]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - drivers/net/vxlan.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
[mirror_ubuntu-artful-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/module.h>
15 #include <linux/errno.h>
16 #include <linux/slab.h>
17 #include <linux/udp.h>
18 #include <linux/igmp.h>
19 #include <linux/if_ether.h>
20 #include <linux/ethtool.h>
21 #include <net/arp.h>
22 #include <net/ndisc.h>
23 #include <net/ip.h>
24 #include <net/icmp.h>
25 #include <net/rtnetlink.h>
26 #include <net/inet_ecn.h>
27 #include <net/net_namespace.h>
28 #include <net/netns/generic.h>
29 #include <net/vxlan.h>
30
31 #if IS_ENABLED(CONFIG_IPV6)
32 #include <net/ip6_tunnel.h>
33 #include <net/ip6_checksum.h>
34 #endif
35
36 #define VXLAN_VERSION "0.1"
37
38 #define PORT_HASH_BITS 8
39 #define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
40 #define FDB_AGE_DEFAULT 300 /* 5 min */
41 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
42
43 /* UDP port for VXLAN traffic.
44 * The IANA assigned port is 4789, but the Linux default is 8472
45 * for compatibility with early adopters.
46 */
47 static unsigned short vxlan_port __read_mostly = 8472;
48 module_param_named(udp_port, vxlan_port, ushort, 0444);
49 MODULE_PARM_DESC(udp_port, "Destination UDP port");
50
51 static bool log_ecn_error = true;
52 module_param(log_ecn_error, bool, 0644);
53 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
54
55 static unsigned int vxlan_net_id;
56 static struct rtnl_link_ops vxlan_link_ops;
57
58 static const u8 all_zeros_mac[ETH_ALEN + 2];
59
60 static int vxlan_sock_add(struct vxlan_dev *vxlan);
61
62 /* per-network namespace private data for this module */
63 struct vxlan_net {
64 struct list_head vxlan_list;
65 struct hlist_head sock_list[PORT_HASH_SIZE];
66 spinlock_t sock_lock;
67 };
68
69 /* Forwarding table entry */
70 struct vxlan_fdb {
71 struct hlist_node hlist; /* linked list of entries */
72 struct rcu_head rcu;
73 unsigned long updated; /* jiffies */
74 unsigned long used;
75 struct list_head remotes;
76 u8 eth_addr[ETH_ALEN];
77 u16 state; /* see ndm_state */
78 __be32 vni;
79 u8 flags; /* see ndm_flags */
80 };
81
82 /* salt for hash table */
83 static u32 vxlan_salt __read_mostly;
84
85 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
86 {
87 return vs->flags & VXLAN_F_COLLECT_METADATA ||
88 ip_tunnel_collect_metadata();
89 }
90
91 #if IS_ENABLED(CONFIG_IPV6)
92 static inline
93 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
94 {
95 if (a->sa.sa_family != b->sa.sa_family)
96 return false;
97 if (a->sa.sa_family == AF_INET6)
98 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
99 else
100 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
101 }
102
103 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
104 {
105 if (ipa->sa.sa_family == AF_INET6)
106 return ipv6_addr_any(&ipa->sin6.sin6_addr);
107 else
108 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
109 }
110
111 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
112 {
113 if (ipa->sa.sa_family == AF_INET6)
114 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
115 else
116 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
117 }
118
119 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
120 {
121 if (nla_len(nla) >= sizeof(struct in6_addr)) {
122 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
123 ip->sa.sa_family = AF_INET6;
124 return 0;
125 } else if (nla_len(nla) >= sizeof(__be32)) {
126 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
127 ip->sa.sa_family = AF_INET;
128 return 0;
129 } else {
130 return -EAFNOSUPPORT;
131 }
132 }
133
134 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
135 const union vxlan_addr *ip)
136 {
137 if (ip->sa.sa_family == AF_INET6)
138 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
139 else
140 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
141 }
142
143 #else /* !CONFIG_IPV6 */
144
145 static inline
146 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
147 {
148 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
149 }
150
151 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
152 {
153 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
154 }
155
156 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
157 {
158 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
159 }
160
161 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
162 {
163 if (nla_len(nla) >= sizeof(struct in6_addr)) {
164 return -EAFNOSUPPORT;
165 } else if (nla_len(nla) >= sizeof(__be32)) {
166 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
167 ip->sa.sa_family = AF_INET;
168 return 0;
169 } else {
170 return -EAFNOSUPPORT;
171 }
172 }
173
174 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
175 const union vxlan_addr *ip)
176 {
177 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
178 }
179 #endif
180
181 /* Virtual Network hash table head */
182 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
183 {
184 return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
185 }
186
187 /* Socket hash table head */
188 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
189 {
190 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
191
192 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
193 }
194
195 /* First remote destination for a forwarding entry.
196 * Guaranteed to be non-NULL because remotes are never deleted.
197 */
198 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
199 {
200 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
201 }
202
203 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
204 {
205 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
206 }
207
208 /* Find VXLAN socket based on network namespace, address family and UDP port
209 * and enabled unshareable flags.
210 */
211 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
212 __be16 port, u32 flags)
213 {
214 struct vxlan_sock *vs;
215
216 flags &= VXLAN_F_RCV_FLAGS;
217
218 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
219 if (inet_sk(vs->sock->sk)->inet_sport == port &&
220 vxlan_get_sk_family(vs) == family &&
221 vs->flags == flags)
222 return vs;
223 }
224 return NULL;
225 }
226
227 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
228 {
229 struct vxlan_dev *vxlan;
230
231 /* For flow based devices, map all packets to VNI 0 */
232 if (vs->flags & VXLAN_F_COLLECT_METADATA)
233 vni = 0;
234
235 hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
236 if (vxlan->default_dst.remote_vni == vni)
237 return vxlan;
238 }
239
240 return NULL;
241 }
242
243 /* Look up VNI in a per net namespace table */
244 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
245 sa_family_t family, __be16 port,
246 u32 flags)
247 {
248 struct vxlan_sock *vs;
249
250 vs = vxlan_find_sock(net, family, port, flags);
251 if (!vs)
252 return NULL;
253
254 return vxlan_vs_find_vni(vs, vni);
255 }
256
257 /* Fill in neighbour message in skbuff. */
258 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
259 const struct vxlan_fdb *fdb,
260 u32 portid, u32 seq, int type, unsigned int flags,
261 const struct vxlan_rdst *rdst)
262 {
263 unsigned long now = jiffies;
264 struct nda_cacheinfo ci;
265 struct nlmsghdr *nlh;
266 struct ndmsg *ndm;
267 bool send_ip, send_eth;
268
269 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
270 if (nlh == NULL)
271 return -EMSGSIZE;
272
273 ndm = nlmsg_data(nlh);
274 memset(ndm, 0, sizeof(*ndm));
275
276 send_eth = send_ip = true;
277
278 if (type == RTM_GETNEIGH) {
279 ndm->ndm_family = AF_INET;
280 send_ip = !vxlan_addr_any(&rdst->remote_ip);
281 send_eth = !is_zero_ether_addr(fdb->eth_addr);
282 } else
283 ndm->ndm_family = AF_BRIDGE;
284 ndm->ndm_state = fdb->state;
285 ndm->ndm_ifindex = vxlan->dev->ifindex;
286 ndm->ndm_flags = fdb->flags;
287 ndm->ndm_type = RTN_UNICAST;
288
289 if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
290 nla_put_s32(skb, NDA_LINK_NETNSID,
291 peernet2id(dev_net(vxlan->dev), vxlan->net)))
292 goto nla_put_failure;
293
294 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
295 goto nla_put_failure;
296
297 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
298 goto nla_put_failure;
299
300 if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
301 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
302 goto nla_put_failure;
303 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
304 nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
305 goto nla_put_failure;
306 if ((vxlan->flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
307 nla_put_u32(skb, NDA_SRC_VNI,
308 be32_to_cpu(fdb->vni)))
309 goto nla_put_failure;
310 if (rdst->remote_ifindex &&
311 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
312 goto nla_put_failure;
313
314 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
315 ci.ndm_confirmed = 0;
316 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
317 ci.ndm_refcnt = 0;
318
319 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
320 goto nla_put_failure;
321
322 nlmsg_end(skb, nlh);
323 return 0;
324
325 nla_put_failure:
326 nlmsg_cancel(skb, nlh);
327 return -EMSGSIZE;
328 }
329
330 static inline size_t vxlan_nlmsg_size(void)
331 {
332 return NLMSG_ALIGN(sizeof(struct ndmsg))
333 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
334 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
335 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
336 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
337 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
338 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
339 + nla_total_size(sizeof(struct nda_cacheinfo));
340 }
341
342 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
343 struct vxlan_rdst *rd, int type)
344 {
345 struct net *net = dev_net(vxlan->dev);
346 struct sk_buff *skb;
347 int err = -ENOBUFS;
348
349 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
350 if (skb == NULL)
351 goto errout;
352
353 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
354 if (err < 0) {
355 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
356 WARN_ON(err == -EMSGSIZE);
357 kfree_skb(skb);
358 goto errout;
359 }
360
361 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
362 return;
363 errout:
364 if (err < 0)
365 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
366 }
367
368 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
369 {
370 struct vxlan_dev *vxlan = netdev_priv(dev);
371 struct vxlan_fdb f = {
372 .state = NUD_STALE,
373 };
374 struct vxlan_rdst remote = {
375 .remote_ip = *ipa, /* goes to NDA_DST */
376 .remote_vni = cpu_to_be32(VXLAN_N_VID),
377 };
378
379 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
380 }
381
382 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
383 {
384 struct vxlan_fdb f = {
385 .state = NUD_STALE,
386 };
387 struct vxlan_rdst remote = { };
388
389 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
390
391 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
392 }
393
394 /* Hash Ethernet address */
395 static u32 eth_hash(const unsigned char *addr)
396 {
397 u64 value = get_unaligned((u64 *)addr);
398
399 /* only want 6 bytes */
400 #ifdef __BIG_ENDIAN
401 value >>= 16;
402 #else
403 value <<= 16;
404 #endif
405 return hash_64(value, FDB_HASH_BITS);
406 }
407
408 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
409 {
410 /* use 1 byte of OUI and 3 bytes of NIC */
411 u32 key = get_unaligned((u32 *)(addr + 2));
412
413 return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
414 }
415
416 /* Hash chain to use given mac address */
417 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
418 const u8 *mac, __be32 vni)
419 {
420 if (vxlan->flags & VXLAN_F_COLLECT_METADATA)
421 return &vxlan->fdb_head[eth_vni_hash(mac, vni)];
422 else
423 return &vxlan->fdb_head[eth_hash(mac)];
424 }
425
426 /* Look up Ethernet address in forwarding table */
427 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
428 const u8 *mac, __be32 vni)
429 {
430 struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
431 struct vxlan_fdb *f;
432
433 hlist_for_each_entry_rcu(f, head, hlist) {
434 if (ether_addr_equal(mac, f->eth_addr)) {
435 if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
436 if (vni == f->vni)
437 return f;
438 } else {
439 return f;
440 }
441 }
442 }
443
444 return NULL;
445 }
446
447 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
448 const u8 *mac, __be32 vni)
449 {
450 struct vxlan_fdb *f;
451
452 f = __vxlan_find_mac(vxlan, mac, vni);
453 if (f)
454 f->used = jiffies;
455
456 return f;
457 }
458
459 /* caller should hold vxlan->hash_lock */
460 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
461 union vxlan_addr *ip, __be16 port,
462 __be32 vni, __u32 ifindex)
463 {
464 struct vxlan_rdst *rd;
465
466 list_for_each_entry(rd, &f->remotes, list) {
467 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
468 rd->remote_port == port &&
469 rd->remote_vni == vni &&
470 rd->remote_ifindex == ifindex)
471 return rd;
472 }
473
474 return NULL;
475 }
476
477 /* Replace destination of unicast mac */
478 static int vxlan_fdb_replace(struct vxlan_fdb *f,
479 union vxlan_addr *ip, __be16 port, __be32 vni,
480 __u32 ifindex)
481 {
482 struct vxlan_rdst *rd;
483
484 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
485 if (rd)
486 return 0;
487
488 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
489 if (!rd)
490 return 0;
491
492 dst_cache_reset(&rd->dst_cache);
493 rd->remote_ip = *ip;
494 rd->remote_port = port;
495 rd->remote_vni = vni;
496 rd->remote_ifindex = ifindex;
497 return 1;
498 }
499
500 /* Add/update destinations for multicast */
501 static int vxlan_fdb_append(struct vxlan_fdb *f,
502 union vxlan_addr *ip, __be16 port, __be32 vni,
503 __u32 ifindex, struct vxlan_rdst **rdp)
504 {
505 struct vxlan_rdst *rd;
506
507 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
508 if (rd)
509 return 0;
510
511 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
512 if (rd == NULL)
513 return -ENOBUFS;
514
515 if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
516 kfree(rd);
517 return -ENOBUFS;
518 }
519
520 rd->remote_ip = *ip;
521 rd->remote_port = port;
522 rd->remote_vni = vni;
523 rd->remote_ifindex = ifindex;
524
525 list_add_tail_rcu(&rd->list, &f->remotes);
526
527 *rdp = rd;
528 return 1;
529 }
530
531 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
532 unsigned int off,
533 struct vxlanhdr *vh, size_t hdrlen,
534 __be32 vni_field,
535 struct gro_remcsum *grc,
536 bool nopartial)
537 {
538 size_t start, offset;
539
540 if (skb->remcsum_offload)
541 return vh;
542
543 if (!NAPI_GRO_CB(skb)->csum_valid)
544 return NULL;
545
546 start = vxlan_rco_start(vni_field);
547 offset = start + vxlan_rco_offset(vni_field);
548
549 vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
550 start, offset, grc, nopartial);
551
552 skb->remcsum_offload = 1;
553
554 return vh;
555 }
556
557 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
558 struct sk_buff **head,
559 struct sk_buff *skb)
560 {
561 struct sk_buff *p, **pp = NULL;
562 struct vxlanhdr *vh, *vh2;
563 unsigned int hlen, off_vx;
564 int flush = 1;
565 struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
566 __be32 flags;
567 struct gro_remcsum grc;
568
569 skb_gro_remcsum_init(&grc);
570
571 off_vx = skb_gro_offset(skb);
572 hlen = off_vx + sizeof(*vh);
573 vh = skb_gro_header_fast(skb, off_vx);
574 if (skb_gro_header_hard(skb, hlen)) {
575 vh = skb_gro_header_slow(skb, hlen, off_vx);
576 if (unlikely(!vh))
577 goto out;
578 }
579
580 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
581
582 flags = vh->vx_flags;
583
584 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
585 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
586 vh->vx_vni, &grc,
587 !!(vs->flags &
588 VXLAN_F_REMCSUM_NOPARTIAL));
589
590 if (!vh)
591 goto out;
592 }
593
594 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
595
596 for (p = *head; p; p = p->next) {
597 if (!NAPI_GRO_CB(p)->same_flow)
598 continue;
599
600 vh2 = (struct vxlanhdr *)(p->data + off_vx);
601 if (vh->vx_flags != vh2->vx_flags ||
602 vh->vx_vni != vh2->vx_vni) {
603 NAPI_GRO_CB(p)->same_flow = 0;
604 continue;
605 }
606 }
607
608 pp = call_gro_receive(eth_gro_receive, head, skb);
609 flush = 0;
610
611 out:
612 skb_gro_remcsum_cleanup(skb, &grc);
613 NAPI_GRO_CB(skb)->flush |= flush;
614
615 return pp;
616 }
617
618 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
619 {
620 /* Sets 'skb->inner_mac_header' since we are always called with
621 * 'skb->encapsulation' set.
622 */
623 return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
624 }
625
626 /* Add new entry to forwarding table -- assumes lock held */
627 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
628 const u8 *mac, union vxlan_addr *ip,
629 __u16 state, __u16 flags,
630 __be16 port, __be32 src_vni, __be32 vni,
631 __u32 ifindex, __u8 ndm_flags)
632 {
633 struct vxlan_rdst *rd = NULL;
634 struct vxlan_fdb *f;
635 int notify = 0;
636 int rc;
637
638 f = __vxlan_find_mac(vxlan, mac, src_vni);
639 if (f) {
640 if (flags & NLM_F_EXCL) {
641 netdev_dbg(vxlan->dev,
642 "lost race to create %pM\n", mac);
643 return -EEXIST;
644 }
645 if (f->state != state) {
646 f->state = state;
647 f->updated = jiffies;
648 notify = 1;
649 }
650 if (f->flags != ndm_flags) {
651 f->flags = ndm_flags;
652 f->updated = jiffies;
653 notify = 1;
654 }
655 if ((flags & NLM_F_REPLACE)) {
656 /* Only change unicasts */
657 if (!(is_multicast_ether_addr(f->eth_addr) ||
658 is_zero_ether_addr(f->eth_addr))) {
659 notify |= vxlan_fdb_replace(f, ip, port, vni,
660 ifindex);
661 } else
662 return -EOPNOTSUPP;
663 }
664 if ((flags & NLM_F_APPEND) &&
665 (is_multicast_ether_addr(f->eth_addr) ||
666 is_zero_ether_addr(f->eth_addr))) {
667 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
668
669 if (rc < 0)
670 return rc;
671 notify |= rc;
672 }
673 } else {
674 if (!(flags & NLM_F_CREATE))
675 return -ENOENT;
676
677 if (vxlan->cfg.addrmax &&
678 vxlan->addrcnt >= vxlan->cfg.addrmax)
679 return -ENOSPC;
680
681 /* Disallow replace to add a multicast entry */
682 if ((flags & NLM_F_REPLACE) &&
683 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
684 return -EOPNOTSUPP;
685
686 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
687 f = kmalloc(sizeof(*f), GFP_ATOMIC);
688 if (!f)
689 return -ENOMEM;
690
691 notify = 1;
692 f->state = state;
693 f->flags = ndm_flags;
694 f->updated = f->used = jiffies;
695 f->vni = src_vni;
696 INIT_LIST_HEAD(&f->remotes);
697 memcpy(f->eth_addr, mac, ETH_ALEN);
698
699 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
700 if (rc < 0) {
701 kfree(f);
702 return rc;
703 }
704
705 ++vxlan->addrcnt;
706 hlist_add_head_rcu(&f->hlist,
707 vxlan_fdb_head(vxlan, mac, src_vni));
708 }
709
710 if (notify) {
711 if (rd == NULL)
712 rd = first_remote_rtnl(f);
713 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
714 }
715
716 return 0;
717 }
718
719 static void vxlan_fdb_free(struct rcu_head *head)
720 {
721 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
722 struct vxlan_rdst *rd, *nd;
723
724 list_for_each_entry_safe(rd, nd, &f->remotes, list) {
725 dst_cache_destroy(&rd->dst_cache);
726 kfree(rd);
727 }
728 kfree(f);
729 }
730
731 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
732 {
733 netdev_dbg(vxlan->dev,
734 "delete %pM\n", f->eth_addr);
735
736 --vxlan->addrcnt;
737 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
738
739 hlist_del_rcu(&f->hlist);
740 call_rcu(&f->rcu, vxlan_fdb_free);
741 }
742
743 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
744 union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
745 __be32 *vni, u32 *ifindex)
746 {
747 struct net *net = dev_net(vxlan->dev);
748 int err;
749
750 if (tb[NDA_DST]) {
751 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
752 if (err)
753 return err;
754 } else {
755 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
756 if (remote->sa.sa_family == AF_INET) {
757 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
758 ip->sa.sa_family = AF_INET;
759 #if IS_ENABLED(CONFIG_IPV6)
760 } else {
761 ip->sin6.sin6_addr = in6addr_any;
762 ip->sa.sa_family = AF_INET6;
763 #endif
764 }
765 }
766
767 if (tb[NDA_PORT]) {
768 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
769 return -EINVAL;
770 *port = nla_get_be16(tb[NDA_PORT]);
771 } else {
772 *port = vxlan->cfg.dst_port;
773 }
774
775 if (tb[NDA_VNI]) {
776 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
777 return -EINVAL;
778 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
779 } else {
780 *vni = vxlan->default_dst.remote_vni;
781 }
782
783 if (tb[NDA_SRC_VNI]) {
784 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
785 return -EINVAL;
786 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
787 } else {
788 *src_vni = vxlan->default_dst.remote_vni;
789 }
790
791 if (tb[NDA_IFINDEX]) {
792 struct net_device *tdev;
793
794 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
795 return -EINVAL;
796 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
797 tdev = __dev_get_by_index(net, *ifindex);
798 if (!tdev)
799 return -EADDRNOTAVAIL;
800 } else {
801 *ifindex = 0;
802 }
803
804 return 0;
805 }
806
807 /* Add static entry (via netlink) */
808 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
809 struct net_device *dev,
810 const unsigned char *addr, u16 vid, u16 flags)
811 {
812 struct vxlan_dev *vxlan = netdev_priv(dev);
813 /* struct net *net = dev_net(vxlan->dev); */
814 union vxlan_addr ip;
815 __be16 port;
816 __be32 src_vni, vni;
817 u32 ifindex;
818 int err;
819
820 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
821 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
822 ndm->ndm_state);
823 return -EINVAL;
824 }
825
826 if (tb[NDA_DST] == NULL)
827 return -EINVAL;
828
829 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
830 if (err)
831 return err;
832
833 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
834 return -EAFNOSUPPORT;
835
836 spin_lock_bh(&vxlan->hash_lock);
837 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
838 port, src_vni, vni, ifindex, ndm->ndm_flags);
839 spin_unlock_bh(&vxlan->hash_lock);
840
841 return err;
842 }
843
844 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
845 const unsigned char *addr, union vxlan_addr ip,
846 __be16 port, __be32 src_vni, u32 vni, u32 ifindex,
847 u16 vid)
848 {
849 struct vxlan_fdb *f;
850 struct vxlan_rdst *rd = NULL;
851 int err = -ENOENT;
852
853 f = vxlan_find_mac(vxlan, addr, src_vni);
854 if (!f)
855 return err;
856
857 if (!vxlan_addr_any(&ip)) {
858 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
859 if (!rd)
860 goto out;
861 }
862
863 /* remove a destination if it's not the only one on the list,
864 * otherwise destroy the fdb entry
865 */
866 if (rd && !list_is_singular(&f->remotes)) {
867 list_del_rcu(&rd->list);
868 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
869 kfree_rcu(rd, rcu);
870 goto out;
871 }
872
873 vxlan_fdb_destroy(vxlan, f);
874
875 out:
876 return 0;
877 }
878
879 /* Delete entry (via netlink) */
880 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
881 struct net_device *dev,
882 const unsigned char *addr, u16 vid)
883 {
884 struct vxlan_dev *vxlan = netdev_priv(dev);
885 union vxlan_addr ip;
886 __be32 src_vni, vni;
887 __be16 port;
888 u32 ifindex;
889 int err;
890
891 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
892 if (err)
893 return err;
894
895 spin_lock_bh(&vxlan->hash_lock);
896 err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
897 vid);
898 spin_unlock_bh(&vxlan->hash_lock);
899
900 return err;
901 }
902
903 /* Dump forwarding table */
904 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
905 struct net_device *dev,
906 struct net_device *filter_dev, int *idx)
907 {
908 struct vxlan_dev *vxlan = netdev_priv(dev);
909 unsigned int h;
910 int err = 0;
911
912 for (h = 0; h < FDB_HASH_SIZE; ++h) {
913 struct vxlan_fdb *f;
914
915 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
916 struct vxlan_rdst *rd;
917
918 list_for_each_entry_rcu(rd, &f->remotes, list) {
919 if (*idx < cb->args[2])
920 goto skip;
921
922 err = vxlan_fdb_info(skb, vxlan, f,
923 NETLINK_CB(cb->skb).portid,
924 cb->nlh->nlmsg_seq,
925 RTM_NEWNEIGH,
926 NLM_F_MULTI, rd);
927 if (err < 0)
928 goto out;
929 skip:
930 *idx += 1;
931 }
932 }
933 }
934 out:
935 return err;
936 }
937
938 /* Watch incoming packets to learn mapping between Ethernet address
939 * and Tunnel endpoint.
940 * Return true if packet is bogus and should be dropped.
941 */
942 static bool vxlan_snoop(struct net_device *dev,
943 union vxlan_addr *src_ip, const u8 *src_mac,
944 __be32 vni)
945 {
946 struct vxlan_dev *vxlan = netdev_priv(dev);
947 struct vxlan_fdb *f;
948
949 f = vxlan_find_mac(vxlan, src_mac, vni);
950 if (likely(f)) {
951 struct vxlan_rdst *rdst = first_remote_rcu(f);
952
953 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
954 return false;
955
956 /* Don't migrate static entries, drop packets */
957 if (f->state & NUD_NOARP)
958 return true;
959
960 if (net_ratelimit())
961 netdev_info(dev,
962 "%pM migrated from %pIS to %pIS\n",
963 src_mac, &rdst->remote_ip.sa, &src_ip->sa);
964
965 rdst->remote_ip = *src_ip;
966 f->updated = jiffies;
967 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
968 } else {
969 /* learned new entry */
970 spin_lock(&vxlan->hash_lock);
971
972 /* close off race between vxlan_flush and incoming packets */
973 if (netif_running(dev))
974 vxlan_fdb_create(vxlan, src_mac, src_ip,
975 NUD_REACHABLE,
976 NLM_F_EXCL|NLM_F_CREATE,
977 vxlan->cfg.dst_port,
978 vni,
979 vxlan->default_dst.remote_vni,
980 0, NTF_SELF);
981 spin_unlock(&vxlan->hash_lock);
982 }
983
984 return false;
985 }
986
987 /* See if multicast group is already in use by other ID */
988 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
989 {
990 struct vxlan_dev *vxlan;
991 struct vxlan_sock *sock4;
992 #if IS_ENABLED(CONFIG_IPV6)
993 struct vxlan_sock *sock6;
994 #endif
995 unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
996
997 sock4 = rtnl_dereference(dev->vn4_sock);
998
999 /* The vxlan_sock is only used by dev, leaving group has
1000 * no effect on other vxlan devices.
1001 */
1002 if (family == AF_INET && sock4 && atomic_read(&sock4->refcnt) == 1)
1003 return false;
1004 #if IS_ENABLED(CONFIG_IPV6)
1005 sock6 = rtnl_dereference(dev->vn6_sock);
1006 if (family == AF_INET6 && sock6 && atomic_read(&sock6->refcnt) == 1)
1007 return false;
1008 #endif
1009
1010 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1011 if (!netif_running(vxlan->dev) || vxlan == dev)
1012 continue;
1013
1014 if (family == AF_INET &&
1015 rtnl_dereference(vxlan->vn4_sock) != sock4)
1016 continue;
1017 #if IS_ENABLED(CONFIG_IPV6)
1018 if (family == AF_INET6 &&
1019 rtnl_dereference(vxlan->vn6_sock) != sock6)
1020 continue;
1021 #endif
1022
1023 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1024 &dev->default_dst.remote_ip))
1025 continue;
1026
1027 if (vxlan->default_dst.remote_ifindex !=
1028 dev->default_dst.remote_ifindex)
1029 continue;
1030
1031 return true;
1032 }
1033
1034 return false;
1035 }
1036
1037 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1038 {
1039 struct vxlan_net *vn;
1040
1041 if (!vs)
1042 return false;
1043 if (!atomic_dec_and_test(&vs->refcnt))
1044 return false;
1045
1046 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1047 spin_lock(&vn->sock_lock);
1048 hlist_del_rcu(&vs->hlist);
1049 udp_tunnel_notify_del_rx_port(vs->sock,
1050 (vs->flags & VXLAN_F_GPE) ?
1051 UDP_TUNNEL_TYPE_VXLAN_GPE :
1052 UDP_TUNNEL_TYPE_VXLAN);
1053 spin_unlock(&vn->sock_lock);
1054
1055 return true;
1056 }
1057
1058 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1059 {
1060 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1061 #if IS_ENABLED(CONFIG_IPV6)
1062 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1063
1064 rcu_assign_pointer(vxlan->vn6_sock, NULL);
1065 #endif
1066
1067 rcu_assign_pointer(vxlan->vn4_sock, NULL);
1068 synchronize_net();
1069
1070 if (__vxlan_sock_release_prep(sock4)) {
1071 udp_tunnel_sock_release(sock4->sock);
1072 kfree(sock4);
1073 }
1074
1075 #if IS_ENABLED(CONFIG_IPV6)
1076 if (__vxlan_sock_release_prep(sock6)) {
1077 udp_tunnel_sock_release(sock6->sock);
1078 kfree(sock6);
1079 }
1080 #endif
1081 }
1082
1083 /* Update multicast group membership when first VNI on
1084 * multicast address is brought up
1085 */
1086 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1087 {
1088 struct sock *sk;
1089 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1090 int ifindex = vxlan->default_dst.remote_ifindex;
1091 int ret = -EINVAL;
1092
1093 if (ip->sa.sa_family == AF_INET) {
1094 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1095 struct ip_mreqn mreq = {
1096 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1097 .imr_ifindex = ifindex,
1098 };
1099
1100 sk = sock4->sock->sk;
1101 lock_sock(sk);
1102 ret = ip_mc_join_group(sk, &mreq);
1103 release_sock(sk);
1104 #if IS_ENABLED(CONFIG_IPV6)
1105 } else {
1106 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1107
1108 sk = sock6->sock->sk;
1109 lock_sock(sk);
1110 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1111 &ip->sin6.sin6_addr);
1112 release_sock(sk);
1113 #endif
1114 }
1115
1116 return ret;
1117 }
1118
1119 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1120 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1121 {
1122 struct sock *sk;
1123 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1124 int ifindex = vxlan->default_dst.remote_ifindex;
1125 int ret = -EINVAL;
1126
1127 if (ip->sa.sa_family == AF_INET) {
1128 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1129 struct ip_mreqn mreq = {
1130 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1131 .imr_ifindex = ifindex,
1132 };
1133
1134 sk = sock4->sock->sk;
1135 lock_sock(sk);
1136 ret = ip_mc_leave_group(sk, &mreq);
1137 release_sock(sk);
1138 #if IS_ENABLED(CONFIG_IPV6)
1139 } else {
1140 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1141
1142 sk = sock6->sock->sk;
1143 lock_sock(sk);
1144 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1145 &ip->sin6.sin6_addr);
1146 release_sock(sk);
1147 #endif
1148 }
1149
1150 return ret;
1151 }
1152
1153 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1154 struct sk_buff *skb, u32 vxflags)
1155 {
1156 size_t start, offset;
1157
1158 if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1159 goto out;
1160
1161 start = vxlan_rco_start(unparsed->vx_vni);
1162 offset = start + vxlan_rco_offset(unparsed->vx_vni);
1163
1164 if (!pskb_may_pull(skb, offset + sizeof(u16)))
1165 return false;
1166
1167 skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1168 !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1169 out:
1170 unparsed->vx_flags &= ~VXLAN_HF_RCO;
1171 unparsed->vx_vni &= VXLAN_VNI_MASK;
1172 return true;
1173 }
1174
1175 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1176 struct sk_buff *skb, u32 vxflags,
1177 struct vxlan_metadata *md)
1178 {
1179 struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1180 struct metadata_dst *tun_dst;
1181
1182 if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1183 goto out;
1184
1185 md->gbp = ntohs(gbp->policy_id);
1186
1187 tun_dst = (struct metadata_dst *)skb_dst(skb);
1188 if (tun_dst) {
1189 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1190 tun_dst->u.tun_info.options_len = sizeof(*md);
1191 }
1192 if (gbp->dont_learn)
1193 md->gbp |= VXLAN_GBP_DONT_LEARN;
1194
1195 if (gbp->policy_applied)
1196 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1197
1198 /* In flow-based mode, GBP is carried in dst_metadata */
1199 if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1200 skb->mark = md->gbp;
1201 out:
1202 unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1203 }
1204
1205 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1206 __be16 *protocol,
1207 struct sk_buff *skb, u32 vxflags)
1208 {
1209 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1210
1211 /* Need to have Next Protocol set for interfaces in GPE mode. */
1212 if (!gpe->np_applied)
1213 return false;
1214 /* "The initial version is 0. If a receiver does not support the
1215 * version indicated it MUST drop the packet.
1216 */
1217 if (gpe->version != 0)
1218 return false;
1219 /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1220 * processing MUST occur." However, we don't implement OAM
1221 * processing, thus drop the packet.
1222 */
1223 if (gpe->oam_flag)
1224 return false;
1225
1226 switch (gpe->next_protocol) {
1227 case VXLAN_GPE_NP_IPV4:
1228 *protocol = htons(ETH_P_IP);
1229 break;
1230 case VXLAN_GPE_NP_IPV6:
1231 *protocol = htons(ETH_P_IPV6);
1232 break;
1233 case VXLAN_GPE_NP_ETHERNET:
1234 *protocol = htons(ETH_P_TEB);
1235 break;
1236 default:
1237 return false;
1238 }
1239
1240 unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1241 return true;
1242 }
1243
1244 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1245 struct vxlan_sock *vs,
1246 struct sk_buff *skb, __be32 vni)
1247 {
1248 union vxlan_addr saddr;
1249
1250 skb_reset_mac_header(skb);
1251 skb->protocol = eth_type_trans(skb, vxlan->dev);
1252 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1253
1254 /* Ignore packet loops (and multicast echo) */
1255 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1256 return false;
1257
1258 /* Get address from the outer IP header */
1259 if (vxlan_get_sk_family(vs) == AF_INET) {
1260 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1261 saddr.sa.sa_family = AF_INET;
1262 #if IS_ENABLED(CONFIG_IPV6)
1263 } else {
1264 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1265 saddr.sa.sa_family = AF_INET6;
1266 #endif
1267 }
1268
1269 if ((vxlan->flags & VXLAN_F_LEARN) &&
1270 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, vni))
1271 return false;
1272
1273 return true;
1274 }
1275
1276 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1277 struct sk_buff *skb)
1278 {
1279 int err = 0;
1280
1281 if (vxlan_get_sk_family(vs) == AF_INET)
1282 err = IP_ECN_decapsulate(oiph, skb);
1283 #if IS_ENABLED(CONFIG_IPV6)
1284 else
1285 err = IP6_ECN_decapsulate(oiph, skb);
1286 #endif
1287
1288 if (unlikely(err) && log_ecn_error) {
1289 if (vxlan_get_sk_family(vs) == AF_INET)
1290 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1291 &((struct iphdr *)oiph)->saddr,
1292 ((struct iphdr *)oiph)->tos);
1293 else
1294 net_info_ratelimited("non-ECT from %pI6\n",
1295 &((struct ipv6hdr *)oiph)->saddr);
1296 }
1297 return err <= 1;
1298 }
1299
1300 /* Callback from net/ipv4/udp.c to receive packets */
1301 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1302 {
1303 struct pcpu_sw_netstats *stats;
1304 struct vxlan_dev *vxlan;
1305 struct vxlan_sock *vs;
1306 struct vxlanhdr unparsed;
1307 struct vxlan_metadata _md;
1308 struct vxlan_metadata *md = &_md;
1309 __be16 protocol = htons(ETH_P_TEB);
1310 bool raw_proto = false;
1311 void *oiph;
1312 __be32 vni = 0;
1313
1314 /* Need UDP and VXLAN header to be present */
1315 if (!pskb_may_pull(skb, VXLAN_HLEN))
1316 goto drop;
1317
1318 unparsed = *vxlan_hdr(skb);
1319 /* VNI flag always required to be set */
1320 if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1321 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1322 ntohl(vxlan_hdr(skb)->vx_flags),
1323 ntohl(vxlan_hdr(skb)->vx_vni));
1324 /* Return non vxlan pkt */
1325 goto drop;
1326 }
1327 unparsed.vx_flags &= ~VXLAN_HF_VNI;
1328 unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1329
1330 vs = rcu_dereference_sk_user_data(sk);
1331 if (!vs)
1332 goto drop;
1333
1334 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1335
1336 if ((vs->flags & VXLAN_F_COLLECT_METADATA) && !vni)
1337 goto drop;
1338
1339 vxlan = vxlan_vs_find_vni(vs, vni);
1340 if (!vxlan)
1341 goto drop;
1342
1343 /* For backwards compatibility, only allow reserved fields to be
1344 * used by VXLAN extensions if explicitly requested.
1345 */
1346 if (vs->flags & VXLAN_F_GPE) {
1347 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1348 goto drop;
1349 raw_proto = true;
1350 }
1351
1352 if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1353 !net_eq(vxlan->net, dev_net(vxlan->dev))))
1354 goto drop;
1355
1356 if (vxlan_collect_metadata(vs)) {
1357 struct metadata_dst *tun_dst;
1358
1359 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1360 key32_to_tunnel_id(vni), sizeof(*md));
1361
1362 if (!tun_dst)
1363 goto drop;
1364
1365 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1366
1367 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1368 } else {
1369 memset(md, 0, sizeof(*md));
1370 }
1371
1372 if (vs->flags & VXLAN_F_REMCSUM_RX)
1373 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1374 goto drop;
1375 if (vs->flags & VXLAN_F_GBP)
1376 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1377 /* Note that GBP and GPE can never be active together. This is
1378 * ensured in vxlan_dev_configure.
1379 */
1380
1381 if (unparsed.vx_flags || unparsed.vx_vni) {
1382 /* If there are any unprocessed flags remaining treat
1383 * this as a malformed packet. This behavior diverges from
1384 * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1385 * in reserved fields are to be ignored. The approach here
1386 * maintains compatibility with previous stack code, and also
1387 * is more robust and provides a little more security in
1388 * adding extensions to VXLAN.
1389 */
1390 goto drop;
1391 }
1392
1393 if (!raw_proto) {
1394 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1395 goto drop;
1396 } else {
1397 skb_reset_mac_header(skb);
1398 skb->dev = vxlan->dev;
1399 skb->pkt_type = PACKET_HOST;
1400 }
1401
1402 oiph = skb_network_header(skb);
1403 skb_reset_network_header(skb);
1404
1405 if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1406 ++vxlan->dev->stats.rx_frame_errors;
1407 ++vxlan->dev->stats.rx_errors;
1408 goto drop;
1409 }
1410
1411 stats = this_cpu_ptr(vxlan->dev->tstats);
1412 u64_stats_update_begin(&stats->syncp);
1413 stats->rx_packets++;
1414 stats->rx_bytes += skb->len;
1415 u64_stats_update_end(&stats->syncp);
1416
1417 gro_cells_receive(&vxlan->gro_cells, skb);
1418 return 0;
1419
1420 drop:
1421 /* Consume bad packet */
1422 kfree_skb(skb);
1423 return 0;
1424 }
1425
1426 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1427 {
1428 struct vxlan_dev *vxlan = netdev_priv(dev);
1429 struct arphdr *parp;
1430 u8 *arpptr, *sha;
1431 __be32 sip, tip;
1432 struct neighbour *n;
1433
1434 if (dev->flags & IFF_NOARP)
1435 goto out;
1436
1437 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1438 dev->stats.tx_dropped++;
1439 goto out;
1440 }
1441 parp = arp_hdr(skb);
1442
1443 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1444 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1445 parp->ar_pro != htons(ETH_P_IP) ||
1446 parp->ar_op != htons(ARPOP_REQUEST) ||
1447 parp->ar_hln != dev->addr_len ||
1448 parp->ar_pln != 4)
1449 goto out;
1450 arpptr = (u8 *)parp + sizeof(struct arphdr);
1451 sha = arpptr;
1452 arpptr += dev->addr_len; /* sha */
1453 memcpy(&sip, arpptr, sizeof(sip));
1454 arpptr += sizeof(sip);
1455 arpptr += dev->addr_len; /* tha */
1456 memcpy(&tip, arpptr, sizeof(tip));
1457
1458 if (ipv4_is_loopback(tip) ||
1459 ipv4_is_multicast(tip))
1460 goto out;
1461
1462 n = neigh_lookup(&arp_tbl, &tip, dev);
1463
1464 if (n) {
1465 struct vxlan_fdb *f;
1466 struct sk_buff *reply;
1467
1468 if (!(n->nud_state & NUD_CONNECTED)) {
1469 neigh_release(n);
1470 goto out;
1471 }
1472
1473 f = vxlan_find_mac(vxlan, n->ha, vni);
1474 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1475 /* bridge-local neighbor */
1476 neigh_release(n);
1477 goto out;
1478 }
1479
1480 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1481 n->ha, sha);
1482
1483 neigh_release(n);
1484
1485 if (reply == NULL)
1486 goto out;
1487
1488 skb_reset_mac_header(reply);
1489 __skb_pull(reply, skb_network_offset(reply));
1490 reply->ip_summed = CHECKSUM_UNNECESSARY;
1491 reply->pkt_type = PACKET_HOST;
1492
1493 if (netif_rx_ni(reply) == NET_RX_DROP)
1494 dev->stats.rx_dropped++;
1495 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1496 union vxlan_addr ipa = {
1497 .sin.sin_addr.s_addr = tip,
1498 .sin.sin_family = AF_INET,
1499 };
1500
1501 vxlan_ip_miss(dev, &ipa);
1502 }
1503 out:
1504 consume_skb(skb);
1505 return NETDEV_TX_OK;
1506 }
1507
1508 #if IS_ENABLED(CONFIG_IPV6)
1509 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1510 struct neighbour *n, bool isrouter)
1511 {
1512 struct net_device *dev = request->dev;
1513 struct sk_buff *reply;
1514 struct nd_msg *ns, *na;
1515 struct ipv6hdr *pip6;
1516 u8 *daddr;
1517 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1518 int ns_olen;
1519 int i, len;
1520
1521 if (dev == NULL)
1522 return NULL;
1523
1524 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1525 sizeof(*na) + na_olen + dev->needed_tailroom;
1526 reply = alloc_skb(len, GFP_ATOMIC);
1527 if (reply == NULL)
1528 return NULL;
1529
1530 reply->protocol = htons(ETH_P_IPV6);
1531 reply->dev = dev;
1532 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1533 skb_push(reply, sizeof(struct ethhdr));
1534 skb_reset_mac_header(reply);
1535
1536 ns = (struct nd_msg *)skb_transport_header(request);
1537
1538 daddr = eth_hdr(request)->h_source;
1539 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1540 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1541 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1542 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1543 break;
1544 }
1545 }
1546
1547 /* Ethernet header */
1548 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1549 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1550 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1551 reply->protocol = htons(ETH_P_IPV6);
1552
1553 skb_pull(reply, sizeof(struct ethhdr));
1554 skb_reset_network_header(reply);
1555 skb_put(reply, sizeof(struct ipv6hdr));
1556
1557 /* IPv6 header */
1558
1559 pip6 = ipv6_hdr(reply);
1560 memset(pip6, 0, sizeof(struct ipv6hdr));
1561 pip6->version = 6;
1562 pip6->priority = ipv6_hdr(request)->priority;
1563 pip6->nexthdr = IPPROTO_ICMPV6;
1564 pip6->hop_limit = 255;
1565 pip6->daddr = ipv6_hdr(request)->saddr;
1566 pip6->saddr = *(struct in6_addr *)n->primary_key;
1567
1568 skb_pull(reply, sizeof(struct ipv6hdr));
1569 skb_reset_transport_header(reply);
1570
1571 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1572
1573 /* Neighbor Advertisement */
1574 memset(na, 0, sizeof(*na)+na_olen);
1575 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1576 na->icmph.icmp6_router = isrouter;
1577 na->icmph.icmp6_override = 1;
1578 na->icmph.icmp6_solicited = 1;
1579 na->target = ns->target;
1580 ether_addr_copy(&na->opt[2], n->ha);
1581 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1582 na->opt[1] = na_olen >> 3;
1583
1584 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1585 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1586 csum_partial(na, sizeof(*na)+na_olen, 0));
1587
1588 pip6->payload_len = htons(sizeof(*na)+na_olen);
1589
1590 skb_push(reply, sizeof(struct ipv6hdr));
1591
1592 reply->ip_summed = CHECKSUM_UNNECESSARY;
1593
1594 return reply;
1595 }
1596
1597 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1598 {
1599 struct vxlan_dev *vxlan = netdev_priv(dev);
1600 struct nd_msg *msg;
1601 const struct ipv6hdr *iphdr;
1602 const struct in6_addr *saddr, *daddr;
1603 struct neighbour *n;
1604 struct inet6_dev *in6_dev;
1605
1606 in6_dev = __in6_dev_get(dev);
1607 if (!in6_dev)
1608 goto out;
1609
1610 iphdr = ipv6_hdr(skb);
1611 saddr = &iphdr->saddr;
1612 daddr = &iphdr->daddr;
1613
1614 msg = (struct nd_msg *)skb_transport_header(skb);
1615 if (msg->icmph.icmp6_code != 0 ||
1616 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1617 goto out;
1618
1619 if (ipv6_addr_loopback(daddr) ||
1620 ipv6_addr_is_multicast(&msg->target))
1621 goto out;
1622
1623 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1624
1625 if (n) {
1626 struct vxlan_fdb *f;
1627 struct sk_buff *reply;
1628
1629 if (!(n->nud_state & NUD_CONNECTED)) {
1630 neigh_release(n);
1631 goto out;
1632 }
1633
1634 f = vxlan_find_mac(vxlan, n->ha, vni);
1635 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1636 /* bridge-local neighbor */
1637 neigh_release(n);
1638 goto out;
1639 }
1640
1641 reply = vxlan_na_create(skb, n,
1642 !!(f ? f->flags & NTF_ROUTER : 0));
1643
1644 neigh_release(n);
1645
1646 if (reply == NULL)
1647 goto out;
1648
1649 if (netif_rx_ni(reply) == NET_RX_DROP)
1650 dev->stats.rx_dropped++;
1651
1652 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1653 union vxlan_addr ipa = {
1654 .sin6.sin6_addr = msg->target,
1655 .sin6.sin6_family = AF_INET6,
1656 };
1657
1658 vxlan_ip_miss(dev, &ipa);
1659 }
1660
1661 out:
1662 consume_skb(skb);
1663 return NETDEV_TX_OK;
1664 }
1665 #endif
1666
1667 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1668 {
1669 struct vxlan_dev *vxlan = netdev_priv(dev);
1670 struct neighbour *n;
1671
1672 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1673 return false;
1674
1675 n = NULL;
1676 switch (ntohs(eth_hdr(skb)->h_proto)) {
1677 case ETH_P_IP:
1678 {
1679 struct iphdr *pip;
1680
1681 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1682 return false;
1683 pip = ip_hdr(skb);
1684 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1685 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1686 union vxlan_addr ipa = {
1687 .sin.sin_addr.s_addr = pip->daddr,
1688 .sin.sin_family = AF_INET,
1689 };
1690
1691 vxlan_ip_miss(dev, &ipa);
1692 return false;
1693 }
1694
1695 break;
1696 }
1697 #if IS_ENABLED(CONFIG_IPV6)
1698 case ETH_P_IPV6:
1699 {
1700 struct ipv6hdr *pip6;
1701
1702 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1703 return false;
1704 pip6 = ipv6_hdr(skb);
1705 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1706 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1707 union vxlan_addr ipa = {
1708 .sin6.sin6_addr = pip6->daddr,
1709 .sin6.sin6_family = AF_INET6,
1710 };
1711
1712 vxlan_ip_miss(dev, &ipa);
1713 return false;
1714 }
1715
1716 break;
1717 }
1718 #endif
1719 default:
1720 return false;
1721 }
1722
1723 if (n) {
1724 bool diff;
1725
1726 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1727 if (diff) {
1728 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1729 dev->addr_len);
1730 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1731 }
1732 neigh_release(n);
1733 return diff;
1734 }
1735
1736 return false;
1737 }
1738
1739 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1740 struct vxlan_metadata *md)
1741 {
1742 struct vxlanhdr_gbp *gbp;
1743
1744 if (!md->gbp)
1745 return;
1746
1747 gbp = (struct vxlanhdr_gbp *)vxh;
1748 vxh->vx_flags |= VXLAN_HF_GBP;
1749
1750 if (md->gbp & VXLAN_GBP_DONT_LEARN)
1751 gbp->dont_learn = 1;
1752
1753 if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1754 gbp->policy_applied = 1;
1755
1756 gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1757 }
1758
1759 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1760 __be16 protocol)
1761 {
1762 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1763
1764 gpe->np_applied = 1;
1765
1766 switch (protocol) {
1767 case htons(ETH_P_IP):
1768 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1769 return 0;
1770 case htons(ETH_P_IPV6):
1771 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1772 return 0;
1773 case htons(ETH_P_TEB):
1774 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1775 return 0;
1776 }
1777 return -EPFNOSUPPORT;
1778 }
1779
1780 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1781 int iphdr_len, __be32 vni,
1782 struct vxlan_metadata *md, u32 vxflags,
1783 bool udp_sum)
1784 {
1785 struct vxlanhdr *vxh;
1786 int min_headroom;
1787 int err;
1788 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1789 __be16 inner_protocol = htons(ETH_P_TEB);
1790
1791 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1792 skb->ip_summed == CHECKSUM_PARTIAL) {
1793 int csum_start = skb_checksum_start_offset(skb);
1794
1795 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1796 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1797 (skb->csum_offset == offsetof(struct udphdr, check) ||
1798 skb->csum_offset == offsetof(struct tcphdr, check)))
1799 type |= SKB_GSO_TUNNEL_REMCSUM;
1800 }
1801
1802 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1803 + VXLAN_HLEN + iphdr_len;
1804
1805 /* Need space for new headers (invalidates iph ptr) */
1806 err = skb_cow_head(skb, min_headroom);
1807 if (unlikely(err))
1808 return err;
1809
1810 err = iptunnel_handle_offloads(skb, type);
1811 if (err)
1812 return err;
1813
1814 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1815 vxh->vx_flags = VXLAN_HF_VNI;
1816 vxh->vx_vni = vxlan_vni_field(vni);
1817
1818 if (type & SKB_GSO_TUNNEL_REMCSUM) {
1819 unsigned int start;
1820
1821 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1822 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1823 vxh->vx_flags |= VXLAN_HF_RCO;
1824
1825 if (!skb_is_gso(skb)) {
1826 skb->ip_summed = CHECKSUM_NONE;
1827 skb->encapsulation = 0;
1828 }
1829 }
1830
1831 if (vxflags & VXLAN_F_GBP)
1832 vxlan_build_gbp_hdr(vxh, vxflags, md);
1833 if (vxflags & VXLAN_F_GPE) {
1834 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1835 if (err < 0)
1836 return err;
1837 inner_protocol = skb->protocol;
1838 }
1839
1840 skb_set_inner_protocol(skb, inner_protocol);
1841 return 0;
1842 }
1843
1844 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1845 struct vxlan_sock *sock4,
1846 struct sk_buff *skb, int oif, u8 tos,
1847 __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1848 struct dst_cache *dst_cache,
1849 const struct ip_tunnel_info *info)
1850 {
1851 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1852 struct rtable *rt = NULL;
1853 struct flowi4 fl4;
1854
1855 if (!sock4)
1856 return ERR_PTR(-EIO);
1857
1858 if (tos && !info)
1859 use_cache = false;
1860 if (use_cache) {
1861 rt = dst_cache_get_ip4(dst_cache, saddr);
1862 if (rt)
1863 return rt;
1864 }
1865
1866 memset(&fl4, 0, sizeof(fl4));
1867 fl4.flowi4_oif = oif;
1868 fl4.flowi4_tos = RT_TOS(tos);
1869 fl4.flowi4_mark = skb->mark;
1870 fl4.flowi4_proto = IPPROTO_UDP;
1871 fl4.daddr = daddr;
1872 fl4.saddr = *saddr;
1873 fl4.fl4_dport = dport;
1874 fl4.fl4_sport = sport;
1875
1876 rt = ip_route_output_key(vxlan->net, &fl4);
1877 if (likely(!IS_ERR(rt))) {
1878 if (rt->dst.dev == dev) {
1879 netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1880 ip_rt_put(rt);
1881 return ERR_PTR(-ELOOP);
1882 }
1883
1884 *saddr = fl4.saddr;
1885 if (use_cache)
1886 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1887 } else {
1888 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1889 return ERR_PTR(-ENETUNREACH);
1890 }
1891 return rt;
1892 }
1893
1894 #if IS_ENABLED(CONFIG_IPV6)
1895 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1896 struct net_device *dev,
1897 struct vxlan_sock *sock6,
1898 struct sk_buff *skb, int oif, u8 tos,
1899 __be32 label,
1900 const struct in6_addr *daddr,
1901 struct in6_addr *saddr,
1902 __be16 dport, __be16 sport,
1903 struct dst_cache *dst_cache,
1904 const struct ip_tunnel_info *info)
1905 {
1906 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1907 struct dst_entry *ndst;
1908 struct flowi6 fl6;
1909 int err;
1910
1911 if (!sock6)
1912 return ERR_PTR(-EIO);
1913
1914 if (tos && !info)
1915 use_cache = false;
1916 if (use_cache) {
1917 ndst = dst_cache_get_ip6(dst_cache, saddr);
1918 if (ndst)
1919 return ndst;
1920 }
1921
1922 memset(&fl6, 0, sizeof(fl6));
1923 fl6.flowi6_oif = oif;
1924 fl6.daddr = *daddr;
1925 fl6.saddr = *saddr;
1926 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1927 fl6.flowi6_mark = skb->mark;
1928 fl6.flowi6_proto = IPPROTO_UDP;
1929 fl6.fl6_dport = dport;
1930 fl6.fl6_sport = sport;
1931
1932 err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1933 sock6->sock->sk,
1934 &ndst, &fl6);
1935 if (unlikely(err < 0)) {
1936 netdev_dbg(dev, "no route to %pI6\n", daddr);
1937 return ERR_PTR(-ENETUNREACH);
1938 }
1939
1940 if (unlikely(ndst->dev == dev)) {
1941 netdev_dbg(dev, "circular route to %pI6\n", daddr);
1942 dst_release(ndst);
1943 return ERR_PTR(-ELOOP);
1944 }
1945
1946 *saddr = fl6.saddr;
1947 if (use_cache)
1948 dst_cache_set_ip6(dst_cache, ndst, saddr);
1949 return ndst;
1950 }
1951 #endif
1952
1953 /* Bypass encapsulation if the destination is local */
1954 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1955 struct vxlan_dev *dst_vxlan, __be32 vni)
1956 {
1957 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1958 union vxlan_addr loopback;
1959 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1960 struct net_device *dev = skb->dev;
1961 int len = skb->len;
1962
1963 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1964 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1965 skb->pkt_type = PACKET_HOST;
1966 skb->encapsulation = 0;
1967 skb->dev = dst_vxlan->dev;
1968 __skb_pull(skb, skb_network_offset(skb));
1969
1970 if (remote_ip->sa.sa_family == AF_INET) {
1971 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1972 loopback.sa.sa_family = AF_INET;
1973 #if IS_ENABLED(CONFIG_IPV6)
1974 } else {
1975 loopback.sin6.sin6_addr = in6addr_loopback;
1976 loopback.sa.sa_family = AF_INET6;
1977 #endif
1978 }
1979
1980 if (dst_vxlan->flags & VXLAN_F_LEARN)
1981 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source, vni);
1982
1983 u64_stats_update_begin(&tx_stats->syncp);
1984 tx_stats->tx_packets++;
1985 tx_stats->tx_bytes += len;
1986 u64_stats_update_end(&tx_stats->syncp);
1987
1988 if (netif_rx(skb) == NET_RX_SUCCESS) {
1989 u64_stats_update_begin(&rx_stats->syncp);
1990 rx_stats->rx_packets++;
1991 rx_stats->rx_bytes += len;
1992 u64_stats_update_end(&rx_stats->syncp);
1993 } else {
1994 dev->stats.rx_dropped++;
1995 }
1996 }
1997
1998 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
1999 struct vxlan_dev *vxlan, union vxlan_addr *daddr,
2000 __be16 dst_port, __be32 vni, struct dst_entry *dst,
2001 u32 rt_flags)
2002 {
2003 #if IS_ENABLED(CONFIG_IPV6)
2004 /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2005 * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2006 * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2007 */
2008 BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2009 #endif
2010 /* Bypass encapsulation if the destination is local */
2011 if (rt_flags & RTCF_LOCAL &&
2012 !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2013 struct vxlan_dev *dst_vxlan;
2014
2015 dst_release(dst);
2016 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2017 daddr->sa.sa_family, dst_port,
2018 vxlan->flags);
2019 if (!dst_vxlan) {
2020 dev->stats.tx_errors++;
2021 kfree_skb(skb);
2022
2023 return -ENOENT;
2024 }
2025 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2026 return 1;
2027 }
2028
2029 return 0;
2030 }
2031
2032 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2033 __be32 default_vni, struct vxlan_rdst *rdst,
2034 bool did_rsc)
2035 {
2036 struct dst_cache *dst_cache;
2037 struct ip_tunnel_info *info;
2038 struct vxlan_dev *vxlan = netdev_priv(dev);
2039 const struct iphdr *old_iph = ip_hdr(skb);
2040 union vxlan_addr *dst;
2041 union vxlan_addr remote_ip, local_ip;
2042 union vxlan_addr *src;
2043 struct vxlan_metadata _md;
2044 struct vxlan_metadata *md = &_md;
2045 __be16 src_port = 0, dst_port;
2046 struct dst_entry *ndst = NULL;
2047 __be32 vni, label;
2048 __u8 tos, ttl;
2049 int err;
2050 u32 flags = vxlan->flags;
2051 bool udp_sum = false;
2052 bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2053
2054 info = skb_tunnel_info(skb);
2055
2056 if (rdst) {
2057 dst = &rdst->remote_ip;
2058 if (vxlan_addr_any(dst)) {
2059 if (did_rsc) {
2060 /* short-circuited back to local bridge */
2061 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2062 return;
2063 }
2064 goto drop;
2065 }
2066
2067 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2068 vni = (rdst->remote_vni) ? : default_vni;
2069 src = &vxlan->cfg.saddr;
2070 dst_cache = &rdst->dst_cache;
2071 md->gbp = skb->mark;
2072 ttl = vxlan->cfg.ttl;
2073 if (!ttl && vxlan_addr_multicast(dst))
2074 ttl = 1;
2075
2076 tos = vxlan->cfg.tos;
2077 if (tos == 1)
2078 tos = ip_tunnel_get_dsfield(old_iph, skb);
2079
2080 if (dst->sa.sa_family == AF_INET)
2081 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2082 else
2083 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2084 label = vxlan->cfg.label;
2085 } else {
2086 if (!info) {
2087 WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2088 dev->name);
2089 goto drop;
2090 }
2091 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2092 if (remote_ip.sa.sa_family == AF_INET) {
2093 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2094 local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2095 } else {
2096 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2097 local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2098 }
2099 dst = &remote_ip;
2100 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2101 vni = tunnel_id_to_key32(info->key.tun_id);
2102 src = &local_ip;
2103 dst_cache = &info->dst_cache;
2104 if (info->options_len)
2105 md = ip_tunnel_info_opts(info);
2106 ttl = info->key.ttl;
2107 tos = info->key.tos;
2108 label = info->key.label;
2109 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2110 }
2111 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2112 vxlan->cfg.port_max, true);
2113
2114 if (dst->sa.sa_family == AF_INET) {
2115 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2116 struct rtable *rt;
2117 __be16 df = 0;
2118
2119 rt = vxlan_get_route(vxlan, dev, sock4, skb,
2120 rdst ? rdst->remote_ifindex : 0, tos,
2121 dst->sin.sin_addr.s_addr,
2122 &src->sin.sin_addr.s_addr,
2123 dst_port, src_port,
2124 dst_cache, info);
2125 if (IS_ERR(rt)) {
2126 err = PTR_ERR(rt);
2127 goto tx_error;
2128 }
2129
2130 /* Bypass encapsulation if the destination is local */
2131 if (!info) {
2132 err = encap_bypass_if_local(skb, dev, vxlan, dst,
2133 dst_port, vni, &rt->dst,
2134 rt->rt_flags);
2135 if (err)
2136 return;
2137 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2138 df = htons(IP_DF);
2139 }
2140
2141 ndst = &rt->dst;
2142 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2143 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2144 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2145 vni, md, flags, udp_sum);
2146 if (err < 0)
2147 goto tx_error;
2148
2149 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, src->sin.sin_addr.s_addr,
2150 dst->sin.sin_addr.s_addr, tos, ttl, df,
2151 src_port, dst_port, xnet, !udp_sum);
2152 #if IS_ENABLED(CONFIG_IPV6)
2153 } else {
2154 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2155
2156 ndst = vxlan6_get_route(vxlan, dev, sock6, skb,
2157 rdst ? rdst->remote_ifindex : 0, tos,
2158 label, &dst->sin6.sin6_addr,
2159 &src->sin6.sin6_addr,
2160 dst_port, src_port,
2161 dst_cache, info);
2162 if (IS_ERR(ndst)) {
2163 err = PTR_ERR(ndst);
2164 ndst = NULL;
2165 goto tx_error;
2166 }
2167
2168 if (!info) {
2169 u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2170
2171 err = encap_bypass_if_local(skb, dev, vxlan, dst,
2172 dst_port, vni, ndst,
2173 rt6i_flags);
2174 if (err)
2175 return;
2176 }
2177
2178 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2179 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2180 skb_scrub_packet(skb, xnet);
2181 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2182 vni, md, flags, udp_sum);
2183 if (err < 0)
2184 goto tx_error;
2185
2186 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2187 &src->sin6.sin6_addr,
2188 &dst->sin6.sin6_addr, tos, ttl,
2189 label, src_port, dst_port, !udp_sum);
2190 #endif
2191 }
2192 return;
2193
2194 drop:
2195 dev->stats.tx_dropped++;
2196 dev_kfree_skb(skb);
2197 return;
2198
2199 tx_error:
2200 if (err == -ELOOP)
2201 dev->stats.collisions++;
2202 else if (err == -ENETUNREACH)
2203 dev->stats.tx_carrier_errors++;
2204 dst_release(ndst);
2205 dev->stats.tx_errors++;
2206 kfree_skb(skb);
2207 }
2208
2209 /* Transmit local packets over Vxlan
2210 *
2211 * Outer IP header inherits ECN and DF from inner header.
2212 * Outer UDP destination is the VXLAN assigned port.
2213 * source port is based on hash of flow
2214 */
2215 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2216 {
2217 struct vxlan_dev *vxlan = netdev_priv(dev);
2218 const struct ip_tunnel_info *info;
2219 struct ethhdr *eth;
2220 bool did_rsc = false;
2221 struct vxlan_rdst *rdst, *fdst = NULL;
2222 struct vxlan_fdb *f;
2223 __be32 vni = 0;
2224
2225 info = skb_tunnel_info(skb);
2226
2227 skb_reset_mac_header(skb);
2228
2229 if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2230 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2231 info->mode & IP_TUNNEL_INFO_TX) {
2232 vni = tunnel_id_to_key32(info->key.tun_id);
2233 } else {
2234 if (info && info->mode & IP_TUNNEL_INFO_TX)
2235 vxlan_xmit_one(skb, dev, vni, NULL, false);
2236 else
2237 kfree_skb(skb);
2238 return NETDEV_TX_OK;
2239 }
2240 }
2241
2242 if (vxlan->flags & VXLAN_F_PROXY) {
2243 eth = eth_hdr(skb);
2244 if (ntohs(eth->h_proto) == ETH_P_ARP)
2245 return arp_reduce(dev, skb, vni);
2246 #if IS_ENABLED(CONFIG_IPV6)
2247 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2248 pskb_may_pull(skb, sizeof(struct ipv6hdr)
2249 + sizeof(struct nd_msg)) &&
2250 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2251 struct nd_msg *msg;
2252
2253 msg = (struct nd_msg *)skb_transport_header(skb);
2254 if (msg->icmph.icmp6_code == 0 &&
2255 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2256 return neigh_reduce(dev, skb, vni);
2257 }
2258 #endif
2259 }
2260
2261 eth = eth_hdr(skb);
2262 f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2263 did_rsc = false;
2264
2265 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2266 (ntohs(eth->h_proto) == ETH_P_IP ||
2267 ntohs(eth->h_proto) == ETH_P_IPV6)) {
2268 did_rsc = route_shortcircuit(dev, skb);
2269 if (did_rsc)
2270 f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2271 }
2272
2273 if (f == NULL) {
2274 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2275 if (f == NULL) {
2276 if ((vxlan->flags & VXLAN_F_L2MISS) &&
2277 !is_multicast_ether_addr(eth->h_dest))
2278 vxlan_fdb_miss(vxlan, eth->h_dest);
2279
2280 dev->stats.tx_dropped++;
2281 kfree_skb(skb);
2282 return NETDEV_TX_OK;
2283 }
2284 }
2285
2286 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2287 struct sk_buff *skb1;
2288
2289 if (!fdst) {
2290 fdst = rdst;
2291 continue;
2292 }
2293 skb1 = skb_clone(skb, GFP_ATOMIC);
2294 if (skb1)
2295 vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2296 }
2297
2298 if (fdst)
2299 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2300 else
2301 kfree_skb(skb);
2302 return NETDEV_TX_OK;
2303 }
2304
2305 /* Walk the forwarding table and purge stale entries */
2306 static void vxlan_cleanup(unsigned long arg)
2307 {
2308 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2309 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2310 unsigned int h;
2311
2312 if (!netif_running(vxlan->dev))
2313 return;
2314
2315 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2316 struct hlist_node *p, *n;
2317
2318 spin_lock_bh(&vxlan->hash_lock);
2319 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2320 struct vxlan_fdb *f
2321 = container_of(p, struct vxlan_fdb, hlist);
2322 unsigned long timeout;
2323
2324 if (f->state & (NUD_PERMANENT | NUD_NOARP))
2325 continue;
2326
2327 timeout = f->used + vxlan->cfg.age_interval * HZ;
2328 if (time_before_eq(timeout, jiffies)) {
2329 netdev_dbg(vxlan->dev,
2330 "garbage collect %pM\n",
2331 f->eth_addr);
2332 f->state = NUD_STALE;
2333 vxlan_fdb_destroy(vxlan, f);
2334 } else if (time_before(timeout, next_timer))
2335 next_timer = timeout;
2336 }
2337 spin_unlock_bh(&vxlan->hash_lock);
2338 }
2339
2340 mod_timer(&vxlan->age_timer, next_timer);
2341 }
2342
2343 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2344 {
2345 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2346 __be32 vni = vxlan->default_dst.remote_vni;
2347
2348 spin_lock(&vn->sock_lock);
2349 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2350 spin_unlock(&vn->sock_lock);
2351 }
2352
2353 /* Setup stats when device is created */
2354 static int vxlan_init(struct net_device *dev)
2355 {
2356 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2357 if (!dev->tstats)
2358 return -ENOMEM;
2359
2360 return 0;
2361 }
2362
2363 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2364 {
2365 struct vxlan_fdb *f;
2366
2367 spin_lock_bh(&vxlan->hash_lock);
2368 f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2369 if (f)
2370 vxlan_fdb_destroy(vxlan, f);
2371 spin_unlock_bh(&vxlan->hash_lock);
2372 }
2373
2374 static void vxlan_uninit(struct net_device *dev)
2375 {
2376 struct vxlan_dev *vxlan = netdev_priv(dev);
2377
2378 vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2379
2380 free_percpu(dev->tstats);
2381 }
2382
2383 /* Start ageing timer and join group when device is brought up */
2384 static int vxlan_open(struct net_device *dev)
2385 {
2386 struct vxlan_dev *vxlan = netdev_priv(dev);
2387 int ret;
2388
2389 ret = vxlan_sock_add(vxlan);
2390 if (ret < 0)
2391 return ret;
2392
2393 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2394 ret = vxlan_igmp_join(vxlan);
2395 if (ret == -EADDRINUSE)
2396 ret = 0;
2397 if (ret) {
2398 vxlan_sock_release(vxlan);
2399 return ret;
2400 }
2401 }
2402
2403 if (vxlan->cfg.age_interval)
2404 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2405
2406 return ret;
2407 }
2408
2409 /* Purge the forwarding table */
2410 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2411 {
2412 unsigned int h;
2413
2414 spin_lock_bh(&vxlan->hash_lock);
2415 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2416 struct hlist_node *p, *n;
2417 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2418 struct vxlan_fdb *f
2419 = container_of(p, struct vxlan_fdb, hlist);
2420 if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2421 continue;
2422 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2423 if (!is_zero_ether_addr(f->eth_addr))
2424 vxlan_fdb_destroy(vxlan, f);
2425 }
2426 }
2427 spin_unlock_bh(&vxlan->hash_lock);
2428 }
2429
2430 /* Cleanup timer and forwarding table on shutdown */
2431 static int vxlan_stop(struct net_device *dev)
2432 {
2433 struct vxlan_dev *vxlan = netdev_priv(dev);
2434 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2435 int ret = 0;
2436
2437 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2438 !vxlan_group_used(vn, vxlan))
2439 ret = vxlan_igmp_leave(vxlan);
2440
2441 del_timer_sync(&vxlan->age_timer);
2442
2443 vxlan_flush(vxlan, false);
2444 vxlan_sock_release(vxlan);
2445
2446 return ret;
2447 }
2448
2449 /* Stub, nothing needs to be done. */
2450 static void vxlan_set_multicast_list(struct net_device *dev)
2451 {
2452 }
2453
2454 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2455 {
2456 struct vxlan_dev *vxlan = netdev_priv(dev);
2457 struct vxlan_rdst *dst = &vxlan->default_dst;
2458 struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2459 dst->remote_ifindex);
2460 bool use_ipv6 = false;
2461
2462 if (dst->remote_ip.sa.sa_family == AF_INET6)
2463 use_ipv6 = true;
2464
2465 /* This check is different than dev->max_mtu, because it looks at
2466 * the lowerdev->mtu, rather than the static dev->max_mtu
2467 */
2468 if (lowerdev) {
2469 int max_mtu = lowerdev->mtu -
2470 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2471 if (new_mtu > max_mtu)
2472 return -EINVAL;
2473 }
2474
2475 dev->mtu = new_mtu;
2476 return 0;
2477 }
2478
2479 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2480 {
2481 struct vxlan_dev *vxlan = netdev_priv(dev);
2482 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2483 __be16 sport, dport;
2484
2485 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2486 vxlan->cfg.port_max, true);
2487 dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2488
2489 if (ip_tunnel_info_af(info) == AF_INET) {
2490 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2491 struct rtable *rt;
2492
2493 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2494 info->key.u.ipv4.dst,
2495 &info->key.u.ipv4.src, dport, sport, NULL, info);
2496 if (IS_ERR(rt))
2497 return PTR_ERR(rt);
2498 ip_rt_put(rt);
2499 } else {
2500 #if IS_ENABLED(CONFIG_IPV6)
2501 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2502 struct dst_entry *ndst;
2503
2504 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2505 info->key.label, &info->key.u.ipv6.dst,
2506 &info->key.u.ipv6.src, dport, sport, NULL, info);
2507 if (IS_ERR(ndst))
2508 return PTR_ERR(ndst);
2509 dst_release(ndst);
2510 #else /* !CONFIG_IPV6 */
2511 return -EPFNOSUPPORT;
2512 #endif
2513 }
2514 info->key.tp_src = sport;
2515 info->key.tp_dst = dport;
2516 return 0;
2517 }
2518
2519 static const struct net_device_ops vxlan_netdev_ether_ops = {
2520 .ndo_init = vxlan_init,
2521 .ndo_uninit = vxlan_uninit,
2522 .ndo_open = vxlan_open,
2523 .ndo_stop = vxlan_stop,
2524 .ndo_start_xmit = vxlan_xmit,
2525 .ndo_get_stats64 = ip_tunnel_get_stats64,
2526 .ndo_set_rx_mode = vxlan_set_multicast_list,
2527 .ndo_change_mtu = vxlan_change_mtu,
2528 .ndo_validate_addr = eth_validate_addr,
2529 .ndo_set_mac_address = eth_mac_addr,
2530 .ndo_fdb_add = vxlan_fdb_add,
2531 .ndo_fdb_del = vxlan_fdb_delete,
2532 .ndo_fdb_dump = vxlan_fdb_dump,
2533 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2534 };
2535
2536 static const struct net_device_ops vxlan_netdev_raw_ops = {
2537 .ndo_init = vxlan_init,
2538 .ndo_uninit = vxlan_uninit,
2539 .ndo_open = vxlan_open,
2540 .ndo_stop = vxlan_stop,
2541 .ndo_start_xmit = vxlan_xmit,
2542 .ndo_get_stats64 = ip_tunnel_get_stats64,
2543 .ndo_change_mtu = vxlan_change_mtu,
2544 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2545 };
2546
2547 /* Info for udev, that this is a virtual tunnel endpoint */
2548 static struct device_type vxlan_type = {
2549 .name = "vxlan",
2550 };
2551
2552 /* Calls the ndo_udp_tunnel_add of the caller in order to
2553 * supply the listening VXLAN udp ports. Callers are expected
2554 * to implement the ndo_udp_tunnel_add.
2555 */
2556 static void vxlan_push_rx_ports(struct net_device *dev)
2557 {
2558 struct vxlan_sock *vs;
2559 struct net *net = dev_net(dev);
2560 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2561 unsigned int i;
2562
2563 spin_lock(&vn->sock_lock);
2564 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2565 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2566 udp_tunnel_push_rx_port(dev, vs->sock,
2567 (vs->flags & VXLAN_F_GPE) ?
2568 UDP_TUNNEL_TYPE_VXLAN_GPE :
2569 UDP_TUNNEL_TYPE_VXLAN);
2570 }
2571 spin_unlock(&vn->sock_lock);
2572 }
2573
2574 /* Initialize the device structure. */
2575 static void vxlan_setup(struct net_device *dev)
2576 {
2577 struct vxlan_dev *vxlan = netdev_priv(dev);
2578 unsigned int h;
2579
2580 eth_hw_addr_random(dev);
2581 ether_setup(dev);
2582
2583 dev->destructor = free_netdev;
2584 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2585
2586 dev->features |= NETIF_F_LLTX;
2587 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2588 dev->features |= NETIF_F_RXCSUM;
2589 dev->features |= NETIF_F_GSO_SOFTWARE;
2590
2591 dev->vlan_features = dev->features;
2592 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2593 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2594 netif_keep_dst(dev);
2595 dev->priv_flags |= IFF_NO_QUEUE;
2596
2597 INIT_LIST_HEAD(&vxlan->next);
2598 spin_lock_init(&vxlan->hash_lock);
2599
2600 init_timer_deferrable(&vxlan->age_timer);
2601 vxlan->age_timer.function = vxlan_cleanup;
2602 vxlan->age_timer.data = (unsigned long) vxlan;
2603
2604 vxlan->cfg.dst_port = htons(vxlan_port);
2605
2606 vxlan->dev = dev;
2607
2608 gro_cells_init(&vxlan->gro_cells, dev);
2609
2610 for (h = 0; h < FDB_HASH_SIZE; ++h)
2611 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2612 }
2613
2614 static void vxlan_ether_setup(struct net_device *dev)
2615 {
2616 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2617 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2618 dev->netdev_ops = &vxlan_netdev_ether_ops;
2619 }
2620
2621 static void vxlan_raw_setup(struct net_device *dev)
2622 {
2623 dev->header_ops = NULL;
2624 dev->type = ARPHRD_NONE;
2625 dev->hard_header_len = 0;
2626 dev->addr_len = 0;
2627 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2628 dev->netdev_ops = &vxlan_netdev_raw_ops;
2629 }
2630
2631 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2632 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2633 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2634 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2635 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2636 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2637 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2638 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2639 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2640 [IFLA_VXLAN_LABEL] = { .type = NLA_U32 },
2641 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2642 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2643 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2644 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2645 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2646 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2647 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2648 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2649 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 },
2650 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2651 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 },
2652 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
2653 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
2654 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2655 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2656 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, },
2657 [IFLA_VXLAN_GPE] = { .type = NLA_FLAG, },
2658 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG },
2659 };
2660
2661 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2662 {
2663 if (tb[IFLA_ADDRESS]) {
2664 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2665 pr_debug("invalid link address (not ethernet)\n");
2666 return -EINVAL;
2667 }
2668
2669 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2670 pr_debug("invalid all zero ethernet address\n");
2671 return -EADDRNOTAVAIL;
2672 }
2673 }
2674
2675 if (!data)
2676 return -EINVAL;
2677
2678 if (data[IFLA_VXLAN_ID]) {
2679 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2680 if (id >= VXLAN_VID_MASK)
2681 return -ERANGE;
2682 }
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
2688 if (ntohs(p->high) < ntohs(p->low)) {
2689 pr_debug("port range %u .. %u not valid\n",
2690 ntohs(p->low), ntohs(p->high));
2691 return -EINVAL;
2692 }
2693 }
2694
2695 return 0;
2696 }
2697
2698 static void vxlan_get_drvinfo(struct net_device *netdev,
2699 struct ethtool_drvinfo *drvinfo)
2700 {
2701 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2702 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2703 }
2704
2705 static const struct ethtool_ops vxlan_ethtool_ops = {
2706 .get_drvinfo = vxlan_get_drvinfo,
2707 .get_link = ethtool_op_get_link,
2708 };
2709
2710 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2711 __be16 port, u32 flags)
2712 {
2713 struct socket *sock;
2714 struct udp_port_cfg udp_conf;
2715 int err;
2716
2717 memset(&udp_conf, 0, sizeof(udp_conf));
2718
2719 if (ipv6) {
2720 udp_conf.family = AF_INET6;
2721 udp_conf.use_udp6_rx_checksums =
2722 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2723 udp_conf.ipv6_v6only = 1;
2724 } else {
2725 udp_conf.family = AF_INET;
2726 }
2727
2728 udp_conf.local_udp_port = port;
2729
2730 /* Open UDP socket */
2731 err = udp_sock_create(net, &udp_conf, &sock);
2732 if (err < 0)
2733 return ERR_PTR(err);
2734
2735 return sock;
2736 }
2737
2738 /* Create new listen socket if needed */
2739 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2740 __be16 port, u32 flags)
2741 {
2742 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2743 struct vxlan_sock *vs;
2744 struct socket *sock;
2745 unsigned int h;
2746 struct udp_tunnel_sock_cfg tunnel_cfg;
2747
2748 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2749 if (!vs)
2750 return ERR_PTR(-ENOMEM);
2751
2752 for (h = 0; h < VNI_HASH_SIZE; ++h)
2753 INIT_HLIST_HEAD(&vs->vni_list[h]);
2754
2755 sock = vxlan_create_sock(net, ipv6, port, flags);
2756 if (IS_ERR(sock)) {
2757 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2758 PTR_ERR(sock));
2759 kfree(vs);
2760 return ERR_CAST(sock);
2761 }
2762
2763 vs->sock = sock;
2764 atomic_set(&vs->refcnt, 1);
2765 vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2766
2767 spin_lock(&vn->sock_lock);
2768 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2769 udp_tunnel_notify_add_rx_port(sock,
2770 (vs->flags & VXLAN_F_GPE) ?
2771 UDP_TUNNEL_TYPE_VXLAN_GPE :
2772 UDP_TUNNEL_TYPE_VXLAN);
2773 spin_unlock(&vn->sock_lock);
2774
2775 /* Mark socket as an encapsulation socket. */
2776 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2777 tunnel_cfg.sk_user_data = vs;
2778 tunnel_cfg.encap_type = 1;
2779 tunnel_cfg.encap_rcv = vxlan_rcv;
2780 tunnel_cfg.encap_destroy = NULL;
2781 tunnel_cfg.gro_receive = vxlan_gro_receive;
2782 tunnel_cfg.gro_complete = vxlan_gro_complete;
2783
2784 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2785
2786 return vs;
2787 }
2788
2789 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2790 {
2791 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2792 struct vxlan_sock *vs = NULL;
2793
2794 if (!vxlan->cfg.no_share) {
2795 spin_lock(&vn->sock_lock);
2796 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2797 vxlan->cfg.dst_port, vxlan->flags);
2798 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2799 spin_unlock(&vn->sock_lock);
2800 return -EBUSY;
2801 }
2802 spin_unlock(&vn->sock_lock);
2803 }
2804 if (!vs)
2805 vs = vxlan_socket_create(vxlan->net, ipv6,
2806 vxlan->cfg.dst_port, vxlan->flags);
2807 if (IS_ERR(vs))
2808 return PTR_ERR(vs);
2809 #if IS_ENABLED(CONFIG_IPV6)
2810 if (ipv6)
2811 rcu_assign_pointer(vxlan->vn6_sock, vs);
2812 else
2813 #endif
2814 rcu_assign_pointer(vxlan->vn4_sock, vs);
2815 vxlan_vs_add_dev(vs, vxlan);
2816 return 0;
2817 }
2818
2819 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2820 {
2821 bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2822 bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2823 int ret = 0;
2824
2825 RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2826 #if IS_ENABLED(CONFIG_IPV6)
2827 RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2828 if (ipv6 || metadata)
2829 ret = __vxlan_sock_add(vxlan, true);
2830 #endif
2831 if (!ret && (!ipv6 || metadata))
2832 ret = __vxlan_sock_add(vxlan, false);
2833 if (ret < 0)
2834 vxlan_sock_release(vxlan);
2835 return ret;
2836 }
2837
2838 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2839 struct vxlan_config *conf)
2840 {
2841 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2842 struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2843 struct vxlan_rdst *dst = &vxlan->default_dst;
2844 unsigned short needed_headroom = ETH_HLEN;
2845 int err;
2846 bool use_ipv6 = false;
2847 __be16 default_port = vxlan->cfg.dst_port;
2848 struct net_device *lowerdev = NULL;
2849
2850 if (conf->flags & VXLAN_F_GPE) {
2851 /* For now, allow GPE only together with COLLECT_METADATA.
2852 * This can be relaxed later; in such case, the other side
2853 * of the PtP link will have to be provided.
2854 */
2855 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2856 !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2857 pr_info("unsupported combination of extensions\n");
2858 return -EINVAL;
2859 }
2860
2861 vxlan_raw_setup(dev);
2862 } else {
2863 vxlan_ether_setup(dev);
2864 }
2865
2866 /* MTU range: 68 - 65535 */
2867 dev->min_mtu = ETH_MIN_MTU;
2868 dev->max_mtu = ETH_MAX_MTU;
2869
2870 vxlan->net = src_net;
2871
2872 dst->remote_vni = conf->vni;
2873
2874 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2875
2876 /* Unless IPv6 is explicitly requested, assume IPv4 */
2877 if (!dst->remote_ip.sa.sa_family)
2878 dst->remote_ip.sa.sa_family = AF_INET;
2879
2880 if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2881 vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2882 if (!IS_ENABLED(CONFIG_IPV6))
2883 return -EPFNOSUPPORT;
2884 use_ipv6 = true;
2885 vxlan->flags |= VXLAN_F_IPV6;
2886 }
2887
2888 if (conf->label && !use_ipv6) {
2889 pr_info("label only supported in use with IPv6\n");
2890 return -EINVAL;
2891 }
2892
2893 if (conf->remote_ifindex) {
2894 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2895 dst->remote_ifindex = conf->remote_ifindex;
2896
2897 if (!lowerdev) {
2898 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2899 return -ENODEV;
2900 }
2901
2902 #if IS_ENABLED(CONFIG_IPV6)
2903 if (use_ipv6) {
2904 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2905 if (idev && idev->cnf.disable_ipv6) {
2906 pr_info("IPv6 is disabled via sysctl\n");
2907 return -EPERM;
2908 }
2909 }
2910 #endif
2911
2912 if (!conf->mtu)
2913 dev->mtu = lowerdev->mtu -
2914 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2915
2916 needed_headroom = lowerdev->hard_header_len;
2917 } else if (vxlan_addr_multicast(&dst->remote_ip)) {
2918 pr_info("multicast destination requires interface to be specified\n");
2919 return -EINVAL;
2920 }
2921
2922 if (conf->mtu) {
2923 int max_mtu = ETH_MAX_MTU;
2924
2925 if (lowerdev)
2926 max_mtu = lowerdev->mtu;
2927
2928 max_mtu -= (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2929
2930 if (conf->mtu < dev->min_mtu || conf->mtu > dev->max_mtu)
2931 return -EINVAL;
2932
2933 dev->mtu = conf->mtu;
2934
2935 if (conf->mtu > max_mtu)
2936 dev->mtu = max_mtu;
2937 }
2938
2939 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2940 needed_headroom += VXLAN6_HEADROOM;
2941 else
2942 needed_headroom += VXLAN_HEADROOM;
2943 dev->needed_headroom = needed_headroom;
2944
2945 memcpy(&vxlan->cfg, conf, sizeof(*conf));
2946 if (!vxlan->cfg.dst_port) {
2947 if (conf->flags & VXLAN_F_GPE)
2948 vxlan->cfg.dst_port = htons(4790); /* IANA VXLAN-GPE port */
2949 else
2950 vxlan->cfg.dst_port = default_port;
2951 }
2952 vxlan->flags |= conf->flags;
2953
2954 if (!vxlan->cfg.age_interval)
2955 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2956
2957 list_for_each_entry(tmp, &vn->vxlan_list, next) {
2958 if (tmp->cfg.vni == conf->vni &&
2959 (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2960 tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2961 tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2962 (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2963 (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2964 pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2965 return -EEXIST;
2966 }
2967 }
2968
2969 dev->ethtool_ops = &vxlan_ethtool_ops;
2970
2971 /* create an fdb entry for a valid default destination */
2972 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2973 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2974 &vxlan->default_dst.remote_ip,
2975 NUD_REACHABLE|NUD_PERMANENT,
2976 NLM_F_EXCL|NLM_F_CREATE,
2977 vxlan->cfg.dst_port,
2978 vxlan->default_dst.remote_vni,
2979 vxlan->default_dst.remote_vni,
2980 vxlan->default_dst.remote_ifindex,
2981 NTF_SELF);
2982 if (err)
2983 return err;
2984 }
2985
2986 err = register_netdevice(dev);
2987 if (err) {
2988 vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2989 return err;
2990 }
2991
2992 list_add(&vxlan->next, &vn->vxlan_list);
2993
2994 return 0;
2995 }
2996
2997 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2998 struct nlattr *tb[], struct nlattr *data[])
2999 {
3000 struct vxlan_config conf;
3001
3002 memset(&conf, 0, sizeof(conf));
3003
3004 if (data[IFLA_VXLAN_ID])
3005 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3006
3007 if (data[IFLA_VXLAN_GROUP]) {
3008 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3009 } else if (data[IFLA_VXLAN_GROUP6]) {
3010 if (!IS_ENABLED(CONFIG_IPV6))
3011 return -EPFNOSUPPORT;
3012
3013 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3014 conf.remote_ip.sa.sa_family = AF_INET6;
3015 }
3016
3017 if (data[IFLA_VXLAN_LOCAL]) {
3018 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3019 conf.saddr.sa.sa_family = AF_INET;
3020 } else if (data[IFLA_VXLAN_LOCAL6]) {
3021 if (!IS_ENABLED(CONFIG_IPV6))
3022 return -EPFNOSUPPORT;
3023
3024 /* TODO: respect scope id */
3025 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3026 conf.saddr.sa.sa_family = AF_INET6;
3027 }
3028
3029 if (data[IFLA_VXLAN_LINK])
3030 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3031
3032 if (data[IFLA_VXLAN_TOS])
3033 conf.tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
3034
3035 if (data[IFLA_VXLAN_TTL])
3036 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3037
3038 if (data[IFLA_VXLAN_LABEL])
3039 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3040 IPV6_FLOWLABEL_MASK;
3041
3042 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3043 conf.flags |= VXLAN_F_LEARN;
3044
3045 if (data[IFLA_VXLAN_AGEING])
3046 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3047
3048 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
3049 conf.flags |= VXLAN_F_PROXY;
3050
3051 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
3052 conf.flags |= VXLAN_F_RSC;
3053
3054 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3055 conf.flags |= VXLAN_F_L2MISS;
3056
3057 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3058 conf.flags |= VXLAN_F_L3MISS;
3059
3060 if (data[IFLA_VXLAN_LIMIT])
3061 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3062
3063 if (data[IFLA_VXLAN_COLLECT_METADATA] &&
3064 nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3065 conf.flags |= VXLAN_F_COLLECT_METADATA;
3066
3067 if (data[IFLA_VXLAN_PORT_RANGE]) {
3068 const struct ifla_vxlan_port_range *p
3069 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3070 conf.port_min = ntohs(p->low);
3071 conf.port_max = ntohs(p->high);
3072 }
3073
3074 if (data[IFLA_VXLAN_PORT])
3075 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3076
3077 if (data[IFLA_VXLAN_UDP_CSUM] &&
3078 !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3079 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3080
3081 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3082 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3083 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3084
3085 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3086 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3087 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3088
3089 if (data[IFLA_VXLAN_REMCSUM_TX] &&
3090 nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3091 conf.flags |= VXLAN_F_REMCSUM_TX;
3092
3093 if (data[IFLA_VXLAN_REMCSUM_RX] &&
3094 nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3095 conf.flags |= VXLAN_F_REMCSUM_RX;
3096
3097 if (data[IFLA_VXLAN_GBP])
3098 conf.flags |= VXLAN_F_GBP;
3099
3100 if (data[IFLA_VXLAN_GPE])
3101 conf.flags |= VXLAN_F_GPE;
3102
3103 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3104 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3105
3106 if (tb[IFLA_MTU])
3107 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3108
3109 return vxlan_dev_configure(src_net, dev, &conf);
3110 }
3111
3112 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3113 {
3114 struct vxlan_dev *vxlan = netdev_priv(dev);
3115 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3116
3117 vxlan_flush(vxlan, true);
3118
3119 spin_lock(&vn->sock_lock);
3120 if (!hlist_unhashed(&vxlan->hlist))
3121 hlist_del_rcu(&vxlan->hlist);
3122 spin_unlock(&vn->sock_lock);
3123
3124 gro_cells_destroy(&vxlan->gro_cells);
3125 list_del(&vxlan->next);
3126 unregister_netdevice_queue(dev, head);
3127 }
3128
3129 static size_t vxlan_get_size(const struct net_device *dev)
3130 {
3131
3132 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
3133 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3134 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3135 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3136 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
3137 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
3138 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3139 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
3140 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
3141 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
3142 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
3143 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
3144 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */
3145 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3146 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3147 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3148 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3149 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3150 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3151 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3152 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3153 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3154 0;
3155 }
3156
3157 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3158 {
3159 const struct vxlan_dev *vxlan = netdev_priv(dev);
3160 const struct vxlan_rdst *dst = &vxlan->default_dst;
3161 struct ifla_vxlan_port_range ports = {
3162 .low = htons(vxlan->cfg.port_min),
3163 .high = htons(vxlan->cfg.port_max),
3164 };
3165
3166 if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3167 goto nla_put_failure;
3168
3169 if (!vxlan_addr_any(&dst->remote_ip)) {
3170 if (dst->remote_ip.sa.sa_family == AF_INET) {
3171 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3172 dst->remote_ip.sin.sin_addr.s_addr))
3173 goto nla_put_failure;
3174 #if IS_ENABLED(CONFIG_IPV6)
3175 } else {
3176 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3177 &dst->remote_ip.sin6.sin6_addr))
3178 goto nla_put_failure;
3179 #endif
3180 }
3181 }
3182
3183 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3184 goto nla_put_failure;
3185
3186 if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3187 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3188 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3189 vxlan->cfg.saddr.sin.sin_addr.s_addr))
3190 goto nla_put_failure;
3191 #if IS_ENABLED(CONFIG_IPV6)
3192 } else {
3193 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3194 &vxlan->cfg.saddr.sin6.sin6_addr))
3195 goto nla_put_failure;
3196 #endif
3197 }
3198 }
3199
3200 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3201 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3202 nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3203 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3204 !!(vxlan->flags & VXLAN_F_LEARN)) ||
3205 nla_put_u8(skb, IFLA_VXLAN_PROXY,
3206 !!(vxlan->flags & VXLAN_F_PROXY)) ||
3207 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3208 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3209 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3210 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3211 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3212 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3213 !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3214 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3215 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3216 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3217 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3218 !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3219 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3220 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3221 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3222 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3223 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3224 !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3225 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3226 !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3227 goto nla_put_failure;
3228
3229 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3230 goto nla_put_failure;
3231
3232 if (vxlan->flags & VXLAN_F_GBP &&
3233 nla_put_flag(skb, IFLA_VXLAN_GBP))
3234 goto nla_put_failure;
3235
3236 if (vxlan->flags & VXLAN_F_GPE &&
3237 nla_put_flag(skb, IFLA_VXLAN_GPE))
3238 goto nla_put_failure;
3239
3240 if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3241 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3242 goto nla_put_failure;
3243
3244 return 0;
3245
3246 nla_put_failure:
3247 return -EMSGSIZE;
3248 }
3249
3250 static struct net *vxlan_get_link_net(const struct net_device *dev)
3251 {
3252 struct vxlan_dev *vxlan = netdev_priv(dev);
3253
3254 return vxlan->net;
3255 }
3256
3257 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3258 .kind = "vxlan",
3259 .maxtype = IFLA_VXLAN_MAX,
3260 .policy = vxlan_policy,
3261 .priv_size = sizeof(struct vxlan_dev),
3262 .setup = vxlan_setup,
3263 .validate = vxlan_validate,
3264 .newlink = vxlan_newlink,
3265 .dellink = vxlan_dellink,
3266 .get_size = vxlan_get_size,
3267 .fill_info = vxlan_fill_info,
3268 .get_link_net = vxlan_get_link_net,
3269 };
3270
3271 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3272 u8 name_assign_type,
3273 struct vxlan_config *conf)
3274 {
3275 struct nlattr *tb[IFLA_MAX + 1];
3276 struct net_device *dev;
3277 int err;
3278
3279 memset(&tb, 0, sizeof(tb));
3280
3281 dev = rtnl_create_link(net, name, name_assign_type,
3282 &vxlan_link_ops, tb);
3283 if (IS_ERR(dev))
3284 return dev;
3285
3286 err = vxlan_dev_configure(net, dev, conf);
3287 if (err < 0) {
3288 free_netdev(dev);
3289 return ERR_PTR(err);
3290 }
3291
3292 err = rtnl_configure_link(dev, NULL);
3293 if (err < 0) {
3294 LIST_HEAD(list_kill);
3295
3296 vxlan_dellink(dev, &list_kill);
3297 unregister_netdevice_many(&list_kill);
3298 return ERR_PTR(err);
3299 }
3300
3301 return dev;
3302 }
3303 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3304
3305 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3306 struct net_device *dev)
3307 {
3308 struct vxlan_dev *vxlan, *next;
3309 LIST_HEAD(list_kill);
3310
3311 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3312 struct vxlan_rdst *dst = &vxlan->default_dst;
3313
3314 /* In case we created vxlan device with carrier
3315 * and we loose the carrier due to module unload
3316 * we also need to remove vxlan device. In other
3317 * cases, it's not necessary and remote_ifindex
3318 * is 0 here, so no matches.
3319 */
3320 if (dst->remote_ifindex == dev->ifindex)
3321 vxlan_dellink(vxlan->dev, &list_kill);
3322 }
3323
3324 unregister_netdevice_many(&list_kill);
3325 }
3326
3327 static int vxlan_netdevice_event(struct notifier_block *unused,
3328 unsigned long event, void *ptr)
3329 {
3330 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3331 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3332
3333 if (event == NETDEV_UNREGISTER)
3334 vxlan_handle_lowerdev_unregister(vn, dev);
3335 else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3336 vxlan_push_rx_ports(dev);
3337
3338 return NOTIFY_DONE;
3339 }
3340
3341 static struct notifier_block vxlan_notifier_block __read_mostly = {
3342 .notifier_call = vxlan_netdevice_event,
3343 };
3344
3345 static __net_init int vxlan_init_net(struct net *net)
3346 {
3347 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3348 unsigned int h;
3349
3350 INIT_LIST_HEAD(&vn->vxlan_list);
3351 spin_lock_init(&vn->sock_lock);
3352
3353 for (h = 0; h < PORT_HASH_SIZE; ++h)
3354 INIT_HLIST_HEAD(&vn->sock_list[h]);
3355
3356 return 0;
3357 }
3358
3359 static void __net_exit vxlan_exit_net(struct net *net)
3360 {
3361 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3362 struct vxlan_dev *vxlan, *next;
3363 struct net_device *dev, *aux;
3364 LIST_HEAD(list);
3365
3366 rtnl_lock();
3367 for_each_netdev_safe(net, dev, aux)
3368 if (dev->rtnl_link_ops == &vxlan_link_ops)
3369 unregister_netdevice_queue(dev, &list);
3370
3371 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3372 /* If vxlan->dev is in the same netns, it has already been added
3373 * to the list by the previous loop.
3374 */
3375 if (!net_eq(dev_net(vxlan->dev), net)) {
3376 gro_cells_destroy(&vxlan->gro_cells);
3377 unregister_netdevice_queue(vxlan->dev, &list);
3378 }
3379 }
3380
3381 unregister_netdevice_many(&list);
3382 rtnl_unlock();
3383 }
3384
3385 static struct pernet_operations vxlan_net_ops = {
3386 .init = vxlan_init_net,
3387 .exit = vxlan_exit_net,
3388 .id = &vxlan_net_id,
3389 .size = sizeof(struct vxlan_net),
3390 };
3391
3392 static int __init vxlan_init_module(void)
3393 {
3394 int rc;
3395
3396 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3397
3398 rc = register_pernet_subsys(&vxlan_net_ops);
3399 if (rc)
3400 goto out1;
3401
3402 rc = register_netdevice_notifier(&vxlan_notifier_block);
3403 if (rc)
3404 goto out2;
3405
3406 rc = rtnl_link_register(&vxlan_link_ops);
3407 if (rc)
3408 goto out3;
3409
3410 return 0;
3411 out3:
3412 unregister_netdevice_notifier(&vxlan_notifier_block);
3413 out2:
3414 unregister_pernet_subsys(&vxlan_net_ops);
3415 out1:
3416 return rc;
3417 }
3418 late_initcall(vxlan_init_module);
3419
3420 static void __exit vxlan_cleanup_module(void)
3421 {
3422 rtnl_link_unregister(&vxlan_link_ops);
3423 unregister_netdevice_notifier(&vxlan_notifier_block);
3424 unregister_pernet_subsys(&vxlan_net_ops);
3425 /* rcu_barrier() is called by netns */
3426 }
3427 module_exit(vxlan_cleanup_module);
3428
3429 MODULE_LICENSE("GPL");
3430 MODULE_VERSION(VXLAN_VERSION);
3431 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3432 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3433 MODULE_ALIAS_RTNL_LINK("vxlan");
Ubuntu Artful kernel mirror