]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/net/vxlan.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'misc' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild
[mirror_ubuntu-bionic-kernel.git] / drivers / net / vxlan.c
1 /*
2 * VXLAN: Virtual eXtensible Local Area Network
3 *
4 * Copyright (c) 2012-2013 Vyatta Inc.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 */
10
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/module.h>
16 #include <linux/errno.h>
17 #include <linux/slab.h>
18 #include <linux/skbuff.h>
19 #include <linux/rculist.h>
20 #include <linux/netdevice.h>
21 #include <linux/in.h>
22 #include <linux/ip.h>
23 #include <linux/udp.h>
24 #include <linux/igmp.h>
25 #include <linux/etherdevice.h>
26 #include <linux/if_ether.h>
27 #include <linux/if_vlan.h>
28 #include <linux/hash.h>
29 #include <linux/ethtool.h>
30 #include <net/arp.h>
31 #include <net/ndisc.h>
32 #include <net/ip.h>
33 #include <net/ip_tunnels.h>
34 #include <net/icmp.h>
35 #include <net/udp.h>
36 #include <net/rtnetlink.h>
37 #include <net/route.h>
38 #include <net/dsfield.h>
39 #include <net/inet_ecn.h>
40 #include <net/net_namespace.h>
41 #include <net/netns/generic.h>
42 #include <net/vxlan.h>
43 #include <net/protocol.h>
44 #if IS_ENABLED(CONFIG_IPV6)
45 #include <net/ipv6.h>
46 #include <net/addrconf.h>
47 #include <net/ip6_tunnel.h>
48 #include <net/ip6_checksum.h>
49 #endif
50
51 #define VXLAN_VERSION "0.1"
52
53 #define PORT_HASH_BITS 8
54 #define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
55 #define VNI_HASH_BITS 10
56 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
57 #define FDB_HASH_BITS 8
58 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
59 #define FDB_AGE_DEFAULT 300 /* 5 min */
60 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
61
62 #define VXLAN_N_VID (1u << 24)
63 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
64 #define VXLAN_HLEN (sizeof(struct udphdr) + sizeof(struct vxlanhdr))
65
66 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
67
68 /* VXLAN protocol header */
69 struct vxlanhdr {
70 __be32 vx_flags;
71 __be32 vx_vni;
72 };
73
74 /* UDP port for VXLAN traffic.
75 * The IANA assigned port is 4789, but the Linux default is 8472
76 * for compatibility with early adopters.
77 */
78 static unsigned short vxlan_port __read_mostly = 8472;
79 module_param_named(udp_port, vxlan_port, ushort, 0444);
80 MODULE_PARM_DESC(udp_port, "Destination UDP port");
81
82 static bool log_ecn_error = true;
83 module_param(log_ecn_error, bool, 0644);
84 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
85
86 static int vxlan_net_id;
87
88 static const u8 all_zeros_mac[ETH_ALEN];
89
90 /* per-network namespace private data for this module */
91 struct vxlan_net {
92 struct list_head vxlan_list;
93 struct hlist_head sock_list[PORT_HASH_SIZE];
94 spinlock_t sock_lock;
95 };
96
97 union vxlan_addr {
98 struct sockaddr_in sin;
99 struct sockaddr_in6 sin6;
100 struct sockaddr sa;
101 };
102
103 struct vxlan_rdst {
104 union vxlan_addr remote_ip;
105 __be16 remote_port;
106 u32 remote_vni;
107 u32 remote_ifindex;
108 struct list_head list;
109 struct rcu_head rcu;
110 };
111
112 /* Forwarding table entry */
113 struct vxlan_fdb {
114 struct hlist_node hlist; /* linked list of entries */
115 struct rcu_head rcu;
116 unsigned long updated; /* jiffies */
117 unsigned long used;
118 struct list_head remotes;
119 u16 state; /* see ndm_state */
120 u8 flags; /* see ndm_flags */
121 u8 eth_addr[ETH_ALEN];
122 };
123
124 /* Pseudo network device */
125 struct vxlan_dev {
126 struct hlist_node hlist; /* vni hash table */
127 struct list_head next; /* vxlan's per namespace list */
128 struct vxlan_sock *vn_sock; /* listening socket */
129 struct net_device *dev;
130 struct net *net; /* netns for packet i/o */
131 struct vxlan_rdst default_dst; /* default destination */
132 union vxlan_addr saddr; /* source address */
133 __be16 dst_port;
134 __u16 port_min; /* source port range */
135 __u16 port_max;
136 __u8 tos; /* TOS override */
137 __u8 ttl;
138 u32 flags; /* VXLAN_F_* in vxlan.h */
139
140 struct work_struct sock_work;
141 struct work_struct igmp_join;
142 struct work_struct igmp_leave;
143
144 unsigned long age_interval;
145 struct timer_list age_timer;
146 spinlock_t hash_lock;
147 unsigned int addrcnt;
148 unsigned int addrmax;
149
150 struct hlist_head fdb_head[FDB_HASH_SIZE];
151 };
152
153 /* salt for hash table */
154 static u32 vxlan_salt __read_mostly;
155 static struct workqueue_struct *vxlan_wq;
156
157 static void vxlan_sock_work(struct work_struct *work);
158
159 #if IS_ENABLED(CONFIG_IPV6)
160 static inline
161 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
162 {
163 if (a->sa.sa_family != b->sa.sa_family)
164 return false;
165 if (a->sa.sa_family == AF_INET6)
166 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
167 else
168 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
169 }
170
171 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
172 {
173 if (ipa->sa.sa_family == AF_INET6)
174 return ipv6_addr_any(&ipa->sin6.sin6_addr);
175 else
176 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
177 }
178
179 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
180 {
181 if (ipa->sa.sa_family == AF_INET6)
182 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
183 else
184 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
185 }
186
187 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
188 {
189 if (nla_len(nla) >= sizeof(struct in6_addr)) {
190 nla_memcpy(&ip->sin6.sin6_addr, nla, sizeof(struct in6_addr));
191 ip->sa.sa_family = AF_INET6;
192 return 0;
193 } else if (nla_len(nla) >= sizeof(__be32)) {
194 ip->sin.sin_addr.s_addr = nla_get_be32(nla);
195 ip->sa.sa_family = AF_INET;
196 return 0;
197 } else {
198 return -EAFNOSUPPORT;
199 }
200 }
201
202 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
203 const union vxlan_addr *ip)
204 {
205 if (ip->sa.sa_family == AF_INET6)
206 return nla_put(skb, attr, sizeof(struct in6_addr), &ip->sin6.sin6_addr);
207 else
208 return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
209 }
210
211 #else /* !CONFIG_IPV6 */
212
213 static inline
214 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
215 {
216 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
217 }
218
219 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
220 {
221 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
222 }
223
224 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
225 {
226 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
227 }
228
229 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
230 {
231 if (nla_len(nla) >= sizeof(struct in6_addr)) {
232 return -EAFNOSUPPORT;
233 } else if (nla_len(nla) >= sizeof(__be32)) {
234 ip->sin.sin_addr.s_addr = nla_get_be32(nla);
235 ip->sa.sa_family = AF_INET;
236 return 0;
237 } else {
238 return -EAFNOSUPPORT;
239 }
240 }
241
242 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
243 const union vxlan_addr *ip)
244 {
245 return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
246 }
247 #endif
248
249 /* Virtual Network hash table head */
250 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
251 {
252 return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
253 }
254
255 /* Socket hash table head */
256 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
257 {
258 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
259
260 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
261 }
262
263 /* First remote destination for a forwarding entry.
264 * Guaranteed to be non-NULL because remotes are never deleted.
265 */
266 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
267 {
268 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
269 }
270
271 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
272 {
273 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
274 }
275
276 /* Find VXLAN socket based on network namespace and UDP port */
277 static struct vxlan_sock *vxlan_find_sock(struct net *net, __be16 port)
278 {
279 struct vxlan_sock *vs;
280
281 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
282 if (inet_sk(vs->sock->sk)->inet_sport == port)
283 return vs;
284 }
285 return NULL;
286 }
287
288 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
289 {
290 struct vxlan_dev *vxlan;
291
292 hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
293 if (vxlan->default_dst.remote_vni == id)
294 return vxlan;
295 }
296
297 return NULL;
298 }
299
300 /* Look up VNI in a per net namespace table */
301 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id, __be16 port)
302 {
303 struct vxlan_sock *vs;
304
305 vs = vxlan_find_sock(net, port);
306 if (!vs)
307 return NULL;
308
309 return vxlan_vs_find_vni(vs, id);
310 }
311
312 /* Fill in neighbour message in skbuff. */
313 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
314 const struct vxlan_fdb *fdb,
315 u32 portid, u32 seq, int type, unsigned int flags,
316 const struct vxlan_rdst *rdst)
317 {
318 unsigned long now = jiffies;
319 struct nda_cacheinfo ci;
320 struct nlmsghdr *nlh;
321 struct ndmsg *ndm;
322 bool send_ip, send_eth;
323
324 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
325 if (nlh == NULL)
326 return -EMSGSIZE;
327
328 ndm = nlmsg_data(nlh);
329 memset(ndm, 0, sizeof(*ndm));
330
331 send_eth = send_ip = true;
332
333 if (type == RTM_GETNEIGH) {
334 ndm->ndm_family = AF_INET;
335 send_ip = !vxlan_addr_any(&rdst->remote_ip);
336 send_eth = !is_zero_ether_addr(fdb->eth_addr);
337 } else
338 ndm->ndm_family = AF_BRIDGE;
339 ndm->ndm_state = fdb->state;
340 ndm->ndm_ifindex = vxlan->dev->ifindex;
341 ndm->ndm_flags = fdb->flags;
342 ndm->ndm_type = NDA_DST;
343
344 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
345 goto nla_put_failure;
346
347 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
348 goto nla_put_failure;
349
350 if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
351 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
352 goto nla_put_failure;
353 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
354 nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
355 goto nla_put_failure;
356 if (rdst->remote_ifindex &&
357 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
358 goto nla_put_failure;
359
360 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
361 ci.ndm_confirmed = 0;
362 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
363 ci.ndm_refcnt = 0;
364
365 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
366 goto nla_put_failure;
367
368 return nlmsg_end(skb, nlh);
369
370 nla_put_failure:
371 nlmsg_cancel(skb, nlh);
372 return -EMSGSIZE;
373 }
374
375 static inline size_t vxlan_nlmsg_size(void)
376 {
377 return NLMSG_ALIGN(sizeof(struct ndmsg))
378 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
379 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
380 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
381 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
382 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
383 + nla_total_size(sizeof(struct nda_cacheinfo));
384 }
385
386 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
387 struct vxlan_rdst *rd, int type)
388 {
389 struct net *net = dev_net(vxlan->dev);
390 struct sk_buff *skb;
391 int err = -ENOBUFS;
392
393 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
394 if (skb == NULL)
395 goto errout;
396
397 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
398 if (err < 0) {
399 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
400 WARN_ON(err == -EMSGSIZE);
401 kfree_skb(skb);
402 goto errout;
403 }
404
405 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
406 return;
407 errout:
408 if (err < 0)
409 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
410 }
411
412 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
413 {
414 struct vxlan_dev *vxlan = netdev_priv(dev);
415 struct vxlan_fdb f = {
416 .state = NUD_STALE,
417 };
418 struct vxlan_rdst remote = {
419 .remote_ip = *ipa, /* goes to NDA_DST */
420 .remote_vni = VXLAN_N_VID,
421 };
422
423 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
424 }
425
426 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
427 {
428 struct vxlan_fdb f = {
429 .state = NUD_STALE,
430 };
431 struct vxlan_rdst remote = { };
432
433 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
434
435 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
436 }
437
438 /* Hash Ethernet address */
439 static u32 eth_hash(const unsigned char *addr)
440 {
441 u64 value = get_unaligned((u64 *)addr);
442
443 /* only want 6 bytes */
444 #ifdef __BIG_ENDIAN
445 value >>= 16;
446 #else
447 value <<= 16;
448 #endif
449 return hash_64(value, FDB_HASH_BITS);
450 }
451
452 /* Hash chain to use given mac address */
453 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
454 const u8 *mac)
455 {
456 return &vxlan->fdb_head[eth_hash(mac)];
457 }
458
459 /* Look up Ethernet address in forwarding table */
460 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
461 const u8 *mac)
462 {
463 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
464 struct vxlan_fdb *f;
465
466 hlist_for_each_entry_rcu(f, head, hlist) {
467 if (ether_addr_equal(mac, f->eth_addr))
468 return f;
469 }
470
471 return NULL;
472 }
473
474 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
475 const u8 *mac)
476 {
477 struct vxlan_fdb *f;
478
479 f = __vxlan_find_mac(vxlan, mac);
480 if (f)
481 f->used = jiffies;
482
483 return f;
484 }
485
486 /* caller should hold vxlan->hash_lock */
487 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
488 union vxlan_addr *ip, __be16 port,
489 __u32 vni, __u32 ifindex)
490 {
491 struct vxlan_rdst *rd;
492
493 list_for_each_entry(rd, &f->remotes, list) {
494 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
495 rd->remote_port == port &&
496 rd->remote_vni == vni &&
497 rd->remote_ifindex == ifindex)
498 return rd;
499 }
500
501 return NULL;
502 }
503
504 /* Replace destination of unicast mac */
505 static int vxlan_fdb_replace(struct vxlan_fdb *f,
506 union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
507 {
508 struct vxlan_rdst *rd;
509
510 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
511 if (rd)
512 return 0;
513
514 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
515 if (!rd)
516 return 0;
517 rd->remote_ip = *ip;
518 rd->remote_port = port;
519 rd->remote_vni = vni;
520 rd->remote_ifindex = ifindex;
521 return 1;
522 }
523
524 /* Add/update destinations for multicast */
525 static int vxlan_fdb_append(struct vxlan_fdb *f,
526 union vxlan_addr *ip, __be16 port, __u32 vni,
527 __u32 ifindex, struct vxlan_rdst **rdp)
528 {
529 struct vxlan_rdst *rd;
530
531 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
532 if (rd)
533 return 0;
534
535 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
536 if (rd == NULL)
537 return -ENOBUFS;
538 rd->remote_ip = *ip;
539 rd->remote_port = port;
540 rd->remote_vni = vni;
541 rd->remote_ifindex = ifindex;
542
543 list_add_tail_rcu(&rd->list, &f->remotes);
544
545 *rdp = rd;
546 return 1;
547 }
548
549 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head, struct sk_buff *skb)
550 {
551 struct sk_buff *p, **pp = NULL;
552 struct vxlanhdr *vh, *vh2;
553 struct ethhdr *eh, *eh2;
554 unsigned int hlen, off_vx, off_eth;
555 const struct packet_offload *ptype;
556 __be16 type;
557 int flush = 1;
558
559 off_vx = skb_gro_offset(skb);
560 hlen = off_vx + sizeof(*vh);
561 vh = skb_gro_header_fast(skb, off_vx);
562 if (skb_gro_header_hard(skb, hlen)) {
563 vh = skb_gro_header_slow(skb, hlen, off_vx);
564 if (unlikely(!vh))
565 goto out;
566 }
567 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
568 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
569
570 off_eth = skb_gro_offset(skb);
571 hlen = off_eth + sizeof(*eh);
572 eh = skb_gro_header_fast(skb, off_eth);
573 if (skb_gro_header_hard(skb, hlen)) {
574 eh = skb_gro_header_slow(skb, hlen, off_eth);
575 if (unlikely(!eh))
576 goto out;
577 }
578
579 flush = 0;
580
581 for (p = *head; p; p = p->next) {
582 if (!NAPI_GRO_CB(p)->same_flow)
583 continue;
584
585 vh2 = (struct vxlanhdr *)(p->data + off_vx);
586 eh2 = (struct ethhdr *)(p->data + off_eth);
587 if (vh->vx_vni != vh2->vx_vni || compare_ether_header(eh, eh2)) {
588 NAPI_GRO_CB(p)->same_flow = 0;
589 continue;
590 }
591 }
592
593 type = eh->h_proto;
594
595 rcu_read_lock();
596 ptype = gro_find_receive_by_type(type);
597 if (ptype == NULL) {
598 flush = 1;
599 goto out_unlock;
600 }
601
602 skb_gro_pull(skb, sizeof(*eh)); /* pull inner eth header */
603 skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
604 pp = ptype->callbacks.gro_receive(head, skb);
605
606 out_unlock:
607 rcu_read_unlock();
608 out:
609 NAPI_GRO_CB(skb)->flush |= flush;
610
611 return pp;
612 }
613
614 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff)
615 {
616 struct ethhdr *eh;
617 struct packet_offload *ptype;
618 __be16 type;
619 int vxlan_len = sizeof(struct vxlanhdr) + sizeof(struct ethhdr);
620 int err = -ENOSYS;
621
622 eh = (struct ethhdr *)(skb->data + nhoff + sizeof(struct vxlanhdr));
623 type = eh->h_proto;
624
625 rcu_read_lock();
626 ptype = gro_find_complete_by_type(type);
627 if (ptype != NULL)
628 err = ptype->callbacks.gro_complete(skb, nhoff + vxlan_len);
629
630 rcu_read_unlock();
631 return err;
632 }
633
634 /* Notify netdevs that UDP port started listening */
635 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
636 {
637 struct net_device *dev;
638 struct sock *sk = vs->sock->sk;
639 struct net *net = sock_net(sk);
640 sa_family_t sa_family = sk->sk_family;
641 __be16 port = inet_sk(sk)->inet_sport;
642 int err;
643
644 if (sa_family == AF_INET) {
645 err = udp_add_offload(&vs->udp_offloads);
646 if (err)
647 pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
648 }
649
650 rcu_read_lock();
651 for_each_netdev_rcu(net, dev) {
652 if (dev->netdev_ops->ndo_add_vxlan_port)
653 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
654 port);
655 }
656 rcu_read_unlock();
657 }
658
659 /* Notify netdevs that UDP port is no more listening */
660 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
661 {
662 struct net_device *dev;
663 struct sock *sk = vs->sock->sk;
664 struct net *net = sock_net(sk);
665 sa_family_t sa_family = sk->sk_family;
666 __be16 port = inet_sk(sk)->inet_sport;
667
668 rcu_read_lock();
669 for_each_netdev_rcu(net, dev) {
670 if (dev->netdev_ops->ndo_del_vxlan_port)
671 dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
672 port);
673 }
674 rcu_read_unlock();
675
676 if (sa_family == AF_INET)
677 udp_del_offload(&vs->udp_offloads);
678 }
679
680 /* Add new entry to forwarding table -- assumes lock held */
681 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
682 const u8 *mac, union vxlan_addr *ip,
683 __u16 state, __u16 flags,
684 __be16 port, __u32 vni, __u32 ifindex,
685 __u8 ndm_flags)
686 {
687 struct vxlan_rdst *rd = NULL;
688 struct vxlan_fdb *f;
689 int notify = 0;
690
691 f = __vxlan_find_mac(vxlan, mac);
692 if (f) {
693 if (flags & NLM_F_EXCL) {
694 netdev_dbg(vxlan->dev,
695 "lost race to create %pM\n", mac);
696 return -EEXIST;
697 }
698 if (f->state != state) {
699 f->state = state;
700 f->updated = jiffies;
701 notify = 1;
702 }
703 if (f->flags != ndm_flags) {
704 f->flags = ndm_flags;
705 f->updated = jiffies;
706 notify = 1;
707 }
708 if ((flags & NLM_F_REPLACE)) {
709 /* Only change unicasts */
710 if (!(is_multicast_ether_addr(f->eth_addr) ||
711 is_zero_ether_addr(f->eth_addr))) {
712 int rc = vxlan_fdb_replace(f, ip, port, vni,
713 ifindex);
714
715 if (rc < 0)
716 return rc;
717 notify |= rc;
718 } else
719 return -EOPNOTSUPP;
720 }
721 if ((flags & NLM_F_APPEND) &&
722 (is_multicast_ether_addr(f->eth_addr) ||
723 is_zero_ether_addr(f->eth_addr))) {
724 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
725 &rd);
726
727 if (rc < 0)
728 return rc;
729 notify |= rc;
730 }
731 } else {
732 if (!(flags & NLM_F_CREATE))
733 return -ENOENT;
734
735 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
736 return -ENOSPC;
737
738 /* Disallow replace to add a multicast entry */
739 if ((flags & NLM_F_REPLACE) &&
740 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
741 return -EOPNOTSUPP;
742
743 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
744 f = kmalloc(sizeof(*f), GFP_ATOMIC);
745 if (!f)
746 return -ENOMEM;
747
748 notify = 1;
749 f->state = state;
750 f->flags = ndm_flags;
751 f->updated = f->used = jiffies;
752 INIT_LIST_HEAD(&f->remotes);
753 memcpy(f->eth_addr, mac, ETH_ALEN);
754
755 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
756
757 ++vxlan->addrcnt;
758 hlist_add_head_rcu(&f->hlist,
759 vxlan_fdb_head(vxlan, mac));
760 }
761
762 if (notify) {
763 if (rd == NULL)
764 rd = first_remote_rtnl(f);
765 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
766 }
767
768 return 0;
769 }
770
771 static void vxlan_fdb_free(struct rcu_head *head)
772 {
773 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
774 struct vxlan_rdst *rd, *nd;
775
776 list_for_each_entry_safe(rd, nd, &f->remotes, list)
777 kfree(rd);
778 kfree(f);
779 }
780
781 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
782 {
783 netdev_dbg(vxlan->dev,
784 "delete %pM\n", f->eth_addr);
785
786 --vxlan->addrcnt;
787 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
788
789 hlist_del_rcu(&f->hlist);
790 call_rcu(&f->rcu, vxlan_fdb_free);
791 }
792
793 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
794 union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
795 {
796 struct net *net = dev_net(vxlan->dev);
797 int err;
798
799 if (tb[NDA_DST]) {
800 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
801 if (err)
802 return err;
803 } else {
804 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
805 if (remote->sa.sa_family == AF_INET) {
806 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
807 ip->sa.sa_family = AF_INET;
808 #if IS_ENABLED(CONFIG_IPV6)
809 } else {
810 ip->sin6.sin6_addr = in6addr_any;
811 ip->sa.sa_family = AF_INET6;
812 #endif
813 }
814 }
815
816 if (tb[NDA_PORT]) {
817 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
818 return -EINVAL;
819 *port = nla_get_be16(tb[NDA_PORT]);
820 } else {
821 *port = vxlan->dst_port;
822 }
823
824 if (tb[NDA_VNI]) {
825 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
826 return -EINVAL;
827 *vni = nla_get_u32(tb[NDA_VNI]);
828 } else {
829 *vni = vxlan->default_dst.remote_vni;
830 }
831
832 if (tb[NDA_IFINDEX]) {
833 struct net_device *tdev;
834
835 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
836 return -EINVAL;
837 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
838 tdev = __dev_get_by_index(net, *ifindex);
839 if (!tdev)
840 return -EADDRNOTAVAIL;
841 } else {
842 *ifindex = 0;
843 }
844
845 return 0;
846 }
847
848 /* Add static entry (via netlink) */
849 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
850 struct net_device *dev,
851 const unsigned char *addr, u16 flags)
852 {
853 struct vxlan_dev *vxlan = netdev_priv(dev);
854 /* struct net *net = dev_net(vxlan->dev); */
855 union vxlan_addr ip;
856 __be16 port;
857 u32 vni, ifindex;
858 int err;
859
860 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
861 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
862 ndm->ndm_state);
863 return -EINVAL;
864 }
865
866 if (tb[NDA_DST] == NULL)
867 return -EINVAL;
868
869 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
870 if (err)
871 return err;
872
873 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
874 return -EAFNOSUPPORT;
875
876 spin_lock_bh(&vxlan->hash_lock);
877 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
878 port, vni, ifindex, ndm->ndm_flags);
879 spin_unlock_bh(&vxlan->hash_lock);
880
881 return err;
882 }
883
884 /* Delete entry (via netlink) */
885 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
886 struct net_device *dev,
887 const unsigned char *addr)
888 {
889 struct vxlan_dev *vxlan = netdev_priv(dev);
890 struct vxlan_fdb *f;
891 struct vxlan_rdst *rd = NULL;
892 union vxlan_addr ip;
893 __be16 port;
894 u32 vni, ifindex;
895 int err;
896
897 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
898 if (err)
899 return err;
900
901 err = -ENOENT;
902
903 spin_lock_bh(&vxlan->hash_lock);
904 f = vxlan_find_mac(vxlan, addr);
905 if (!f)
906 goto out;
907
908 if (!vxlan_addr_any(&ip)) {
909 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
910 if (!rd)
911 goto out;
912 }
913
914 err = 0;
915
916 /* remove a destination if it's not the only one on the list,
917 * otherwise destroy the fdb entry
918 */
919 if (rd && !list_is_singular(&f->remotes)) {
920 list_del_rcu(&rd->list);
921 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
922 kfree_rcu(rd, rcu);
923 goto out;
924 }
925
926 vxlan_fdb_destroy(vxlan, f);
927
928 out:
929 spin_unlock_bh(&vxlan->hash_lock);
930
931 return err;
932 }
933
934 /* Dump forwarding table */
935 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
936 struct net_device *dev, int idx)
937 {
938 struct vxlan_dev *vxlan = netdev_priv(dev);
939 unsigned int h;
940
941 for (h = 0; h < FDB_HASH_SIZE; ++h) {
942 struct vxlan_fdb *f;
943 int err;
944
945 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
946 struct vxlan_rdst *rd;
947
948 if (idx < cb->args[0])
949 goto skip;
950
951 list_for_each_entry_rcu(rd, &f->remotes, list) {
952 err = vxlan_fdb_info(skb, vxlan, f,
953 NETLINK_CB(cb->skb).portid,
954 cb->nlh->nlmsg_seq,
955 RTM_NEWNEIGH,
956 NLM_F_MULTI, rd);
957 if (err < 0)
958 goto out;
959 }
960 skip:
961 ++idx;
962 }
963 }
964 out:
965 return idx;
966 }
967
968 /* Watch incoming packets to learn mapping between Ethernet address
969 * and Tunnel endpoint.
970 * Return true if packet is bogus and should be droppped.
971 */
972 static bool vxlan_snoop(struct net_device *dev,
973 union vxlan_addr *src_ip, const u8 *src_mac)
974 {
975 struct vxlan_dev *vxlan = netdev_priv(dev);
976 struct vxlan_fdb *f;
977
978 f = vxlan_find_mac(vxlan, src_mac);
979 if (likely(f)) {
980 struct vxlan_rdst *rdst = first_remote_rcu(f);
981
982 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
983 return false;
984
985 /* Don't migrate static entries, drop packets */
986 if (f->state & NUD_NOARP)
987 return true;
988
989 if (net_ratelimit())
990 netdev_info(dev,
991 "%pM migrated from %pIS to %pIS\n",
992 src_mac, &rdst->remote_ip, &src_ip);
993
994 rdst->remote_ip = *src_ip;
995 f->updated = jiffies;
996 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
997 } else {
998 /* learned new entry */
999 spin_lock(&vxlan->hash_lock);
1000
1001 /* close off race between vxlan_flush and incoming packets */
1002 if (netif_running(dev))
1003 vxlan_fdb_create(vxlan, src_mac, src_ip,
1004 NUD_REACHABLE,
1005 NLM_F_EXCL|NLM_F_CREATE,
1006 vxlan->dst_port,
1007 vxlan->default_dst.remote_vni,
1008 0, NTF_SELF);
1009 spin_unlock(&vxlan->hash_lock);
1010 }
1011
1012 return false;
1013 }
1014
1015 /* See if multicast group is already in use by other ID */
1016 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1017 {
1018 struct vxlan_dev *vxlan;
1019
1020 /* The vxlan_sock is only used by dev, leaving group has
1021 * no effect on other vxlan devices.
1022 */
1023 if (atomic_read(&dev->vn_sock->refcnt) == 1)
1024 return false;
1025
1026 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1027 if (!netif_running(vxlan->dev) || vxlan == dev)
1028 continue;
1029
1030 if (vxlan->vn_sock != dev->vn_sock)
1031 continue;
1032
1033 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1034 &dev->default_dst.remote_ip))
1035 continue;
1036
1037 if (vxlan->default_dst.remote_ifindex !=
1038 dev->default_dst.remote_ifindex)
1039 continue;
1040
1041 return true;
1042 }
1043
1044 return false;
1045 }
1046
1047 static void vxlan_sock_hold(struct vxlan_sock *vs)
1048 {
1049 atomic_inc(&vs->refcnt);
1050 }
1051
1052 void vxlan_sock_release(struct vxlan_sock *vs)
1053 {
1054 struct sock *sk = vs->sock->sk;
1055 struct net *net = sock_net(sk);
1056 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1057
1058 if (!atomic_dec_and_test(&vs->refcnt))
1059 return;
1060
1061 spin_lock(&vn->sock_lock);
1062 hlist_del_rcu(&vs->hlist);
1063 rcu_assign_sk_user_data(vs->sock->sk, NULL);
1064 vxlan_notify_del_rx_port(vs);
1065 spin_unlock(&vn->sock_lock);
1066
1067 queue_work(vxlan_wq, &vs->del_work);
1068 }
1069 EXPORT_SYMBOL_GPL(vxlan_sock_release);
1070
1071 /* Callback to update multicast group membership when first VNI on
1072 * multicast asddress is brought up
1073 * Done as workqueue because ip_mc_join_group acquires RTNL.
1074 */
1075 static void vxlan_igmp_join(struct work_struct *work)
1076 {
1077 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join);
1078 struct vxlan_sock *vs = vxlan->vn_sock;
1079 struct sock *sk = vs->sock->sk;
1080 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1081 int ifindex = vxlan->default_dst.remote_ifindex;
1082
1083 lock_sock(sk);
1084 if (ip->sa.sa_family == AF_INET) {
1085 struct ip_mreqn mreq = {
1086 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1087 .imr_ifindex = ifindex,
1088 };
1089
1090 ip_mc_join_group(sk, &mreq);
1091 #if IS_ENABLED(CONFIG_IPV6)
1092 } else {
1093 ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1094 &ip->sin6.sin6_addr);
1095 #endif
1096 }
1097 release_sock(sk);
1098
1099 vxlan_sock_release(vs);
1100 dev_put(vxlan->dev);
1101 }
1102
1103 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1104 static void vxlan_igmp_leave(struct work_struct *work)
1105 {
1106 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave);
1107 struct vxlan_sock *vs = vxlan->vn_sock;
1108 struct sock *sk = vs->sock->sk;
1109 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1110 int ifindex = vxlan->default_dst.remote_ifindex;
1111
1112 lock_sock(sk);
1113 if (ip->sa.sa_family == AF_INET) {
1114 struct ip_mreqn mreq = {
1115 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1116 .imr_ifindex = ifindex,
1117 };
1118
1119 ip_mc_leave_group(sk, &mreq);
1120 #if IS_ENABLED(CONFIG_IPV6)
1121 } else {
1122 ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1123 &ip->sin6.sin6_addr);
1124 #endif
1125 }
1126
1127 release_sock(sk);
1128
1129 vxlan_sock_release(vs);
1130 dev_put(vxlan->dev);
1131 }
1132
1133 /* Callback from net/ipv4/udp.c to receive packets */
1134 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1135 {
1136 struct vxlan_sock *vs;
1137 struct vxlanhdr *vxh;
1138
1139 /* Need Vxlan and inner Ethernet header to be present */
1140 if (!pskb_may_pull(skb, VXLAN_HLEN))
1141 goto error;
1142
1143 /* Return packets with reserved bits set */
1144 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1145 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
1146 (vxh->vx_vni & htonl(0xff))) {
1147 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1148 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1149 goto error;
1150 }
1151
1152 if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1153 goto drop;
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_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1203 skb->protocol = eth_type_trans(skb, vxlan->dev);
1204
1205 /* Ignore packet loops (and multicast echo) */
1206 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1207 goto drop;
1208
1209 /* Re-examine inner Ethernet packet */
1210 if (remote_ip->sa.sa_family == AF_INET) {
1211 oip = ip_hdr(skb);
1212 saddr.sin.sin_addr.s_addr = oip->saddr;
1213 saddr.sa.sa_family = AF_INET;
1214 #if IS_ENABLED(CONFIG_IPV6)
1215 } else {
1216 oip6 = ipv6_hdr(skb);
1217 saddr.sin6.sin6_addr = oip6->saddr;
1218 saddr.sa.sa_family = AF_INET6;
1219 #endif
1220 }
1221
1222 if ((vxlan->flags & VXLAN_F_LEARN) &&
1223 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1224 goto drop;
1225
1226 skb_reset_network_header(skb);
1227
1228 if (oip6)
1229 err = IP6_ECN_decapsulate(oip6, skb);
1230 if (oip)
1231 err = IP_ECN_decapsulate(oip, skb);
1232
1233 if (unlikely(err)) {
1234 if (log_ecn_error) {
1235 if (oip6)
1236 net_info_ratelimited("non-ECT from %pI6\n",
1237 &oip6->saddr);
1238 if (oip)
1239 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1240 &oip->saddr, oip->tos);
1241 }
1242 if (err > 1) {
1243 ++vxlan->dev->stats.rx_frame_errors;
1244 ++vxlan->dev->stats.rx_errors;
1245 goto drop;
1246 }
1247 }
1248
1249 stats = this_cpu_ptr(vxlan->dev->tstats);
1250 u64_stats_update_begin(&stats->syncp);
1251 stats->rx_packets++;
1252 stats->rx_bytes += skb->len;
1253 u64_stats_update_end(&stats->syncp);
1254
1255 netif_rx(skb);
1256
1257 return;
1258 drop:
1259 /* Consume bad packet */
1260 kfree_skb(skb);
1261 }
1262
1263 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1264 {
1265 struct vxlan_dev *vxlan = netdev_priv(dev);
1266 struct arphdr *parp;
1267 u8 *arpptr, *sha;
1268 __be32 sip, tip;
1269 struct neighbour *n;
1270
1271 if (dev->flags & IFF_NOARP)
1272 goto out;
1273
1274 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1275 dev->stats.tx_dropped++;
1276 goto out;
1277 }
1278 parp = arp_hdr(skb);
1279
1280 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1281 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1282 parp->ar_pro != htons(ETH_P_IP) ||
1283 parp->ar_op != htons(ARPOP_REQUEST) ||
1284 parp->ar_hln != dev->addr_len ||
1285 parp->ar_pln != 4)
1286 goto out;
1287 arpptr = (u8 *)parp + sizeof(struct arphdr);
1288 sha = arpptr;
1289 arpptr += dev->addr_len; /* sha */
1290 memcpy(&sip, arpptr, sizeof(sip));
1291 arpptr += sizeof(sip);
1292 arpptr += dev->addr_len; /* tha */
1293 memcpy(&tip, arpptr, sizeof(tip));
1294
1295 if (ipv4_is_loopback(tip) ||
1296 ipv4_is_multicast(tip))
1297 goto out;
1298
1299 n = neigh_lookup(&arp_tbl, &tip, dev);
1300
1301 if (n) {
1302 struct vxlan_fdb *f;
1303 struct sk_buff *reply;
1304
1305 if (!(n->nud_state & NUD_CONNECTED)) {
1306 neigh_release(n);
1307 goto out;
1308 }
1309
1310 f = vxlan_find_mac(vxlan, n->ha);
1311 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1312 /* bridge-local neighbor */
1313 neigh_release(n);
1314 goto out;
1315 }
1316
1317 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1318 n->ha, sha);
1319
1320 neigh_release(n);
1321
1322 if (reply == NULL)
1323 goto out;
1324
1325 skb_reset_mac_header(reply);
1326 __skb_pull(reply, skb_network_offset(reply));
1327 reply->ip_summed = CHECKSUM_UNNECESSARY;
1328 reply->pkt_type = PACKET_HOST;
1329
1330 if (netif_rx_ni(reply) == NET_RX_DROP)
1331 dev->stats.rx_dropped++;
1332 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1333 union vxlan_addr ipa = {
1334 .sin.sin_addr.s_addr = tip,
1335 .sa.sa_family = AF_INET,
1336 };
1337
1338 vxlan_ip_miss(dev, &ipa);
1339 }
1340 out:
1341 consume_skb(skb);
1342 return NETDEV_TX_OK;
1343 }
1344
1345 #if IS_ENABLED(CONFIG_IPV6)
1346
1347 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1348 struct neighbour *n, bool isrouter)
1349 {
1350 struct net_device *dev = request->dev;
1351 struct sk_buff *reply;
1352 struct nd_msg *ns, *na;
1353 struct ipv6hdr *pip6;
1354 u8 *daddr;
1355 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1356 int ns_olen;
1357 int i, len;
1358
1359 if (dev == NULL)
1360 return NULL;
1361
1362 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1363 sizeof(*na) + na_olen + dev->needed_tailroom;
1364 reply = alloc_skb(len, GFP_ATOMIC);
1365 if (reply == NULL)
1366 return NULL;
1367
1368 reply->protocol = htons(ETH_P_IPV6);
1369 reply->dev = dev;
1370 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1371 skb_push(reply, sizeof(struct ethhdr));
1372 skb_set_mac_header(reply, 0);
1373
1374 ns = (struct nd_msg *)skb_transport_header(request);
1375
1376 daddr = eth_hdr(request)->h_source;
1377 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1378 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1379 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1380 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1381 break;
1382 }
1383 }
1384
1385 /* Ethernet header */
1386 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1387 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1388 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1389 reply->protocol = htons(ETH_P_IPV6);
1390
1391 skb_pull(reply, sizeof(struct ethhdr));
1392 skb_set_network_header(reply, 0);
1393 skb_put(reply, sizeof(struct ipv6hdr));
1394
1395 /* IPv6 header */
1396
1397 pip6 = ipv6_hdr(reply);
1398 memset(pip6, 0, sizeof(struct ipv6hdr));
1399 pip6->version = 6;
1400 pip6->priority = ipv6_hdr(request)->priority;
1401 pip6->nexthdr = IPPROTO_ICMPV6;
1402 pip6->hop_limit = 255;
1403 pip6->daddr = ipv6_hdr(request)->saddr;
1404 pip6->saddr = *(struct in6_addr *)n->primary_key;
1405
1406 skb_pull(reply, sizeof(struct ipv6hdr));
1407 skb_set_transport_header(reply, 0);
1408
1409 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1410
1411 /* Neighbor Advertisement */
1412 memset(na, 0, sizeof(*na)+na_olen);
1413 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1414 na->icmph.icmp6_router = isrouter;
1415 na->icmph.icmp6_override = 1;
1416 na->icmph.icmp6_solicited = 1;
1417 na->target = ns->target;
1418 ether_addr_copy(&na->opt[2], n->ha);
1419 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1420 na->opt[1] = na_olen >> 3;
1421
1422 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1423 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1424 csum_partial(na, sizeof(*na)+na_olen, 0));
1425
1426 pip6->payload_len = htons(sizeof(*na)+na_olen);
1427
1428 skb_push(reply, sizeof(struct ipv6hdr));
1429
1430 reply->ip_summed = CHECKSUM_UNNECESSARY;
1431
1432 return reply;
1433 }
1434
1435 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1436 {
1437 struct vxlan_dev *vxlan = netdev_priv(dev);
1438 struct nd_msg *msg;
1439 const struct ipv6hdr *iphdr;
1440 const struct in6_addr *saddr, *daddr;
1441 struct neighbour *n;
1442 struct inet6_dev *in6_dev;
1443
1444 in6_dev = __in6_dev_get(dev);
1445 if (!in6_dev)
1446 goto out;
1447
1448 if (!pskb_may_pull(skb, skb->len))
1449 goto out;
1450
1451 iphdr = ipv6_hdr(skb);
1452 saddr = &iphdr->saddr;
1453 daddr = &iphdr->daddr;
1454
1455 msg = (struct nd_msg *)skb_transport_header(skb);
1456 if (msg->icmph.icmp6_code != 0 ||
1457 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1458 goto out;
1459
1460 if (ipv6_addr_loopback(daddr) ||
1461 ipv6_addr_is_multicast(&msg->target))
1462 goto out;
1463
1464 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1465
1466 if (n) {
1467 struct vxlan_fdb *f;
1468 struct sk_buff *reply;
1469
1470 if (!(n->nud_state & NUD_CONNECTED)) {
1471 neigh_release(n);
1472 goto out;
1473 }
1474
1475 f = vxlan_find_mac(vxlan, n->ha);
1476 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1477 /* bridge-local neighbor */
1478 neigh_release(n);
1479 goto out;
1480 }
1481
1482 reply = vxlan_na_create(skb, n,
1483 !!(f ? f->flags & NTF_ROUTER : 0));
1484
1485 neigh_release(n);
1486
1487 if (reply == NULL)
1488 goto out;
1489
1490 if (netif_rx_ni(reply) == NET_RX_DROP)
1491 dev->stats.rx_dropped++;
1492
1493 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1494 union vxlan_addr ipa = {
1495 .sin6.sin6_addr = msg->target,
1496 .sa.sa_family = AF_INET6,
1497 };
1498
1499 vxlan_ip_miss(dev, &ipa);
1500 }
1501
1502 out:
1503 consume_skb(skb);
1504 return NETDEV_TX_OK;
1505 }
1506 #endif
1507
1508 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1509 {
1510 struct vxlan_dev *vxlan = netdev_priv(dev);
1511 struct neighbour *n;
1512
1513 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1514 return false;
1515
1516 n = NULL;
1517 switch (ntohs(eth_hdr(skb)->h_proto)) {
1518 case ETH_P_IP:
1519 {
1520 struct iphdr *pip;
1521
1522 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1523 return false;
1524 pip = ip_hdr(skb);
1525 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1526 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1527 union vxlan_addr ipa = {
1528 .sin.sin_addr.s_addr = pip->daddr,
1529 .sa.sa_family = AF_INET,
1530 };
1531
1532 vxlan_ip_miss(dev, &ipa);
1533 return false;
1534 }
1535
1536 break;
1537 }
1538 #if IS_ENABLED(CONFIG_IPV6)
1539 case ETH_P_IPV6:
1540 {
1541 struct ipv6hdr *pip6;
1542
1543 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1544 return false;
1545 pip6 = ipv6_hdr(skb);
1546 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1547 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1548 union vxlan_addr ipa = {
1549 .sin6.sin6_addr = pip6->daddr,
1550 .sa.sa_family = AF_INET6,
1551 };
1552
1553 vxlan_ip_miss(dev, &ipa);
1554 return false;
1555 }
1556
1557 break;
1558 }
1559 #endif
1560 default:
1561 return false;
1562 }
1563
1564 if (n) {
1565 bool diff;
1566
1567 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1568 if (diff) {
1569 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1570 dev->addr_len);
1571 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1572 }
1573 neigh_release(n);
1574 return diff;
1575 }
1576
1577 return false;
1578 }
1579
1580 /* Compute source port for outgoing packet
1581 * first choice to use L4 flow hash since it will spread
1582 * better and maybe available from hardware
1583 * secondary choice is to use jhash on the Ethernet header
1584 */
1585 __be16 vxlan_src_port(__u16 port_min, __u16 port_max, struct sk_buff *skb)
1586 {
1587 unsigned int range = (port_max - port_min) + 1;
1588 u32 hash;
1589
1590 hash = skb_get_hash(skb);
1591 if (!hash)
1592 hash = jhash(skb->data, 2 * ETH_ALEN,
1593 (__force u32) skb->protocol);
1594
1595 return htons((((u64) hash * range) >> 32) + port_min);
1596 }
1597 EXPORT_SYMBOL_GPL(vxlan_src_port);
1598
1599 static inline struct sk_buff *vxlan_handle_offloads(struct sk_buff *skb,
1600 bool udp_csum)
1601 {
1602 int type = udp_csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1603 return iptunnel_handle_offloads(skb, udp_csum, type);
1604 }
1605
1606 #if IS_ENABLED(CONFIG_IPV6)
1607 static int vxlan6_xmit_skb(struct vxlan_sock *vs,
1608 struct dst_entry *dst, struct sk_buff *skb,
1609 struct net_device *dev, struct in6_addr *saddr,
1610 struct in6_addr *daddr, __u8 prio, __u8 ttl,
1611 __be16 src_port, __be16 dst_port, __be32 vni,
1612 bool xnet)
1613 {
1614 struct ipv6hdr *ip6h;
1615 struct vxlanhdr *vxh;
1616 struct udphdr *uh;
1617 int min_headroom;
1618 int err;
1619
1620 skb = vxlan_handle_offloads(skb, !udp_get_no_check6_tx(vs->sock->sk));
1621 if (IS_ERR(skb))
1622 return -EINVAL;
1623
1624 skb_scrub_packet(skb, xnet);
1625
1626 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1627 + VXLAN_HLEN + sizeof(struct ipv6hdr)
1628 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1629
1630 /* Need space for new headers (invalidates iph ptr) */
1631 err = skb_cow_head(skb, min_headroom);
1632 if (unlikely(err))
1633 return err;
1634
1635 if (vlan_tx_tag_present(skb)) {
1636 if (WARN_ON(!__vlan_put_tag(skb,
1637 skb->vlan_proto,
1638 vlan_tx_tag_get(skb))))
1639 return -ENOMEM;
1640
1641 skb->vlan_tci = 0;
1642 }
1643
1644 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1645 vxh->vx_flags = htonl(VXLAN_FLAGS);
1646 vxh->vx_vni = vni;
1647
1648 __skb_push(skb, sizeof(*uh));
1649 skb_reset_transport_header(skb);
1650 uh = udp_hdr(skb);
1651
1652 uh->dest = dst_port;
1653 uh->source = src_port;
1654
1655 uh->len = htons(skb->len);
1656
1657 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1658 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1659 IPSKB_REROUTED);
1660 skb_dst_set(skb, dst);
1661
1662 udp6_set_csum(udp_get_no_check6_tx(vs->sock->sk), skb,
1663 saddr, daddr, skb->len);
1664
1665 __skb_push(skb, sizeof(*ip6h));
1666 skb_reset_network_header(skb);
1667 ip6h = ipv6_hdr(skb);
1668 ip6h->version = 6;
1669 ip6h->priority = prio;
1670 ip6h->flow_lbl[0] = 0;
1671 ip6h->flow_lbl[1] = 0;
1672 ip6h->flow_lbl[2] = 0;
1673 ip6h->payload_len = htons(skb->len);
1674 ip6h->nexthdr = IPPROTO_UDP;
1675 ip6h->hop_limit = ttl;
1676 ip6h->daddr = *daddr;
1677 ip6h->saddr = *saddr;
1678
1679 ip6tunnel_xmit(skb, dev);
1680 return 0;
1681 }
1682 #endif
1683
1684 int vxlan_xmit_skb(struct vxlan_sock *vs,
1685 struct rtable *rt, struct sk_buff *skb,
1686 __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1687 __be16 src_port, __be16 dst_port, __be32 vni, bool xnet)
1688 {
1689 struct vxlanhdr *vxh;
1690 struct udphdr *uh;
1691 int min_headroom;
1692 int err;
1693
1694 skb = vxlan_handle_offloads(skb, !vs->sock->sk->sk_no_check_tx);
1695 if (IS_ERR(skb))
1696 return -EINVAL;
1697
1698 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1699 + VXLAN_HLEN + sizeof(struct iphdr)
1700 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1701
1702 /* Need space for new headers (invalidates iph ptr) */
1703 err = skb_cow_head(skb, min_headroom);
1704 if (unlikely(err))
1705 return err;
1706
1707 if (vlan_tx_tag_present(skb)) {
1708 if (WARN_ON(!__vlan_put_tag(skb,
1709 skb->vlan_proto,
1710 vlan_tx_tag_get(skb))))
1711 return -ENOMEM;
1712
1713 skb->vlan_tci = 0;
1714 }
1715
1716 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1717 vxh->vx_flags = htonl(VXLAN_FLAGS);
1718 vxh->vx_vni = vni;
1719
1720 __skb_push(skb, sizeof(*uh));
1721 skb_reset_transport_header(skb);
1722 uh = udp_hdr(skb);
1723
1724 uh->dest = dst_port;
1725 uh->source = src_port;
1726
1727 uh->len = htons(skb->len);
1728
1729 udp_set_csum(vs->sock->sk->sk_no_check_tx, skb,
1730 src, dst, skb->len);
1731
1732 return iptunnel_xmit(vs->sock->sk, rt, skb, src, dst, IPPROTO_UDP,
1733 tos, ttl, df, xnet);
1734 }
1735 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1736
1737 /* Bypass encapsulation if the destination is local */
1738 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1739 struct vxlan_dev *dst_vxlan)
1740 {
1741 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1742 union vxlan_addr loopback;
1743 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1744
1745 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1746 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1747 skb->pkt_type = PACKET_HOST;
1748 skb->encapsulation = 0;
1749 skb->dev = dst_vxlan->dev;
1750 __skb_pull(skb, skb_network_offset(skb));
1751
1752 if (remote_ip->sa.sa_family == AF_INET) {
1753 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1754 loopback.sa.sa_family = AF_INET;
1755 #if IS_ENABLED(CONFIG_IPV6)
1756 } else {
1757 loopback.sin6.sin6_addr = in6addr_loopback;
1758 loopback.sa.sa_family = AF_INET6;
1759 #endif
1760 }
1761
1762 if (dst_vxlan->flags & VXLAN_F_LEARN)
1763 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1764
1765 u64_stats_update_begin(&tx_stats->syncp);
1766 tx_stats->tx_packets++;
1767 tx_stats->tx_bytes += skb->len;
1768 u64_stats_update_end(&tx_stats->syncp);
1769
1770 if (netif_rx(skb) == NET_RX_SUCCESS) {
1771 u64_stats_update_begin(&rx_stats->syncp);
1772 rx_stats->rx_packets++;
1773 rx_stats->rx_bytes += skb->len;
1774 u64_stats_update_end(&rx_stats->syncp);
1775 } else {
1776 skb->dev->stats.rx_dropped++;
1777 }
1778 }
1779
1780 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1781 struct vxlan_rdst *rdst, bool did_rsc)
1782 {
1783 struct vxlan_dev *vxlan = netdev_priv(dev);
1784 struct rtable *rt = NULL;
1785 const struct iphdr *old_iph;
1786 struct flowi4 fl4;
1787 union vxlan_addr *dst;
1788 __be16 src_port = 0, dst_port;
1789 u32 vni;
1790 __be16 df = 0;
1791 __u8 tos, ttl;
1792 int err;
1793
1794 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1795 vni = rdst->remote_vni;
1796 dst = &rdst->remote_ip;
1797
1798 if (vxlan_addr_any(dst)) {
1799 if (did_rsc) {
1800 /* short-circuited back to local bridge */
1801 vxlan_encap_bypass(skb, vxlan, vxlan);
1802 return;
1803 }
1804 goto drop;
1805 }
1806
1807 old_iph = ip_hdr(skb);
1808
1809 ttl = vxlan->ttl;
1810 if (!ttl && vxlan_addr_multicast(dst))
1811 ttl = 1;
1812
1813 tos = vxlan->tos;
1814 if (tos == 1)
1815 tos = ip_tunnel_get_dsfield(old_iph, skb);
1816
1817 src_port = vxlan_src_port(vxlan->port_min, vxlan->port_max, skb);
1818
1819 if (dst->sa.sa_family == AF_INET) {
1820 memset(&fl4, 0, sizeof(fl4));
1821 fl4.flowi4_oif = rdst->remote_ifindex;
1822 fl4.flowi4_tos = RT_TOS(tos);
1823 fl4.daddr = dst->sin.sin_addr.s_addr;
1824 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1825
1826 rt = ip_route_output_key(vxlan->net, &fl4);
1827 if (IS_ERR(rt)) {
1828 netdev_dbg(dev, "no route to %pI4\n",
1829 &dst->sin.sin_addr.s_addr);
1830 dev->stats.tx_carrier_errors++;
1831 goto tx_error;
1832 }
1833
1834 if (rt->dst.dev == dev) {
1835 netdev_dbg(dev, "circular route to %pI4\n",
1836 &dst->sin.sin_addr.s_addr);
1837 dev->stats.collisions++;
1838 goto rt_tx_error;
1839 }
1840
1841 /* Bypass encapsulation if the destination is local */
1842 if (rt->rt_flags & RTCF_LOCAL &&
1843 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1844 struct vxlan_dev *dst_vxlan;
1845
1846 ip_rt_put(rt);
1847 dst_vxlan = vxlan_find_vni(vxlan->net, vni, dst_port);
1848 if (!dst_vxlan)
1849 goto tx_error;
1850 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1851 return;
1852 }
1853
1854 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1855 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1856
1857 err = vxlan_xmit_skb(vxlan->vn_sock, rt, skb,
1858 fl4.saddr, dst->sin.sin_addr.s_addr,
1859 tos, ttl, df, src_port, dst_port,
1860 htonl(vni << 8),
1861 !net_eq(vxlan->net, dev_net(vxlan->dev)));
1862
1863 if (err < 0)
1864 goto rt_tx_error;
1865 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1866 #if IS_ENABLED(CONFIG_IPV6)
1867 } else {
1868 struct sock *sk = vxlan->vn_sock->sock->sk;
1869 struct dst_entry *ndst;
1870 struct flowi6 fl6;
1871 u32 flags;
1872
1873 memset(&fl6, 0, sizeof(fl6));
1874 fl6.flowi6_oif = rdst->remote_ifindex;
1875 fl6.daddr = dst->sin6.sin6_addr;
1876 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
1877 fl6.flowi6_proto = IPPROTO_UDP;
1878
1879 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
1880 netdev_dbg(dev, "no route to %pI6\n",
1881 &dst->sin6.sin6_addr);
1882 dev->stats.tx_carrier_errors++;
1883 goto tx_error;
1884 }
1885
1886 if (ndst->dev == dev) {
1887 netdev_dbg(dev, "circular route to %pI6\n",
1888 &dst->sin6.sin6_addr);
1889 dst_release(ndst);
1890 dev->stats.collisions++;
1891 goto tx_error;
1892 }
1893
1894 /* Bypass encapsulation if the destination is local */
1895 flags = ((struct rt6_info *)ndst)->rt6i_flags;
1896 if (flags & RTF_LOCAL &&
1897 !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1898 struct vxlan_dev *dst_vxlan;
1899
1900 dst_release(ndst);
1901 dst_vxlan = vxlan_find_vni(vxlan->net, vni, dst_port);
1902 if (!dst_vxlan)
1903 goto tx_error;
1904 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1905 return;
1906 }
1907
1908 ttl = ttl ? : ip6_dst_hoplimit(ndst);
1909
1910 err = vxlan6_xmit_skb(vxlan->vn_sock, ndst, skb,
1911 dev, &fl6.saddr, &fl6.daddr, 0, ttl,
1912 src_port, dst_port, htonl(vni << 8),
1913 !net_eq(vxlan->net, dev_net(vxlan->dev)));
1914 #endif
1915 }
1916
1917 return;
1918
1919 drop:
1920 dev->stats.tx_dropped++;
1921 goto tx_free;
1922
1923 rt_tx_error:
1924 ip_rt_put(rt);
1925 tx_error:
1926 dev->stats.tx_errors++;
1927 tx_free:
1928 dev_kfree_skb(skb);
1929 }
1930
1931 /* Transmit local packets over Vxlan
1932 *
1933 * Outer IP header inherits ECN and DF from inner header.
1934 * Outer UDP destination is the VXLAN assigned port.
1935 * source port is based on hash of flow
1936 */
1937 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1938 {
1939 struct vxlan_dev *vxlan = netdev_priv(dev);
1940 struct ethhdr *eth;
1941 bool did_rsc = false;
1942 struct vxlan_rdst *rdst, *fdst = NULL;
1943 struct vxlan_fdb *f;
1944
1945 skb_reset_mac_header(skb);
1946 eth = eth_hdr(skb);
1947
1948 if ((vxlan->flags & VXLAN_F_PROXY)) {
1949 if (ntohs(eth->h_proto) == ETH_P_ARP)
1950 return arp_reduce(dev, skb);
1951 #if IS_ENABLED(CONFIG_IPV6)
1952 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
1953 skb->len >= sizeof(struct ipv6hdr) + sizeof(struct nd_msg) &&
1954 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
1955 struct nd_msg *msg;
1956
1957 msg = (struct nd_msg *)skb_transport_header(skb);
1958 if (msg->icmph.icmp6_code == 0 &&
1959 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
1960 return neigh_reduce(dev, skb);
1961 }
1962 #endif
1963 }
1964
1965 f = vxlan_find_mac(vxlan, eth->h_dest);
1966 did_rsc = false;
1967
1968 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1969 (ntohs(eth->h_proto) == ETH_P_IP ||
1970 ntohs(eth->h_proto) == ETH_P_IPV6)) {
1971 did_rsc = route_shortcircuit(dev, skb);
1972 if (did_rsc)
1973 f = vxlan_find_mac(vxlan, eth->h_dest);
1974 }
1975
1976 if (f == NULL) {
1977 f = vxlan_find_mac(vxlan, all_zeros_mac);
1978 if (f == NULL) {
1979 if ((vxlan->flags & VXLAN_F_L2MISS) &&
1980 !is_multicast_ether_addr(eth->h_dest))
1981 vxlan_fdb_miss(vxlan, eth->h_dest);
1982
1983 dev->stats.tx_dropped++;
1984 kfree_skb(skb);
1985 return NETDEV_TX_OK;
1986 }
1987 }
1988
1989 list_for_each_entry_rcu(rdst, &f->remotes, list) {
1990 struct sk_buff *skb1;
1991
1992 if (!fdst) {
1993 fdst = rdst;
1994 continue;
1995 }
1996 skb1 = skb_clone(skb, GFP_ATOMIC);
1997 if (skb1)
1998 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1999 }
2000
2001 if (fdst)
2002 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2003 else
2004 kfree_skb(skb);
2005 return NETDEV_TX_OK;
2006 }
2007
2008 /* Walk the forwarding table and purge stale entries */
2009 static void vxlan_cleanup(unsigned long arg)
2010 {
2011 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2012 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2013 unsigned int h;
2014
2015 if (!netif_running(vxlan->dev))
2016 return;
2017
2018 spin_lock_bh(&vxlan->hash_lock);
2019 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2020 struct hlist_node *p, *n;
2021 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2022 struct vxlan_fdb *f
2023 = container_of(p, struct vxlan_fdb, hlist);
2024 unsigned long timeout;
2025
2026 if (f->state & NUD_PERMANENT)
2027 continue;
2028
2029 timeout = f->used + vxlan->age_interval * HZ;
2030 if (time_before_eq(timeout, jiffies)) {
2031 netdev_dbg(vxlan->dev,
2032 "garbage collect %pM\n",
2033 f->eth_addr);
2034 f->state = NUD_STALE;
2035 vxlan_fdb_destroy(vxlan, f);
2036 } else if (time_before(timeout, next_timer))
2037 next_timer = timeout;
2038 }
2039 }
2040 spin_unlock_bh(&vxlan->hash_lock);
2041
2042 mod_timer(&vxlan->age_timer, next_timer);
2043 }
2044
2045 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2046 {
2047 __u32 vni = vxlan->default_dst.remote_vni;
2048
2049 vxlan->vn_sock = vs;
2050 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2051 }
2052
2053 /* Setup stats when device is created */
2054 static int vxlan_init(struct net_device *dev)
2055 {
2056 struct vxlan_dev *vxlan = netdev_priv(dev);
2057 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2058 struct vxlan_sock *vs;
2059
2060 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2061 if (!dev->tstats)
2062 return -ENOMEM;
2063
2064 spin_lock(&vn->sock_lock);
2065 vs = vxlan_find_sock(vxlan->net, vxlan->dst_port);
2066 if (vs) {
2067 /* If we have a socket with same port already, reuse it */
2068 atomic_inc(&vs->refcnt);
2069 vxlan_vs_add_dev(vs, vxlan);
2070 } else {
2071 /* otherwise make new socket outside of RTNL */
2072 dev_hold(dev);
2073 queue_work(vxlan_wq, &vxlan->sock_work);
2074 }
2075 spin_unlock(&vn->sock_lock);
2076
2077 return 0;
2078 }
2079
2080 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2081 {
2082 struct vxlan_fdb *f;
2083
2084 spin_lock_bh(&vxlan->hash_lock);
2085 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2086 if (f)
2087 vxlan_fdb_destroy(vxlan, f);
2088 spin_unlock_bh(&vxlan->hash_lock);
2089 }
2090
2091 static void vxlan_uninit(struct net_device *dev)
2092 {
2093 struct vxlan_dev *vxlan = netdev_priv(dev);
2094 struct vxlan_sock *vs = vxlan->vn_sock;
2095
2096 vxlan_fdb_delete_default(vxlan);
2097
2098 if (vs)
2099 vxlan_sock_release(vs);
2100 free_percpu(dev->tstats);
2101 }
2102
2103 /* Start ageing timer and join group when device is brought up */
2104 static int vxlan_open(struct net_device *dev)
2105 {
2106 struct vxlan_dev *vxlan = netdev_priv(dev);
2107 struct vxlan_sock *vs = vxlan->vn_sock;
2108
2109 /* socket hasn't been created */
2110 if (!vs)
2111 return -ENOTCONN;
2112
2113 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2114 vxlan_sock_hold(vs);
2115 dev_hold(dev);
2116 queue_work(vxlan_wq, &vxlan->igmp_join);
2117 }
2118
2119 if (vxlan->age_interval)
2120 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2121
2122 return 0;
2123 }
2124
2125 /* Purge the forwarding table */
2126 static void vxlan_flush(struct vxlan_dev *vxlan)
2127 {
2128 unsigned int h;
2129
2130 spin_lock_bh(&vxlan->hash_lock);
2131 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2132 struct hlist_node *p, *n;
2133 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2134 struct vxlan_fdb *f
2135 = container_of(p, struct vxlan_fdb, hlist);
2136 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2137 if (!is_zero_ether_addr(f->eth_addr))
2138 vxlan_fdb_destroy(vxlan, f);
2139 }
2140 }
2141 spin_unlock_bh(&vxlan->hash_lock);
2142 }
2143
2144 /* Cleanup timer and forwarding table on shutdown */
2145 static int vxlan_stop(struct net_device *dev)
2146 {
2147 struct vxlan_dev *vxlan = netdev_priv(dev);
2148 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2149 struct vxlan_sock *vs = vxlan->vn_sock;
2150
2151 if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2152 !vxlan_group_used(vn, vxlan)) {
2153 vxlan_sock_hold(vs);
2154 dev_hold(dev);
2155 queue_work(vxlan_wq, &vxlan->igmp_leave);
2156 }
2157
2158 del_timer_sync(&vxlan->age_timer);
2159
2160 vxlan_flush(vxlan);
2161
2162 return 0;
2163 }
2164
2165 /* Stub, nothing needs to be done. */
2166 static void vxlan_set_multicast_list(struct net_device *dev)
2167 {
2168 }
2169
2170 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2171 {
2172 struct vxlan_dev *vxlan = netdev_priv(dev);
2173 struct vxlan_rdst *dst = &vxlan->default_dst;
2174 struct net_device *lowerdev;
2175 int max_mtu;
2176
2177 lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2178 if (lowerdev == NULL)
2179 return eth_change_mtu(dev, new_mtu);
2180
2181 if (dst->remote_ip.sa.sa_family == AF_INET6)
2182 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2183 else
2184 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2185
2186 if (new_mtu < 68 || new_mtu > max_mtu)
2187 return -EINVAL;
2188
2189 dev->mtu = new_mtu;
2190 return 0;
2191 }
2192
2193 static const struct net_device_ops vxlan_netdev_ops = {
2194 .ndo_init = vxlan_init,
2195 .ndo_uninit = vxlan_uninit,
2196 .ndo_open = vxlan_open,
2197 .ndo_stop = vxlan_stop,
2198 .ndo_start_xmit = vxlan_xmit,
2199 .ndo_get_stats64 = ip_tunnel_get_stats64,
2200 .ndo_set_rx_mode = vxlan_set_multicast_list,
2201 .ndo_change_mtu = vxlan_change_mtu,
2202 .ndo_validate_addr = eth_validate_addr,
2203 .ndo_set_mac_address = eth_mac_addr,
2204 .ndo_fdb_add = vxlan_fdb_add,
2205 .ndo_fdb_del = vxlan_fdb_delete,
2206 .ndo_fdb_dump = vxlan_fdb_dump,
2207 };
2208
2209 /* Info for udev, that this is a virtual tunnel endpoint */
2210 static struct device_type vxlan_type = {
2211 .name = "vxlan",
2212 };
2213
2214 /* Calls the ndo_add_vxlan_port of the caller in order to
2215 * supply the listening VXLAN udp ports. Callers are expected
2216 * to implement the ndo_add_vxlan_port.
2217 */
2218 void vxlan_get_rx_port(struct net_device *dev)
2219 {
2220 struct vxlan_sock *vs;
2221 struct net *net = dev_net(dev);
2222 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2223 sa_family_t sa_family;
2224 __be16 port;
2225 unsigned int i;
2226
2227 spin_lock(&vn->sock_lock);
2228 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2229 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2230 port = inet_sk(vs->sock->sk)->inet_sport;
2231 sa_family = vs->sock->sk->sk_family;
2232 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2233 port);
2234 }
2235 }
2236 spin_unlock(&vn->sock_lock);
2237 }
2238 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2239
2240 /* Initialize the device structure. */
2241 static void vxlan_setup(struct net_device *dev)
2242 {
2243 struct vxlan_dev *vxlan = netdev_priv(dev);
2244 unsigned int h;
2245 int low, high;
2246
2247 eth_hw_addr_random(dev);
2248 ether_setup(dev);
2249 if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2250 dev->hard_header_len = ETH_HLEN + VXLAN6_HEADROOM;
2251 else
2252 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
2253
2254 dev->netdev_ops = &vxlan_netdev_ops;
2255 dev->destructor = free_netdev;
2256 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2257
2258 dev->tx_queue_len = 0;
2259 dev->features |= NETIF_F_LLTX;
2260 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2261 dev->features |= NETIF_F_RXCSUM;
2262 dev->features |= NETIF_F_GSO_SOFTWARE;
2263
2264 dev->vlan_features = dev->features;
2265 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2266 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2267 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2268 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2269 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
2270 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2271
2272 INIT_LIST_HEAD(&vxlan->next);
2273 spin_lock_init(&vxlan->hash_lock);
2274 INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join);
2275 INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave);
2276 INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
2277
2278 init_timer_deferrable(&vxlan->age_timer);
2279 vxlan->age_timer.function = vxlan_cleanup;
2280 vxlan->age_timer.data = (unsigned long) vxlan;
2281
2282 inet_get_local_port_range(dev_net(dev), &low, &high);
2283 vxlan->port_min = low;
2284 vxlan->port_max = high;
2285 vxlan->dst_port = htons(vxlan_port);
2286
2287 vxlan->dev = dev;
2288
2289 for (h = 0; h < FDB_HASH_SIZE; ++h)
2290 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2291 }
2292
2293 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2294 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2295 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2296 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2297 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2298 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2299 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2300 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2301 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2302 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2303 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2304 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2305 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2306 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2307 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2308 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2309 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2310 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2311 };
2312
2313 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2314 {
2315 if (tb[IFLA_ADDRESS]) {
2316 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2317 pr_debug("invalid link address (not ethernet)\n");
2318 return -EINVAL;
2319 }
2320
2321 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2322 pr_debug("invalid all zero ethernet address\n");
2323 return -EADDRNOTAVAIL;
2324 }
2325 }
2326
2327 if (!data)
2328 return -EINVAL;
2329
2330 if (data[IFLA_VXLAN_ID]) {
2331 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2332 if (id >= VXLAN_VID_MASK)
2333 return -ERANGE;
2334 }
2335
2336 if (data[IFLA_VXLAN_PORT_RANGE]) {
2337 const struct ifla_vxlan_port_range *p
2338 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2339
2340 if (ntohs(p->high) < ntohs(p->low)) {
2341 pr_debug("port range %u .. %u not valid\n",
2342 ntohs(p->low), ntohs(p->high));
2343 return -EINVAL;
2344 }
2345 }
2346
2347 return 0;
2348 }
2349
2350 static void vxlan_get_drvinfo(struct net_device *netdev,
2351 struct ethtool_drvinfo *drvinfo)
2352 {
2353 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2354 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2355 }
2356
2357 static const struct ethtool_ops vxlan_ethtool_ops = {
2358 .get_drvinfo = vxlan_get_drvinfo,
2359 .get_link = ethtool_op_get_link,
2360 };
2361
2362 static void vxlan_del_work(struct work_struct *work)
2363 {
2364 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2365
2366 sk_release_kernel(vs->sock->sk);
2367 kfree_rcu(vs, rcu);
2368 }
2369
2370 #if IS_ENABLED(CONFIG_IPV6)
2371 /* Create UDP socket for encapsulation receive. AF_INET6 socket
2372 * could be used for both IPv4 and IPv6 communications, but
2373 * users may set bindv6only=1.
2374 */
2375 static struct socket *create_v6_sock(struct net *net, __be16 port, u32 flags)
2376 {
2377 struct sock *sk;
2378 struct socket *sock;
2379 struct sockaddr_in6 vxlan_addr = {
2380 .sin6_family = AF_INET6,
2381 .sin6_port = port,
2382 };
2383 int rc, val = 1;
2384
2385 rc = sock_create_kern(AF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
2386 if (rc < 0) {
2387 pr_debug("UDPv6 socket create failed\n");
2388 return ERR_PTR(rc);
2389 }
2390
2391 /* Put in proper namespace */
2392 sk = sock->sk;
2393 sk_change_net(sk, net);
2394
2395 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
2396 (char *)&val, sizeof(val));
2397 rc = kernel_bind(sock, (struct sockaddr *)&vxlan_addr,
2398 sizeof(struct sockaddr_in6));
2399 if (rc < 0) {
2400 pr_debug("bind for UDPv6 socket %pI6:%u (%d)\n",
2401 &vxlan_addr.sin6_addr, ntohs(vxlan_addr.sin6_port), rc);
2402 sk_release_kernel(sk);
2403 return ERR_PTR(rc);
2404 }
2405 /* At this point, IPv6 module should have been loaded in
2406 * sock_create_kern().
2407 */
2408 BUG_ON(!ipv6_stub);
2409
2410 /* Disable multicast loopback */
2411 inet_sk(sk)->mc_loop = 0;
2412
2413 if (flags & VXLAN_F_UDP_ZERO_CSUM6_TX)
2414 udp_set_no_check6_tx(sk, true);
2415
2416 if (flags & VXLAN_F_UDP_ZERO_CSUM6_RX)
2417 udp_set_no_check6_rx(sk, true);
2418
2419 return sock;
2420 }
2421
2422 #else
2423
2424 static struct socket *create_v6_sock(struct net *net, __be16 port, u32 flags)
2425 {
2426 return ERR_PTR(-EPFNOSUPPORT);
2427 }
2428 #endif
2429
2430 static struct socket *create_v4_sock(struct net *net, __be16 port, u32 flags)
2431 {
2432 struct sock *sk;
2433 struct socket *sock;
2434 struct sockaddr_in vxlan_addr = {
2435 .sin_family = AF_INET,
2436 .sin_addr.s_addr = htonl(INADDR_ANY),
2437 .sin_port = port,
2438 };
2439 int rc;
2440
2441 /* Create UDP socket for encapsulation receive. */
2442 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
2443 if (rc < 0) {
2444 pr_debug("UDP socket create failed\n");
2445 return ERR_PTR(rc);
2446 }
2447
2448 /* Put in proper namespace */
2449 sk = sock->sk;
2450 sk_change_net(sk, net);
2451
2452 rc = kernel_bind(sock, (struct sockaddr *) &vxlan_addr,
2453 sizeof(vxlan_addr));
2454 if (rc < 0) {
2455 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
2456 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
2457 sk_release_kernel(sk);
2458 return ERR_PTR(rc);
2459 }
2460
2461 /* Disable multicast loopback */
2462 inet_sk(sk)->mc_loop = 0;
2463
2464 if (!(flags & VXLAN_F_UDP_CSUM))
2465 sock->sk->sk_no_check_tx = 1;
2466
2467 return sock;
2468 }
2469
2470 /* Create new listen socket if needed */
2471 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2472 vxlan_rcv_t *rcv, void *data,
2473 u32 flags)
2474 {
2475 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2476 struct vxlan_sock *vs;
2477 struct socket *sock;
2478 struct sock *sk;
2479 unsigned int h;
2480 bool ipv6 = !!(flags & VXLAN_F_IPV6);
2481
2482 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2483 if (!vs)
2484 return ERR_PTR(-ENOMEM);
2485
2486 for (h = 0; h < VNI_HASH_SIZE; ++h)
2487 INIT_HLIST_HEAD(&vs->vni_list[h]);
2488
2489 INIT_WORK(&vs->del_work, vxlan_del_work);
2490
2491 if (ipv6)
2492 sock = create_v6_sock(net, port, flags);
2493 else
2494 sock = create_v4_sock(net, port, flags);
2495 if (IS_ERR(sock)) {
2496 kfree(vs);
2497 return ERR_CAST(sock);
2498 }
2499
2500 vs->sock = sock;
2501 sk = sock->sk;
2502 atomic_set(&vs->refcnt, 1);
2503 vs->rcv = rcv;
2504 vs->data = data;
2505 rcu_assign_sk_user_data(vs->sock->sk, vs);
2506
2507 /* Initialize the vxlan udp offloads structure */
2508 vs->udp_offloads.port = port;
2509 vs->udp_offloads.callbacks.gro_receive = vxlan_gro_receive;
2510 vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2511
2512 spin_lock(&vn->sock_lock);
2513 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2514 vxlan_notify_add_rx_port(vs);
2515 spin_unlock(&vn->sock_lock);
2516
2517 /* Mark socket as an encapsulation socket. */
2518 udp_sk(sk)->encap_type = 1;
2519 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
2520 #if IS_ENABLED(CONFIG_IPV6)
2521 if (ipv6)
2522 ipv6_stub->udpv6_encap_enable();
2523 else
2524 #endif
2525 udp_encap_enable();
2526
2527 return vs;
2528 }
2529
2530 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2531 vxlan_rcv_t *rcv, void *data,
2532 bool no_share, u32 flags)
2533 {
2534 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2535 struct vxlan_sock *vs;
2536
2537 vs = vxlan_socket_create(net, port, rcv, data, flags);
2538 if (!IS_ERR(vs))
2539 return vs;
2540
2541 if (no_share) /* Return error if sharing is not allowed. */
2542 return vs;
2543
2544 spin_lock(&vn->sock_lock);
2545 vs = vxlan_find_sock(net, port);
2546 if (vs) {
2547 if (vs->rcv == rcv)
2548 atomic_inc(&vs->refcnt);
2549 else
2550 vs = ERR_PTR(-EBUSY);
2551 }
2552 spin_unlock(&vn->sock_lock);
2553
2554 if (!vs)
2555 vs = ERR_PTR(-EINVAL);
2556
2557 return vs;
2558 }
2559 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2560
2561 /* Scheduled at device creation to bind to a socket */
2562 static void vxlan_sock_work(struct work_struct *work)
2563 {
2564 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2565 struct net *net = vxlan->net;
2566 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2567 __be16 port = vxlan->dst_port;
2568 struct vxlan_sock *nvs;
2569
2570 nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags);
2571 spin_lock(&vn->sock_lock);
2572 if (!IS_ERR(nvs))
2573 vxlan_vs_add_dev(nvs, vxlan);
2574 spin_unlock(&vn->sock_lock);
2575
2576 dev_put(vxlan->dev);
2577 }
2578
2579 static int vxlan_newlink(struct net *net, struct net_device *dev,
2580 struct nlattr *tb[], struct nlattr *data[])
2581 {
2582 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2583 struct vxlan_dev *vxlan = netdev_priv(dev);
2584 struct vxlan_rdst *dst = &vxlan->default_dst;
2585 __u32 vni;
2586 int err;
2587 bool use_ipv6 = false;
2588
2589 if (!data[IFLA_VXLAN_ID])
2590 return -EINVAL;
2591
2592 vxlan->net = dev_net(dev);
2593
2594 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2595 dst->remote_vni = vni;
2596
2597 /* Unless IPv6 is explicitly requested, assume IPv4 */
2598 dst->remote_ip.sa.sa_family = AF_INET;
2599 if (data[IFLA_VXLAN_GROUP]) {
2600 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2601 } else if (data[IFLA_VXLAN_GROUP6]) {
2602 if (!IS_ENABLED(CONFIG_IPV6))
2603 return -EPFNOSUPPORT;
2604
2605 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2606 sizeof(struct in6_addr));
2607 dst->remote_ip.sa.sa_family = AF_INET6;
2608 use_ipv6 = true;
2609 }
2610
2611 if (data[IFLA_VXLAN_LOCAL]) {
2612 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2613 vxlan->saddr.sa.sa_family = AF_INET;
2614 } else if (data[IFLA_VXLAN_LOCAL6]) {
2615 if (!IS_ENABLED(CONFIG_IPV6))
2616 return -EPFNOSUPPORT;
2617
2618 /* TODO: respect scope id */
2619 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2620 sizeof(struct in6_addr));
2621 vxlan->saddr.sa.sa_family = AF_INET6;
2622 use_ipv6 = true;
2623 }
2624
2625 if (data[IFLA_VXLAN_LINK] &&
2626 (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2627 struct net_device *lowerdev
2628 = __dev_get_by_index(net, dst->remote_ifindex);
2629
2630 if (!lowerdev) {
2631 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2632 return -ENODEV;
2633 }
2634
2635 #if IS_ENABLED(CONFIG_IPV6)
2636 if (use_ipv6) {
2637 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2638 if (idev && idev->cnf.disable_ipv6) {
2639 pr_info("IPv6 is disabled via sysctl\n");
2640 return -EPERM;
2641 }
2642 vxlan->flags |= VXLAN_F_IPV6;
2643 }
2644 #endif
2645
2646 if (!tb[IFLA_MTU])
2647 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2648
2649 /* update header length based on lower device */
2650 dev->hard_header_len = lowerdev->hard_header_len +
2651 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2652 } else if (use_ipv6)
2653 vxlan->flags |= VXLAN_F_IPV6;
2654
2655 if (data[IFLA_VXLAN_TOS])
2656 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2657
2658 if (data[IFLA_VXLAN_TTL])
2659 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2660
2661 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2662 vxlan->flags |= VXLAN_F_LEARN;
2663
2664 if (data[IFLA_VXLAN_AGEING])
2665 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2666 else
2667 vxlan->age_interval = FDB_AGE_DEFAULT;
2668
2669 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2670 vxlan->flags |= VXLAN_F_PROXY;
2671
2672 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2673 vxlan->flags |= VXLAN_F_RSC;
2674
2675 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2676 vxlan->flags |= VXLAN_F_L2MISS;
2677
2678 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2679 vxlan->flags |= VXLAN_F_L3MISS;
2680
2681 if (data[IFLA_VXLAN_LIMIT])
2682 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2683
2684 if (data[IFLA_VXLAN_PORT_RANGE]) {
2685 const struct ifla_vxlan_port_range *p
2686 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2687 vxlan->port_min = ntohs(p->low);
2688 vxlan->port_max = ntohs(p->high);
2689 }
2690
2691 if (data[IFLA_VXLAN_PORT])
2692 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2693
2694 if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2695 vxlan->flags |= VXLAN_F_UDP_CSUM;
2696
2697 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2698 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2699 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2700
2701 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2702 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2703 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2704
2705 if (vxlan_find_vni(net, vni, vxlan->dst_port)) {
2706 pr_info("duplicate VNI %u\n", vni);
2707 return -EEXIST;
2708 }
2709
2710 dev->ethtool_ops = &vxlan_ethtool_ops;
2711
2712 /* create an fdb entry for a valid default destination */
2713 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2714 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2715 &vxlan->default_dst.remote_ip,
2716 NUD_REACHABLE|NUD_PERMANENT,
2717 NLM_F_EXCL|NLM_F_CREATE,
2718 vxlan->dst_port,
2719 vxlan->default_dst.remote_vni,
2720 vxlan->default_dst.remote_ifindex,
2721 NTF_SELF);
2722 if (err)
2723 return err;
2724 }
2725
2726 err = register_netdevice(dev);
2727 if (err) {
2728 vxlan_fdb_delete_default(vxlan);
2729 return err;
2730 }
2731
2732 list_add(&vxlan->next, &vn->vxlan_list);
2733
2734 return 0;
2735 }
2736
2737 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2738 {
2739 struct vxlan_dev *vxlan = netdev_priv(dev);
2740 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2741
2742 spin_lock(&vn->sock_lock);
2743 if (!hlist_unhashed(&vxlan->hlist))
2744 hlist_del_rcu(&vxlan->hlist);
2745 spin_unlock(&vn->sock_lock);
2746
2747 list_del(&vxlan->next);
2748 unregister_netdevice_queue(dev, head);
2749 }
2750
2751 static size_t vxlan_get_size(const struct net_device *dev)
2752 {
2753
2754 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
2755 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2756 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2757 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2758 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
2759 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
2760 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
2761 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
2762 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
2763 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
2764 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
2765 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2766 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2767 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2768 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2769 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2770 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2771 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2772 0;
2773 }
2774
2775 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2776 {
2777 const struct vxlan_dev *vxlan = netdev_priv(dev);
2778 const struct vxlan_rdst *dst = &vxlan->default_dst;
2779 struct ifla_vxlan_port_range ports = {
2780 .low = htons(vxlan->port_min),
2781 .high = htons(vxlan->port_max),
2782 };
2783
2784 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2785 goto nla_put_failure;
2786
2787 if (!vxlan_addr_any(&dst->remote_ip)) {
2788 if (dst->remote_ip.sa.sa_family == AF_INET) {
2789 if (nla_put_be32(skb, IFLA_VXLAN_GROUP,
2790 dst->remote_ip.sin.sin_addr.s_addr))
2791 goto nla_put_failure;
2792 #if IS_ENABLED(CONFIG_IPV6)
2793 } else {
2794 if (nla_put(skb, IFLA_VXLAN_GROUP6, sizeof(struct in6_addr),
2795 &dst->remote_ip.sin6.sin6_addr))
2796 goto nla_put_failure;
2797 #endif
2798 }
2799 }
2800
2801 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2802 goto nla_put_failure;
2803
2804 if (!vxlan_addr_any(&vxlan->saddr)) {
2805 if (vxlan->saddr.sa.sa_family == AF_INET) {
2806 if (nla_put_be32(skb, IFLA_VXLAN_LOCAL,
2807 vxlan->saddr.sin.sin_addr.s_addr))
2808 goto nla_put_failure;
2809 #if IS_ENABLED(CONFIG_IPV6)
2810 } else {
2811 if (nla_put(skb, IFLA_VXLAN_LOCAL6, sizeof(struct in6_addr),
2812 &vxlan->saddr.sin6.sin6_addr))
2813 goto nla_put_failure;
2814 #endif
2815 }
2816 }
2817
2818 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2819 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2820 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2821 !!(vxlan->flags & VXLAN_F_LEARN)) ||
2822 nla_put_u8(skb, IFLA_VXLAN_PROXY,
2823 !!(vxlan->flags & VXLAN_F_PROXY)) ||
2824 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2825 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2826 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2827 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2828 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2829 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2830 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2831 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port) ||
2832 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2833 !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2834 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2835 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2836 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2837 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)))
2838 goto nla_put_failure;
2839
2840 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2841 goto nla_put_failure;
2842
2843 return 0;
2844
2845 nla_put_failure:
2846 return -EMSGSIZE;
2847 }
2848
2849 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2850 .kind = "vxlan",
2851 .maxtype = IFLA_VXLAN_MAX,
2852 .policy = vxlan_policy,
2853 .priv_size = sizeof(struct vxlan_dev),
2854 .setup = vxlan_setup,
2855 .validate = vxlan_validate,
2856 .newlink = vxlan_newlink,
2857 .dellink = vxlan_dellink,
2858 .get_size = vxlan_get_size,
2859 .fill_info = vxlan_fill_info,
2860 };
2861
2862 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
2863 struct net_device *dev)
2864 {
2865 struct vxlan_dev *vxlan, *next;
2866 LIST_HEAD(list_kill);
2867
2868 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2869 struct vxlan_rdst *dst = &vxlan->default_dst;
2870
2871 /* In case we created vxlan device with carrier
2872 * and we loose the carrier due to module unload
2873 * we also need to remove vxlan device. In other
2874 * cases, it's not necessary and remote_ifindex
2875 * is 0 here, so no matches.
2876 */
2877 if (dst->remote_ifindex == dev->ifindex)
2878 vxlan_dellink(vxlan->dev, &list_kill);
2879 }
2880
2881 unregister_netdevice_many(&list_kill);
2882 }
2883
2884 static int vxlan_lowerdev_event(struct notifier_block *unused,
2885 unsigned long event, void *ptr)
2886 {
2887 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2888 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2889
2890 if (event == NETDEV_UNREGISTER)
2891 vxlan_handle_lowerdev_unregister(vn, dev);
2892
2893 return NOTIFY_DONE;
2894 }
2895
2896 static struct notifier_block vxlan_notifier_block __read_mostly = {
2897 .notifier_call = vxlan_lowerdev_event,
2898 };
2899
2900 static __net_init int vxlan_init_net(struct net *net)
2901 {
2902 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2903 unsigned int h;
2904
2905 INIT_LIST_HEAD(&vn->vxlan_list);
2906 spin_lock_init(&vn->sock_lock);
2907
2908 for (h = 0; h < PORT_HASH_SIZE; ++h)
2909 INIT_HLIST_HEAD(&vn->sock_list[h]);
2910
2911 return 0;
2912 }
2913
2914 static void __net_exit vxlan_exit_net(struct net *net)
2915 {
2916 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2917 struct vxlan_dev *vxlan, *next;
2918 struct net_device *dev, *aux;
2919 LIST_HEAD(list);
2920
2921 rtnl_lock();
2922 for_each_netdev_safe(net, dev, aux)
2923 if (dev->rtnl_link_ops == &vxlan_link_ops)
2924 unregister_netdevice_queue(dev, &list);
2925
2926 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2927 /* If vxlan->dev is in the same netns, it has already been added
2928 * to the list by the previous loop.
2929 */
2930 if (!net_eq(dev_net(vxlan->dev), net))
2931 unregister_netdevice_queue(dev, &list);
2932 }
2933
2934 unregister_netdevice_many(&list);
2935 rtnl_unlock();
2936 }
2937
2938 static struct pernet_operations vxlan_net_ops = {
2939 .init = vxlan_init_net,
2940 .exit = vxlan_exit_net,
2941 .id = &vxlan_net_id,
2942 .size = sizeof(struct vxlan_net),
2943 };
2944
2945 static int __init vxlan_init_module(void)
2946 {
2947 int rc;
2948
2949 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
2950 if (!vxlan_wq)
2951 return -ENOMEM;
2952
2953 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
2954
2955 rc = register_pernet_subsys(&vxlan_net_ops);
2956 if (rc)
2957 goto out1;
2958
2959 rc = register_netdevice_notifier(&vxlan_notifier_block);
2960 if (rc)
2961 goto out2;
2962
2963 rc = rtnl_link_register(&vxlan_link_ops);
2964 if (rc)
2965 goto out3;
2966
2967 return 0;
2968 out3:
2969 unregister_netdevice_notifier(&vxlan_notifier_block);
2970 out2:
2971 unregister_pernet_subsys(&vxlan_net_ops);
2972 out1:
2973 destroy_workqueue(vxlan_wq);
2974 return rc;
2975 }
2976 late_initcall(vxlan_init_module);
2977
2978 static void __exit vxlan_cleanup_module(void)
2979 {
2980 rtnl_link_unregister(&vxlan_link_ops);
2981 unregister_netdevice_notifier(&vxlan_notifier_block);
2982 destroy_workqueue(vxlan_wq);
2983 unregister_pernet_subsys(&vxlan_net_ops);
2984 /* rcu_barrier() is called by netns */
2985 }
2986 module_exit(vxlan_cleanup_module);
2987
2988 MODULE_LICENSE("GPL");
2989 MODULE_VERSION(VXLAN_VERSION);
2990 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
2991 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
2992 MODULE_ALIAS_RTNL_LINK("vxlan");
ubuntu-bionic-kernel mirror