struct device_type vxlan_type = {
2133 .name = "vxlan",
2134 };
2135
2136 /* Calls the ndo_add_vxlan_port of the caller in order to
2137 * supply the listening VXLAN udp ports. Callers are expected
2138 * to implement the ndo_add_vxlan_port.
2139 */
2140 void vxlan_get_rx_port(struct net_device *dev)
2141 {
2142 struct vxlan_sock *vs;
2143 struct net *net = dev_net(dev);
2144 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2145 sa_family_t sa_family;
2146 __be16 port;
2147 unsigned int i;
2148
2149 spin_lock(&vn->sock_lock);
2150 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2151 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2152 port = inet_sk(vs->sock->sk)->inet_sport;
2153 sa_family = vs->sock->sk->sk_family;
2154 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2155 port);
2156 }
2157 }
2158 spin_unlock(&vn->sock_lock);
2159 }
2160 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2161
2162 /* Initialize the device structure. */
2163 static void vxlan_setup(struct net_device *dev)
2164 {
2165 struct vxlan_dev *vxlan = netdev_priv(dev);
2166 unsigned int h;
2167 int low, high;
2168
2169 eth_hw_addr_random(dev);
2170 ether_setup(dev);
2171 if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2172 dev->hard_header_len = ETH_HLEN + VXLAN6_HEADROOM;
2173 else
2174 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
2175
2176 dev->netdev_ops = &vxlan_netdev_ops;
2177 dev->destructor = free_netdev;
2178 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2179
2180 dev->tx_queue_len = 0;
2181 dev->features |= NETIF_F_LLTX;
2182 dev->features |= NETIF_F_NETNS_LOCAL;
2183 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2184 dev->features |= NETIF_F_RXCSUM;
2185 dev->features |= NETIF_F_GSO_SOFTWARE;
2186
2187 dev->vlan_features = dev->features;
2188 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2189 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2190 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2191 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2192 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
2193 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2194
2195 INIT_LIST_HEAD(&vxlan->next);
2196 spin_lock_init(&vxlan->hash_lock);
2197 INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join);
2198 INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave);
2199 INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
2200
2201 init_timer_deferrable(&vxlan->age_timer);
2202 vxlan->age_timer.function = vxlan_cleanup;
2203 vxlan->age_timer.data = (unsigned long) vxlan;
2204
2205 inet_get_local_port_range(dev_net(dev), &low, &high);
2206 vxlan->port_min = low;
2207 vxlan->port_max = high;
2208 vxlan->dst_port = htons(vxlan_port);
2209
2210 vxlan->dev = dev;
2211
2212 for (h = 0; h < FDB_HASH_SIZE; ++h)
2213 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2214 }
2215
2216 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2217 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2218 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2219 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2220 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2221 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2222 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2223 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2224 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2225 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2226 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2227 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2228 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2229 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2230 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2231 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2232 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2233 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2234 };
2235
2236 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2237 {
2238 if (tb[IFLA_ADDRESS]) {
2239 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2240 pr_debug("invalid link address (not ethernet)\n");
2241 return -EINVAL;
2242 }
2243
2244 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2245 pr_debug("invalid all zero ethernet address\n");
2246 return -EADDRNOTAVAIL;
2247 }
2248 }
2249
2250 if (!data)
2251 return -EINVAL;
2252
2253 if (data[IFLA_VXLAN_ID]) {
2254 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2255 if (id >= VXLAN_VID_MASK)
2256 return -ERANGE;
2257 }
2258
2259 if (data[IFLA_VXLAN_PORT_RANGE]) {
2260 const struct ifla_vxlan_port_range *p
2261 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2262
2263 if (ntohs(p->high) < ntohs(p->low)) {
2264 pr_debug("port range %u .. %u not valid\n",
2265 ntohs(p->low), ntohs(p->high));
2266 return -EINVAL;
2267 }
2268 }
2269
2270 return 0;
2271 }
2272
2273 static void vxlan_get_drvinfo(struct net_device *netdev,
2274 struct ethtool_drvinfo *drvinfo)
2275 {
2276 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2277 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2278 }
2279
2280 static const struct ethtool_ops vxlan_ethtool_ops = {
2281 .get_drvinfo = vxlan_get_drvinfo,
2282 .get_link = ethtool_op_get_link,
2283 };
2284
2285 static void vxlan_del_work(struct work_struct *work)
2286 {
2287 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2288
2289 sk_release_kernel(vs->sock->sk);
2290 kfree_rcu(vs, rcu);
2291 }
2292
2293 #if IS_ENABLED(CONFIG_IPV6)
2294 /* Create UDP socket for encapsulation receive. AF_INET6 socket
2295 * could be used for both IPv4 and IPv6 communications, but
2296 * users may set bindv6only=1.
2297 */
2298 static struct socket *create_v6_sock(struct net *net, __be16 port)
2299 {
2300 struct sock *sk;
2301 struct socket *sock;
2302 struct sockaddr_in6 vxlan_addr = {
2303 .sin6_family = AF_INET6,
2304 .sin6_port = port,
2305 };
2306 int rc, val = 1;
2307
2308 rc = sock_create_kern(AF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
2309 if (rc < 0) {
2310 pr_debug("UDPv6 socket create failed\n");
2311 return ERR_PTR(rc);
2312 }
2313
2314 /* Put in proper namespace */
2315 sk = sock->sk;
2316 sk_change_net(sk, net);
2317
2318 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
2319 (char *)&val, sizeof(val));
2320 rc = kernel_bind(sock, (struct sockaddr *)&vxlan_addr,
2321 sizeof(struct sockaddr_in6));
2322 if (rc < 0) {
2323 pr_debug("bind for UDPv6 socket %pI6:%u (%d)\n",
2324 &vxlan_addr.sin6_addr, ntohs(vxlan_addr.sin6_port), rc);
2325 sk_release_kernel(sk);
2326 return ERR_PTR(rc);
2327 }
2328 /* At this point, IPv6 module should have been loaded in
2329 * sock_create_kern().
2330 */
2331 BUG_ON(!ipv6_stub);
2332
2333 /* Disable multicast loopback */
2334 inet_sk(sk)->mc_loop = 0;
2335 return sock;
2336 }
2337
2338 #else
2339
2340 static struct socket *create_v6_sock(struct net *net, __be16 port)
2341 {
2342 return ERR_PTR(-EPFNOSUPPORT);
2343 }
2344 #endif
2345
2346 static struct socket *create_v4_sock(struct net *net, __be16 port)
2347 {
2348 struct sock *sk;
2349 struct socket *sock;
2350 struct sockaddr_in vxlan_addr = {
2351 .sin_family = AF_INET,
2352 .sin_addr.s_addr = htonl(INADDR_ANY),
2353 .sin_port = port,
2354 };
2355 int rc;
2356
2357 /* Create UDP socket for encapsulation receive. */
2358 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
2359 if (rc < 0) {
2360 pr_debug("UDP socket create failed\n");
2361 return ERR_PTR(rc);
2362 }
2363
2364 /* Put in proper namespace */
2365 sk = sock->sk;
2366 sk_change_net(sk, net);
2367
2368 rc = kernel_bind(sock, (struct sockaddr *) &vxlan_addr,
2369 sizeof(vxlan_addr));
2370 if (rc < 0) {
2371 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
2372 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
2373 sk_release_kernel(sk);
2374 return ERR_PTR(rc);
2375 }
2376
2377 /* Disable multicast loopback */
2378 inet_sk(sk)->mc_loop = 0;
2379 return sock;
2380 }
2381
2382 /* Create new listen socket if needed */
2383 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2384 vxlan_rcv_t *rcv, void *data, bool ipv6)
2385 {
2386 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2387 struct vxlan_sock *vs;
2388 struct socket *sock;
2389 struct sock *sk;
2390 unsigned int h;
2391
2392 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2393 if (!vs)
2394 return ERR_PTR(-ENOMEM);
2395
2396 for (h = 0; h < VNI_HASH_SIZE; ++h)
2397 INIT_HLIST_HEAD(&vs->vni_list[h]);
2398
2399 INIT_WORK(&vs->del_work, vxlan_del_work);
2400
2401 if (ipv6)
2402 sock = create_v6_sock(net, port);
2403 else
2404 sock = create_v4_sock(net, port);
2405 if (IS_ERR(sock)) {
2406 kfree(vs);
2407 return ERR_CAST(sock);
2408 }
2409
2410 vs->sock = sock;
2411 sk = sock->sk;
2412 atomic_set(&vs->refcnt, 1);
2413 vs->rcv = rcv;
2414 vs->data = data;
2415 rcu_assign_sk_user_data(vs->sock->sk, vs);
2416
2417 /* Initialize the vxlan udp offloads structure */
2418 vs->udp_offloads.port = port;
2419 vs->udp_offloads.callbacks.gro_receive = vxlan_gro_receive;
2420 vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2421
2422 spin_lock(&vn->sock_lock);
2423 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2424 vxlan_notify_add_rx_port(vs);
2425 spin_unlock(&vn->sock_lock);
2426
2427 /* Mark socket as an encapsulation socket. */
2428 udp_sk(sk)->encap_type = 1;
2429 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
2430 #if IS_ENABLED(CONFIG_IPV6)
2431 if (ipv6)
2432 ipv6_stub->udpv6_encap_enable();
2433 else
2434 #endif
2435 udp_encap_enable();
2436
2437 return vs;
2438 }
2439
2440 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2441 vxlan_rcv_t *rcv, void *data,
2442 bool no_share, bool ipv6)
2443 {
2444 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2445 struct vxlan_sock *vs;
2446
2447 vs = vxlan_socket_create(net, port, rcv, data, ipv6);
2448 if (!IS_ERR(vs))
2449 return vs;
2450
2451 if (no_share) /* Return error if sharing is not allowed. */
2452 return vs;
2453
2454 spin_lock(&vn->sock_lock);
2455 vs = vxlan_find_sock(net, port);
2456 if (vs) {
2457 if (vs->rcv == rcv)
2458 atomic_inc(&vs->refcnt);
2459 else
2460 vs = ERR_PTR(-EBUSY);
2461 }
2462 spin_unlock(&vn->sock_lock);
2463
2464 if (!vs)
2465 vs = ERR_PTR(-EINVAL);
2466
2467 return vs;
2468 }
2469 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2470
2471 /* Scheduled at device creation to bind to a socket */
2472 static void vxlan_sock_work(struct work_struct *work)
2473 {
2474 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2475 struct net *net = dev_net(vxlan->dev);
2476 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2477 __be16 port = vxlan->dst_port;
2478 struct vxlan_sock *nvs;
2479
2480 nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags & VXLAN_F_IPV6);
2481 spin_lock(&vn->sock_lock);
2482 if (!IS_ERR(nvs))
2483 vxlan_vs_add_dev(nvs, vxlan);
2484 spin_unlock(&vn->sock_lock);
2485
2486 dev_put(vxlan->dev);
2487 }
2488
2489 static int vxlan_newlink(struct net *net, struct net_device *dev,
2490 struct nlattr *tb[], struct nlattr *data[])
2491 {
2492 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2493 struct vxlan_dev *vxlan = netdev_priv(dev);
2494 struct vxlan_rdst *dst = &vxlan->default_dst;
2495 __u32 vni;
2496 int err;
2497 bool use_ipv6 = false;
2498
2499 if (!data[IFLA_VXLAN_ID])
2500 return -EINVAL;
2501
2502 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2503 dst->remote_vni = vni;
2504
2505 if (data[IFLA_VXLAN_GROUP]) {
2506 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2507 dst->remote_ip.sa.sa_family = AF_INET;
2508 } else if (data[IFLA_VXLAN_GROUP6]) {
2509 if (!IS_ENABLED(CONFIG_IPV6))
2510 return -EPFNOSUPPORT;
2511
2512 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2513 sizeof(struct in6_addr));
2514 dst->remote_ip.sa.sa_family = AF_INET6;
2515 use_ipv6 = true;
2516 }
2517
2518 if (data[IFLA_VXLAN_LOCAL]) {
2519 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2520 vxlan->saddr.sa.sa_family = AF_INET;
2521 } else if (data[IFLA_VXLAN_LOCAL6]) {
2522 if (!IS_ENABLED(CONFIG_IPV6))
2523 return -EPFNOSUPPORT;
2524
2525 /* TODO: respect scope id */
2526 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2527 sizeof(struct in6_addr));
2528 vxlan->saddr.sa.sa_family = AF_INET6;
2529 use_ipv6 = true;
2530 }
2531
2532 if (data[IFLA_VXLAN_LINK] &&
2533 (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2534 struct net_device *lowerdev
2535 = __dev_get_by_index(net, dst->remote_ifindex);
2536
2537 if (!lowerdev) {
2538 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2539 return -ENODEV;
2540 }
2541
2542 #if IS_ENABLED(CONFIG_IPV6)
2543 if (use_ipv6) {
2544 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2545 if (idev && idev->cnf.disable_ipv6) {
2546 pr_info("IPv6 is disabled via sysctl\n");
2547 return -EPERM;
2548 }
2549 vxlan->flags |= VXLAN_F_IPV6;
2550 }
2551 #endif
2552
2553 if (!tb[IFLA_MTU])
2554 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2555
2556 /* update header length based on lower device */
2557 dev->hard_header_len = lowerdev->hard_header_len +
2558 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2559 } else if (use_ipv6)
2560 vxlan->flags |= VXLAN_F_IPV6;
2561
2562 if (data[IFLA_VXLAN_TOS])
2563 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2564
2565 if (data[IFLA_VXLAN_TTL])
2566 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2567
2568 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2569 vxlan->flags |= VXLAN_F_LEARN;
2570
2571 if (data[IFLA_VXLAN_AGEING])
2572 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2573 else
2574 vxlan->age_interval = FDB_AGE_DEFAULT;
2575
2576 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2577 vxlan->flags |= VXLAN_F_PROXY;
2578
2579 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2580 vxlan->flags |= VXLAN_F_RSC;
2581
2582 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2583 vxlan->flags |= VXLAN_F_L2MISS;
2584
2585 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2586 vxlan->flags |= VXLAN_F_L3MISS;
2587
2588 if (data[IFLA_VXLAN_LIMIT])
2589 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2590
2591 if (data[IFLA_VXLAN_PORT_RANGE]) {
2592 const struct ifla_vxlan_port_range *p
2593 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2594 vxlan->port_min = ntohs(p->low);
2595 vxlan->port_max = ntohs(p->high);
2596 }
2597
2598 if (data[IFLA_VXLAN_PORT])
2599 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2600
2601 if (vxlan_find_vni(net, vni, vxlan->dst_port)) {
2602 pr_info("duplicate VNI %u\n", vni);
2603 return -EEXIST;
2604 }
2605
2606 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
2607
2608 /* create an fdb entry for a valid default destination */
2609 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2610 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2611 &vxlan->default_dst.remote_ip,
2612 NUD_REACHABLE|NUD_PERMANENT,
2613 NLM_F_EXCL|NLM_F_CREATE,
2614 vxlan->dst_port,
2615 vxlan->default_dst.remote_vni,
2616 vxlan->default_dst.remote_ifindex,
2617 NTF_SELF);
2618 if (err)
2619 return err;
2620 }
2621
2622 err = register_netdevice(dev);
2623 if (err) {
2624 vxlan_fdb_delete_default(vxlan);
2625 return err;
2626 }
2627
2628 list_add(&vxlan->next, &vn->vxlan_list);
2629
2630 return 0;
2631 }
2632
2633 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2634 {
2635 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2636 struct vxlan_dev *vxlan = netdev_priv(dev);
2637
2638 spin_lock(&vn->sock_lock);
2639 if (!hlist_unhashed(&vxlan->hlist))
2640 hlist_del_rcu(&vxlan->hlist);
2641 spin_unlock(&vn->sock_lock);
2642
2643 list_del(&vxlan->next);
2644 unregister_netdevice_queue(dev, head);
2645 }
2646
2647 static size_t vxlan_get_size(const struct net_device *dev)
2648 {
2649
2650 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
2651 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2652 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2653 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2654 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
2655 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
2656 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
2657 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
2658 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
2659 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
2660 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
2661 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2662 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2663 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2664 nla_total_size(sizeof(__be16))+ /* IFLA_VXLAN_PORT */
2665 0;
2666 }
2667
2668 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2669 {
2670 const struct vxlan_dev *vxlan = netdev_priv(dev);
2671 const struct vxlan_rdst *dst = &vxlan->default_dst;
2672 struct ifla_vxlan_port_range ports = {
2673 .low = htons(vxlan->port_min),
2674 .high = htons(vxlan->port_max),
2675 };
2676
2677 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2678 goto nla_put_failure;
2679
2680 if (!vxlan_addr_any(&dst->remote_ip)) {
2681 if (dst->remote_ip.sa.sa_family == AF_INET) {
2682 if (nla_put_be32(skb, IFLA_VXLAN_GROUP,
2683 dst->remote_ip.sin.sin_addr.s_addr))
2684 goto nla_put_failure;
2685 #if IS_ENABLED(CONFIG_IPV6)
2686 } else {
2687 if (nla_put(skb, IFLA_VXLAN_GROUP6, sizeof(struct in6_addr),
2688 &dst->remote_ip.sin6.sin6_addr))
2689 goto nla_put_failure;
2690 #endif
2691 }
2692 }
2693
2694 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2695 goto nla_put_failure;
2696
2697 if (!vxlan_addr_any(&vxlan->saddr)) {
2698 if (vxlan->saddr.sa.sa_family == AF_INET) {
2699 if (nla_put_be32(skb, IFLA_VXLAN_LOCAL,
2700 vxlan->saddr.sin.sin_addr.s_addr))
2701 goto nla_put_failure;
2702 #if IS_ENABLED(CONFIG_IPV6)
2703 } else {
2704 if (nla_put(skb, IFLA_VXLAN_LOCAL6, sizeof(struct in6_addr),
2705 &vxlan->saddr.sin6.sin6_addr))
2706 goto nla_put_failure;
2707 #endif
2708 }
2709 }
2710
2711 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2712 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2713 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2714 !!(vxlan->flags & VXLAN_F_LEARN)) ||
2715 nla_put_u8(skb, IFLA_VXLAN_PROXY,
2716 !!(vxlan->flags & VXLAN_F_PROXY)) ||
2717 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2718 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2719 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2720 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2721 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2722 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2723 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2724 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port))
2725 goto nla_put_failure;
2726
2727 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2728 goto nla_put_failure;
2729
2730 return 0;
2731
2732 nla_put_failure:
2733 return -EMSGSIZE;
2734 }
2735
2736 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2737 .kind = "vxlan",
2738 .maxtype = IFLA_VXLAN_MAX,
2739 .policy = vxlan_policy,
2740 .priv_size = sizeof(struct vxlan_dev),
2741 .setup = vxlan_setup,
2742 .validate = vxlan_validate,
2743 .newlink = vxlan_newlink,
2744 .dellink = vxlan_dellink,
2745 .get_size = vxlan_get_size,
2746 .fill_info = vxlan_fill_info,
2747 };
2748
2749 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
2750 struct net_device *dev)
2751 {
2752 struct vxlan_dev *vxlan, *next;
2753 LIST_HEAD(list_kill);
2754
2755 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2756 struct vxlan_rdst *dst = &vxlan->default_dst;
2757
2758 /* In case we created vxlan device with carrier
2759 * and we loose the carrier due to module unload
2760 * we also need to remove vxlan device. In other
2761 * cases, it's not necessary and remote_ifindex
2762 * is 0 here, so no matches.
2763 */
2764 if (dst->remote_ifindex == dev->ifindex)
2765 vxlan_dellink(vxlan->dev, &list_kill);
2766 }
2767
2768 unregister_netdevice_many(&list_kill);
2769 }
2770
2771 static int vxlan_lowerdev_event(struct notifier_block *unused,
2772 unsigned long event, void *ptr)
2773 {
2774 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2775 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2776
2777 if (event == NETDEV_UNREGISTER)
2778 vxlan_handle_lowerdev_unregister(vn, dev);
2779
2780 return NOTIFY_DONE;
2781 }
2782
2783 static struct notifier_block vxlan_notifier_block __read_mostly = {
2784 .notifier_call = vxlan_lowerdev_event,
2785 };
2786
2787 static __net_init int vxlan_init_net(struct net *net)
2788 {
2789 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2790 unsigned int h;
2791
2792 INIT_LIST_HEAD(&vn->vxlan_list);
2793 spin_lock_init(&vn->sock_lock);
2794
2795 for (h = 0; h < PORT_HASH_SIZE; ++h)
2796 INIT_HLIST_HEAD(&vn->sock_list[h]);
2797
2798 return 0;
2799 }
2800
2801 static struct pernet_operations vxlan_net_ops = {
2802 .init = vxlan_init_net,
2803 .id = &vxlan_net_id,
2804 .size = sizeof(struct vxlan_net),
2805 };
2806
2807 static int __init vxlan_init_module(void)
2808 {
2809 int rc;
2810
2811 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
2812 if (!vxlan_wq)
2813 return -ENOMEM;
2814
2815 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
2816
2817 rc = register_pernet_subsys(&vxlan_net_ops);
2818 if (rc)
2819 goto out1;
2820
2821 rc = register_netdevice_notifier(&vxlan_notifier_block);
2822 if (rc)
2823 goto out2;
2824
2825 rc = rtnl_link_register(&vxlan_link_ops);
2826 if (rc)
2827 goto out3;
2828
2829 return 0;
2830 out3:
2831 unregister_netdevice_notifier(&vxlan_notifier_block);
2832 out2:
2833 unregister_pernet_subsys(&vxlan_net_ops);
2834 out1:
2835 destroy_workqueue(vxlan_wq);
2836 return rc;
2837 }
2838 late_initcall(vxlan_init_module);
2839
2840 static void __exit vxlan_cleanup_module(void)
2841 {
2842 rtnl_link_unregister(&vxlan_link_ops);
2843 unregister_netdevice_notifier(&vxlan_notifier_block);
2844 destroy_workqueue(vxlan_wq);
2845 unregister_pernet_subsys(&vxlan_net_ops);
2846 /* rcu_barrier() is called by netns */
2847 }
2848 module_exit(vxlan_cleanup_module);
2849
2850 MODULE_LICENSE("GPL");
2851 MODULE_VERSION(VXLAN_VERSION);
2852 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
2853 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
2854 MODULE_ALIAS_RTNL_LINK("vxlan");
Linux kernel - Deliverables
This page took 0.088631 seconds and 5 git commands to generate.