projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
[deliverable/linux.git] / drivers / net / vxlan.c
1 /*
2 * VXLAN: Virtual eXtensible Local Area Network
3 *
4 * Copyright (c) 2012-2013 Vyatta Inc.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 *
10 * TODO
11 * - IPv6 (not in RFC)
12 */
13
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15
16 #include <linux/kernel.h>
17 #include <linux/types.h>
18 #include <linux/module.h>
19 #include <linux/errno.h>
20 #include <linux/slab.h>
21 #include <linux/skbuff.h>
22 #include <linux/rculist.h>
23 #include <linux/netdevice.h>
24 #include <linux/in.h>
25 #include <linux/ip.h>
26 #include <linux/udp.h>
27 #include <linux/igmp.h>
28 #include <linux/etherdevice.h>
29 #include <linux/if_ether.h>
30 #include <linux/hash.h>
31 #include <linux/ethtool.h>
32 #include <net/arp.h>
33 #include <net/ndisc.h>
34 #include <net/ip.h>
35 #include <net/ip_tunnels.h>
36 #include <net/icmp.h>
37 #include <net/udp.h>
38 #include <net/rtnetlink.h>
39 #include <net/route.h>
40 #include <net/dsfield.h>
41 #include <net/inet_ecn.h>
42 #include <net/net_namespace.h>
43 #include <net/netns/generic.h>
44
45 #define VXLAN_VERSION "0.1"
46
47 #define PORT_HASH_BITS 8
48 #define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
49 #define VNI_HASH_BITS 10
50 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
51 #define FDB_HASH_BITS 8
52 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
53 #define FDB_AGE_DEFAULT 300 /* 5 min */
54 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
55
56 #define VXLAN_N_VID (1u << 24)
57 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
58 /* IP header + UDP + VXLAN + Ethernet header */
59 #define VXLAN_HEADROOM (20 + 8 + 8 + 14)
60
61 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
62
63 /* VXLAN protocol header */
64 struct vxlanhdr {
65 __be32 vx_flags;
66 __be32 vx_vni;
67 };
68
69 /* UDP port for VXLAN traffic.
70 * The IANA assigned port is 4789, but the Linux default is 8472
71 * for compatability with early adopters.
72 */
73 static unsigned int vxlan_port __read_mostly = 8472;
74 module_param_named(udp_port, vxlan_port, uint, 0444);
75 MODULE_PARM_DESC(udp_port, "Destination UDP port");
76
77 static bool log_ecn_error = true;
78 module_param(log_ecn_error, bool, 0644);
79 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
80
81 static unsigned int vxlan_net_id;
82
83 /* per UDP socket information */
84 struct vxlan_sock {
85 struct hlist_node hlist;
86 struct rcu_head rcu;
87 struct work_struct del_work;
88 unsigned int refcnt;
89 struct socket *sock;
90 struct hlist_head vni_list[VNI_HASH_SIZE];
91 };
92
93 /* per-network namespace private data for this module */
94 struct vxlan_net {
95 struct list_head vxlan_list;
96 struct hlist_head sock_list[PORT_HASH_SIZE];
97 };
98
99 struct vxlan_rdst {
100 __be32 remote_ip;
101 __be16 remote_port;
102 u32 remote_vni;
103 u32 remote_ifindex;
104 struct vxlan_rdst *remote_next;
105 };
106
107 /* Forwarding table entry */
108 struct vxlan_fdb {
109 struct hlist_node hlist; /* linked list of entries */
110 struct rcu_head rcu;
111 unsigned long updated; /* jiffies */
112 unsigned long used;
113 struct vxlan_rdst remote;
114 u16 state; /* see ndm_state */
115 u8 flags; /* see ndm_flags */
116 u8 eth_addr[ETH_ALEN];
117 };
118
119 /* Pseudo network device */
120 struct vxlan_dev {
121 struct hlist_node hlist; /* vni hash table */
122 struct list_head next; /* vxlan's per namespace list */
123 struct vxlan_sock *vn_sock; /* listening socket */
124 struct net_device *dev;
125 struct vxlan_rdst default_dst; /* default destination */
126 __be32 saddr; /* source address */
127 __be16 dst_port;
128 __u16 port_min; /* source port range */
129 __u16 port_max;
130 __u8 tos; /* TOS override */
131 __u8 ttl;
132 u32 flags; /* VXLAN_F_* below */
133
134 unsigned long age_interval;
135 struct timer_list age_timer;
136 spinlock_t hash_lock;
137 unsigned int addrcnt;
138 unsigned int addrmax;
139
140 struct hlist_head fdb_head[FDB_HASH_SIZE];
141 };
142
143 #define VXLAN_F_LEARN 0x01
144 #define VXLAN_F_PROXY 0x02
145 #define VXLAN_F_RSC 0x04
146 #define VXLAN_F_L2MISS 0x08
147 #define VXLAN_F_L3MISS 0x10
148
149 /* salt for hash table */
150 static u32 vxlan_salt __read_mostly;
151
152 /* Virtual Network hash table head */
153 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
154 {
155 return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
156 }
157
158 /* Socket hash table head */
159 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
160 {
161 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
162
163 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
164 }
165
166 /* Find VXLAN socket based on network namespace and UDP port */
167 static struct vxlan_sock *vxlan_find_port(struct net *net, __be16 port)
168 {
169 struct vxlan_sock *vs;
170
171 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
172 if (inet_sk(vs->sock->sk)->inet_sport == port)
173 return vs;
174 }
175 return NULL;
176 }
177
178 /* Look up VNI in a per net namespace table */
179 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id, __be16 port)
180 {
181 struct vxlan_sock *vs;
182 struct vxlan_dev *vxlan;
183
184 vs = vxlan_find_port(net, port);
185 if (!vs)
186 return NULL;
187
188 hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
189 if (vxlan->default_dst.remote_vni == id)
190 return vxlan;
191 }
192
193 return NULL;
194 }
195
196 /* Fill in neighbour message in skbuff. */
197 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
198 const struct vxlan_fdb *fdb,
199 u32 portid, u32 seq, int type, unsigned int flags,
200 const struct vxlan_rdst *rdst)
201 {
202 unsigned long now = jiffies;
203 struct nda_cacheinfo ci;
204 struct nlmsghdr *nlh;
205 struct ndmsg *ndm;
206 bool send_ip, send_eth;
207
208 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
209 if (nlh == NULL)
210 return -EMSGSIZE;
211
212 ndm = nlmsg_data(nlh);
213 memset(ndm, 0, sizeof(*ndm));
214
215 send_eth = send_ip = true;
216
217 if (type == RTM_GETNEIGH) {
218 ndm->ndm_family = AF_INET;
219 send_ip = rdst->remote_ip != htonl(INADDR_ANY);
220 send_eth = !is_zero_ether_addr(fdb->eth_addr);
221 } else
222 ndm->ndm_family = AF_BRIDGE;
223 ndm->ndm_state = fdb->state;
224 ndm->ndm_ifindex = vxlan->dev->ifindex;
225 ndm->ndm_flags = fdb->flags;
226 ndm->ndm_type = NDA_DST;
227
228 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
229 goto nla_put_failure;
230
231 if (send_ip && nla_put_be32(skb, NDA_DST, rdst->remote_ip))
232 goto nla_put_failure;
233
234 if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
235 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
236 goto nla_put_failure;
237 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
238 nla_put_be32(skb, NDA_VNI, rdst->remote_vni))
239 goto nla_put_failure;
240 if (rdst->remote_ifindex &&
241 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
242 goto nla_put_failure;
243
244 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
245 ci.ndm_confirmed = 0;
246 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
247 ci.ndm_refcnt = 0;
248
249 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
250 goto nla_put_failure;
251
252 return nlmsg_end(skb, nlh);
253
254 nla_put_failure:
255 nlmsg_cancel(skb, nlh);
256 return -EMSGSIZE;
257 }
258
259 static inline size_t vxlan_nlmsg_size(void)
260 {
261 return NLMSG_ALIGN(sizeof(struct ndmsg))
262 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
263 + nla_total_size(sizeof(__be32)) /* NDA_DST */
264 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
265 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
266 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
267 + nla_total_size(sizeof(struct nda_cacheinfo));
268 }
269
270 static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
271 const struct vxlan_fdb *fdb, int type)
272 {
273 struct net *net = dev_net(vxlan->dev);
274 struct sk_buff *skb;
275 int err = -ENOBUFS;
276
277 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
278 if (skb == NULL)
279 goto errout;
280
281 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, &fdb->remote);
282 if (err < 0) {
283 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
284 WARN_ON(err == -EMSGSIZE);
285 kfree_skb(skb);
286 goto errout;
287 }
288
289 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
290 return;
291 errout:
292 if (err < 0)
293 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
294 }
295
296 static void vxlan_ip_miss(struct net_device *dev, __be32 ipa)
297 {
298 struct vxlan_dev *vxlan = netdev_priv(dev);
299 struct vxlan_fdb f;
300
301 memset(&f, 0, sizeof f);
302 f.state = NUD_STALE;
303 f.remote.remote_ip = ipa; /* goes to NDA_DST */
304 f.remote.remote_vni = VXLAN_N_VID;
305
306 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
307 }
308
309 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
310 {
311 struct vxlan_fdb f;
312
313 memset(&f, 0, sizeof f);
314 f.state = NUD_STALE;
315 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
316
317 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
318 }
319
320 /* Hash Ethernet address */
321 static u32 eth_hash(const unsigned char *addr)
322 {
323 u64 value = get_unaligned((u64 *)addr);
324
325 /* only want 6 bytes */
326 #ifdef __BIG_ENDIAN
327 value >>= 16;
328 #else
329 value <<= 16;
330 #endif
331 return hash_64(value, FDB_HASH_BITS);
332 }
333
334 /* Hash chain to use given mac address */
335 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
336 const u8 *mac)
337 {
338 return &vxlan->fdb_head[eth_hash(mac)];
339 }
340
341 /* Look up Ethernet address in forwarding table */
342 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
343 const u8 *mac)
344
345 {
346 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
347 struct vxlan_fdb *f;
348
349 hlist_for_each_entry_rcu(f, head, hlist) {
350 if (compare_ether_addr(mac, f->eth_addr) == 0)
351 return f;
352 }
353
354 return NULL;
355 }
356
357 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
358 const u8 *mac)
359 {
360 struct vxlan_fdb *f;
361
362 f = __vxlan_find_mac(vxlan, mac);
363 if (f)
364 f->used = jiffies;
365
366 return f;
367 }
368
369 /* Add/update destinations for multicast */
370 static int vxlan_fdb_append(struct vxlan_fdb *f,
371 __be32 ip, __be16 port, __u32 vni, __u32 ifindex)
372 {
373 struct vxlan_rdst *rd_prev, *rd;
374
375 rd_prev = NULL;
376 for (rd = &f->remote; rd; rd = rd->remote_next) {
377 if (rd->remote_ip == ip &&
378 rd->remote_port == port &&
379 rd->remote_vni == vni &&
380 rd->remote_ifindex == ifindex)
381 return 0;
382 rd_prev = rd;
383 }
384 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
385 if (rd == NULL)
386 return -ENOBUFS;
387 rd->remote_ip = ip;
388 rd->remote_port = port;
389 rd->remote_vni = vni;
390 rd->remote_ifindex = ifindex;
391 rd->remote_next = NULL;
392 rd_prev->remote_next = rd;
393 return 1;
394 }
395
396 /* Add new entry to forwarding table -- assumes lock held */
397 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
398 const u8 *mac, __be32 ip,
399 __u16 state, __u16 flags,
400 __be16 port, __u32 vni, __u32 ifindex,
401 __u8 ndm_flags)
402 {
403 struct vxlan_fdb *f;
404 int notify = 0;
405
406 f = __vxlan_find_mac(vxlan, mac);
407 if (f) {
408 if (flags & NLM_F_EXCL) {
409 netdev_dbg(vxlan->dev,
410 "lost race to create %pM\n", mac);
411 return -EEXIST;
412 }
413 if (f->state != state) {
414 f->state = state;
415 f->updated = jiffies;
416 notify = 1;
417 }
418 if (f->flags != ndm_flags) {
419 f->flags = ndm_flags;
420 f->updated = jiffies;
421 notify = 1;
422 }
423 if ((flags & NLM_F_APPEND) &&
424 is_multicast_ether_addr(f->eth_addr)) {
425 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex);
426
427 if (rc < 0)
428 return rc;
429 notify |= rc;
430 }
431 } else {
432 if (!(flags & NLM_F_CREATE))
433 return -ENOENT;
434
435 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
436 return -ENOSPC;
437
438 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
439 f = kmalloc(sizeof(*f), GFP_ATOMIC);
440 if (!f)
441 return -ENOMEM;
442
443 notify = 1;
444 f->remote.remote_ip = ip;
445 f->remote.remote_port = port;
446 f->remote.remote_vni = vni;
447 f->remote.remote_ifindex = ifindex;
448 f->remote.remote_next = NULL;
449 f->state = state;
450 f->flags = ndm_flags;
451 f->updated = f->used = jiffies;
452 memcpy(f->eth_addr, mac, ETH_ALEN);
453
454 ++vxlan->addrcnt;
455 hlist_add_head_rcu(&f->hlist,
456 vxlan_fdb_head(vxlan, mac));
457 }
458
459 if (notify)
460 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
461
462 return 0;
463 }
464
465 static void vxlan_fdb_free(struct rcu_head *head)
466 {
467 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
468
469 while (f->remote.remote_next) {
470 struct vxlan_rdst *rd = f->remote.remote_next;
471
472 f->remote.remote_next = rd->remote_next;
473 kfree(rd);
474 }
475 kfree(f);
476 }
477
478 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
479 {
480 netdev_dbg(vxlan->dev,
481 "delete %pM\n", f->eth_addr);
482
483 --vxlan->addrcnt;
484 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
485
486 hlist_del_rcu(&f->hlist);
487 call_rcu(&f->rcu, vxlan_fdb_free);
488 }
489
490 /* Add static entry (via netlink) */
491 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
492 struct net_device *dev,
493 const unsigned char *addr, u16 flags)
494 {
495 struct vxlan_dev *vxlan = netdev_priv(dev);
496 struct net *net = dev_net(vxlan->dev);
497 __be32 ip;
498 __be16 port;
499 u32 vni, ifindex;
500 int err;
501
502 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
503 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
504 ndm->ndm_state);
505 return -EINVAL;
506 }
507
508 if (tb[NDA_DST] == NULL)
509 return -EINVAL;
510
511 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
512 return -EAFNOSUPPORT;
513
514 ip = nla_get_be32(tb[NDA_DST]);
515
516 if (tb[NDA_PORT]) {
517 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
518 return -EINVAL;
519 port = nla_get_be16(tb[NDA_PORT]);
520 } else
521 port = vxlan->dst_port;
522
523 if (tb[NDA_VNI]) {
524 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
525 return -EINVAL;
526 vni = nla_get_u32(tb[NDA_VNI]);
527 } else
528 vni = vxlan->default_dst.remote_vni;
529
530 if (tb[NDA_IFINDEX]) {
531 struct net_device *tdev;
532
533 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
534 return -EINVAL;
535 ifindex = nla_get_u32(tb[NDA_IFINDEX]);
536 tdev = dev_get_by_index(net, ifindex);
537 if (!tdev)
538 return -EADDRNOTAVAIL;
539 dev_put(tdev);
540 } else
541 ifindex = 0;
542
543 spin_lock_bh(&vxlan->hash_lock);
544 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags,
545 port, vni, ifindex, ndm->ndm_flags);
546 spin_unlock_bh(&vxlan->hash_lock);
547
548 return err;
549 }
550
551 /* Delete entry (via netlink) */
552 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
553 struct net_device *dev,
554 const unsigned char *addr)
555 {
556 struct vxlan_dev *vxlan = netdev_priv(dev);
557 struct vxlan_fdb *f;
558 int err = -ENOENT;
559
560 spin_lock_bh(&vxlan->hash_lock);
561 f = vxlan_find_mac(vxlan, addr);
562 if (f) {
563 vxlan_fdb_destroy(vxlan, f);
564 err = 0;
565 }
566 spin_unlock_bh(&vxlan->hash_lock);
567
568 return err;
569 }
570
571 /* Dump forwarding table */
572 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
573 struct net_device *dev, int idx)
574 {
575 struct vxlan_dev *vxlan = netdev_priv(dev);
576 unsigned int h;
577
578 for (h = 0; h < FDB_HASH_SIZE; ++h) {
579 struct vxlan_fdb *f;
580 int err;
581
582 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
583 struct vxlan_rdst *rd;
584 for (rd = &f->remote; rd; rd = rd->remote_next) {
585 if (idx < cb->args[0])
586 goto skip;
587
588 err = vxlan_fdb_info(skb, vxlan, f,
589 NETLINK_CB(cb->skb).portid,
590 cb->nlh->nlmsg_seq,
591 RTM_NEWNEIGH,
592 NLM_F_MULTI, rd);
593 if (err < 0)
594 break;
595 skip:
596 ++idx;
597 }
598 }
599 }
600
601 return idx;
602 }
603
604 /* Watch incoming packets to learn mapping between Ethernet address
605 * and Tunnel endpoint.
606 * Return true if packet is bogus and should be droppped.
607 */
608 static bool vxlan_snoop(struct net_device *dev,
609 __be32 src_ip, const u8 *src_mac)
610 {
611 struct vxlan_dev *vxlan = netdev_priv(dev);
612 struct vxlan_fdb *f;
613
614 f = vxlan_find_mac(vxlan, src_mac);
615 if (likely(f)) {
616 if (likely(f->remote.remote_ip == src_ip))
617 return false;
618
619 /* Don't migrate static entries, drop packets */
620 if (f->state & NUD_NOARP)
621 return true;
622
623 if (net_ratelimit())
624 netdev_info(dev,
625 "%pM migrated from %pI4 to %pI4\n",
626 src_mac, &f->remote.remote_ip, &src_ip);
627
628 f->remote.remote_ip = src_ip;
629 f->updated = jiffies;
630 } else {
631 /* learned new entry */
632 spin_lock(&vxlan->hash_lock);
633
634 /* close off race between vxlan_flush and incoming packets */
635 if (netif_running(dev))
636 vxlan_fdb_create(vxlan, src_mac, src_ip,
637 NUD_REACHABLE,
638 NLM_F_EXCL|NLM_F_CREATE,
639 vxlan->dst_port,
640 vxlan->default_dst.remote_vni,
641 0, NTF_SELF);
642 spin_unlock(&vxlan->hash_lock);
643 }
644
645 return false;
646 }
647
648
649 /* See if multicast group is already in use by other ID */
650 static bool vxlan_group_used(struct vxlan_net *vn,
651 const struct vxlan_dev *this)
652 {
653 struct vxlan_dev *vxlan;
654
655 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
656 if (vxlan == this)
657 continue;
658
659 if (!netif_running(vxlan->dev))
660 continue;
661
662 if (vxlan->default_dst.remote_ip == this->default_dst.remote_ip)
663 return true;
664 }
665
666 return false;
667 }
668
669 /* kernel equivalent to IP_ADD_MEMBERSHIP */
670 static int vxlan_join_group(struct net_device *dev)
671 {
672 struct vxlan_dev *vxlan = netdev_priv(dev);
673 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
674 struct sock *sk = vxlan->vn_sock->sock->sk;
675 struct ip_mreqn mreq = {
676 .imr_multiaddr.s_addr = vxlan->default_dst.remote_ip,
677 .imr_ifindex = vxlan->default_dst.remote_ifindex,
678 };
679 int err;
680
681 /* Already a member of group */
682 if (vxlan_group_used(vn, vxlan))
683 return 0;
684
685 /* Need to drop RTNL to call multicast join */
686 rtnl_unlock();
687 lock_sock(sk);
688 err = ip_mc_join_group(sk, &mreq);
689 release_sock(sk);
690 rtnl_lock();
691
692 return err;
693 }
694
695
696 /* kernel equivalent to IP_DROP_MEMBERSHIP */
697 static int vxlan_leave_group(struct net_device *dev)
698 {
699 struct vxlan_dev *vxlan = netdev_priv(dev);
700 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
701 int err = 0;
702 struct sock *sk = vxlan->vn_sock->sock->sk;
703 struct ip_mreqn mreq = {
704 .imr_multiaddr.s_addr = vxlan->default_dst.remote_ip,
705 .imr_ifindex = vxlan->default_dst.remote_ifindex,
706 };
707
708 /* Only leave group when last vxlan is done. */
709 if (vxlan_group_used(vn, vxlan))
710 return 0;
711
712 /* Need to drop RTNL to call multicast leave */
713 rtnl_unlock();
714 lock_sock(sk);
715 err = ip_mc_leave_group(sk, &mreq);
716 release_sock(sk);
717 rtnl_lock();
718
719 return err;
720 }
721
722 /* Callback from net/ipv4/udp.c to receive packets */
723 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
724 {
725 struct iphdr *oip;
726 struct vxlanhdr *vxh;
727 struct vxlan_dev *vxlan;
728 struct pcpu_tstats *stats;
729 __be16 port;
730 __u32 vni;
731 int err;
732
733 /* pop off outer UDP header */
734 __skb_pull(skb, sizeof(struct udphdr));
735
736 /* Need Vxlan and inner Ethernet header to be present */
737 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
738 goto error;
739
740 /* Drop packets with reserved bits set */
741 vxh = (struct vxlanhdr *) skb->data;
742 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
743 (vxh->vx_vni & htonl(0xff))) {
744 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
745 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
746 goto error;
747 }
748
749 __skb_pull(skb, sizeof(struct vxlanhdr));
750
751 /* Is this VNI defined? */
752 vni = ntohl(vxh->vx_vni) >> 8;
753 port = inet_sk(sk)->inet_sport;
754 vxlan = vxlan_find_vni(sock_net(sk), vni, port);
755 if (!vxlan) {
756 netdev_dbg(skb->dev, "unknown vni %d port %u\n",
757 vni, ntohs(port));
758 goto drop;
759 }
760
761 if (!pskb_may_pull(skb, ETH_HLEN)) {
762 vxlan->dev->stats.rx_length_errors++;
763 vxlan->dev->stats.rx_errors++;
764 goto drop;
765 }
766
767 skb_reset_mac_header(skb);
768
769 /* Re-examine inner Ethernet packet */
770 oip = ip_hdr(skb);
771 skb->protocol = eth_type_trans(skb, vxlan->dev);
772
773 /* Ignore packet loops (and multicast echo) */
774 if (compare_ether_addr(eth_hdr(skb)->h_source,
775 vxlan->dev->dev_addr) == 0)
776 goto drop;
777
778 if ((vxlan->flags & VXLAN_F_LEARN) &&
779 vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source))
780 goto drop;
781
782 __skb_tunnel_rx(skb, vxlan->dev);
783 skb_reset_network_header(skb);
784
785 /* If the NIC driver gave us an encapsulated packet with
786 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
787 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
788 * for us. Otherwise force the upper layers to verify it.
789 */
790 if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
791 !(vxlan->dev->features & NETIF_F_RXCSUM))
792 skb->ip_summed = CHECKSUM_NONE;
793
794 skb->encapsulation = 0;
795
796 err = IP_ECN_decapsulate(oip, skb);
797 if (unlikely(err)) {
798 if (log_ecn_error)
799 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
800 &oip->saddr, oip->tos);
801 if (err > 1) {
802 ++vxlan->dev->stats.rx_frame_errors;
803 ++vxlan->dev->stats.rx_errors;
804 goto drop;
805 }
806 }
807
808 stats = this_cpu_ptr(vxlan->dev->tstats);
809 u64_stats_update_begin(&stats->syncp);
810 stats->rx_packets++;
811 stats->rx_bytes += skb->len;
812 u64_stats_update_end(&stats->syncp);
813
814 netif_rx(skb);
815
816 return 0;
817 error:
818 /* Put UDP header back */
819 __skb_push(skb, sizeof(struct udphdr));
820
821 return 1;
822 drop:
823 /* Consume bad packet */
824 kfree_skb(skb);
825 return 0;
826 }
827
828 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
829 {
830 struct vxlan_dev *vxlan = netdev_priv(dev);
831 struct arphdr *parp;
832 u8 *arpptr, *sha;
833 __be32 sip, tip;
834 struct neighbour *n;
835
836 if (dev->flags & IFF_NOARP)
837 goto out;
838
839 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
840 dev->stats.tx_dropped++;
841 goto out;
842 }
843 parp = arp_hdr(skb);
844
845 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
846 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
847 parp->ar_pro != htons(ETH_P_IP) ||
848 parp->ar_op != htons(ARPOP_REQUEST) ||
849 parp->ar_hln != dev->addr_len ||
850 parp->ar_pln != 4)
851 goto out;
852 arpptr = (u8 *)parp + sizeof(struct arphdr);
853 sha = arpptr;
854 arpptr += dev->addr_len; /* sha */
855 memcpy(&sip, arpptr, sizeof(sip));
856 arpptr += sizeof(sip);
857 arpptr += dev->addr_len; /* tha */
858 memcpy(&tip, arpptr, sizeof(tip));
859
860 if (ipv4_is_loopback(tip) ||
861 ipv4_is_multicast(tip))
862 goto out;
863
864 n = neigh_lookup(&arp_tbl, &tip, dev);
865
866 if (n) {
867 struct vxlan_fdb *f;
868 struct sk_buff *reply;
869
870 if (!(n->nud_state & NUD_CONNECTED)) {
871 neigh_release(n);
872 goto out;
873 }
874
875 f = vxlan_find_mac(vxlan, n->ha);
876 if (f && f->remote.remote_ip == htonl(INADDR_ANY)) {
877 /* bridge-local neighbor */
878 neigh_release(n);
879 goto out;
880 }
881
882 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
883 n->ha, sha);
884
885 neigh_release(n);
886
887 skb_reset_mac_header(reply);
888 __skb_pull(reply, skb_network_offset(reply));
889 reply->ip_summed = CHECKSUM_UNNECESSARY;
890 reply->pkt_type = PACKET_HOST;
891
892 if (netif_rx_ni(reply) == NET_RX_DROP)
893 dev->stats.rx_dropped++;
894 } else if (vxlan->flags & VXLAN_F_L3MISS)
895 vxlan_ip_miss(dev, tip);
896 out:
897 consume_skb(skb);
898 return NETDEV_TX_OK;
899 }
900
901 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
902 {
903 struct vxlan_dev *vxlan = netdev_priv(dev);
904 struct neighbour *n;
905 struct iphdr *pip;
906
907 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
908 return false;
909
910 n = NULL;
911 switch (ntohs(eth_hdr(skb)->h_proto)) {
912 case ETH_P_IP:
913 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
914 return false;
915 pip = ip_hdr(skb);
916 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
917 break;
918 default:
919 return false;
920 }
921
922 if (n) {
923 bool diff;
924
925 diff = compare_ether_addr(eth_hdr(skb)->h_dest, n->ha) != 0;
926 if (diff) {
927 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
928 dev->addr_len);
929 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
930 }
931 neigh_release(n);
932 return diff;
933 } else if (vxlan->flags & VXLAN_F_L3MISS)
934 vxlan_ip_miss(dev, pip->daddr);
935 return false;
936 }
937
938 static void vxlan_sock_put(struct sk_buff *skb)
939 {
940 sock_put(skb->sk);
941 }
942
943 /* On transmit, associate with the tunnel socket */
944 static void vxlan_set_owner(struct net_device *dev, struct sk_buff *skb)
945 {
946 struct vxlan_dev *vxlan = netdev_priv(dev);
947 struct sock *sk = vxlan->vn_sock->sock->sk;
948
949 skb_orphan(skb);
950 sock_hold(sk);
951 skb->sk = sk;
952 skb->destructor = vxlan_sock_put;
953 }
954
955 /* Compute source port for outgoing packet
956 * first choice to use L4 flow hash since it will spread
957 * better and maybe available from hardware
958 * secondary choice is to use jhash on the Ethernet header
959 */
960 static __be16 vxlan_src_port(const struct vxlan_dev *vxlan, struct sk_buff *skb)
961 {
962 unsigned int range = (vxlan->port_max - vxlan->port_min) + 1;
963 u32 hash;
964
965 hash = skb_get_rxhash(skb);
966 if (!hash)
967 hash = jhash(skb->data, 2 * ETH_ALEN,
968 (__force u32) skb->protocol);
969
970 return htons((((u64) hash * range) >> 32) + vxlan->port_min);
971 }
972
973 static int handle_offloads(struct sk_buff *skb)
974 {
975 if (skb_is_gso(skb)) {
976 int err = skb_unclone(skb, GFP_ATOMIC);
977 if (unlikely(err))
978 return err;
979
980 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
981 } else if (skb->ip_summed != CHECKSUM_PARTIAL)
982 skb->ip_summed = CHECKSUM_NONE;
983
984 return 0;
985 }
986
987 /* Bypass encapsulation if the destination is local */
988 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
989 struct vxlan_dev *dst_vxlan)
990 {
991 struct pcpu_tstats *tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
992 struct pcpu_tstats *rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
993
994 skb->pkt_type = PACKET_HOST;
995 skb->encapsulation = 0;
996 skb->dev = dst_vxlan->dev;
997 __skb_pull(skb, skb_network_offset(skb));
998
999 if (dst_vxlan->flags & VXLAN_F_LEARN)
1000 vxlan_snoop(skb->dev, htonl(INADDR_LOOPBACK),
1001 eth_hdr(skb)->h_source);
1002
1003 u64_stats_update_begin(&tx_stats->syncp);
1004 tx_stats->tx_packets++;
1005 tx_stats->tx_bytes += skb->len;
1006 u64_stats_update_end(&tx_stats->syncp);
1007
1008 if (netif_rx(skb) == NET_RX_SUCCESS) {
1009 u64_stats_update_begin(&rx_stats->syncp);
1010 rx_stats->rx_packets++;
1011 rx_stats->rx_bytes += skb->len;
1012 u64_stats_update_end(&rx_stats->syncp);
1013 } else {
1014 skb->dev->stats.rx_dropped++;
1015 }
1016 }
1017
1018 static netdev_tx_t vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1019 struct vxlan_rdst *rdst, bool did_rsc)
1020 {
1021 struct vxlan_dev *vxlan = netdev_priv(dev);
1022 struct rtable *rt;
1023 const struct iphdr *old_iph;
1024 struct vxlanhdr *vxh;
1025 struct udphdr *uh;
1026 struct flowi4 fl4;
1027 __be32 dst;
1028 __be16 src_port, dst_port;
1029 u32 vni;
1030 __be16 df = 0;
1031 __u8 tos, ttl;
1032 int err;
1033
1034 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1035 vni = rdst->remote_vni;
1036 dst = rdst->remote_ip;
1037
1038 if (!dst) {
1039 if (did_rsc) {
1040 /* short-circuited back to local bridge */
1041 vxlan_encap_bypass(skb, vxlan, vxlan);
1042 return NETDEV_TX_OK;
1043 }
1044 goto drop;
1045 }
1046
1047 if (!skb->encapsulation) {
1048 skb_reset_inner_headers(skb);
1049 skb->encapsulation = 1;
1050 }
1051
1052 /* Need space for new headers (invalidates iph ptr) */
1053 if (skb_cow_head(skb, VXLAN_HEADROOM))
1054 goto drop;
1055
1056 old_iph = ip_hdr(skb);
1057
1058 ttl = vxlan->ttl;
1059 if (!ttl && IN_MULTICAST(ntohl(dst)))
1060 ttl = 1;
1061
1062 tos = vxlan->tos;
1063 if (tos == 1)
1064 tos = ip_tunnel_get_dsfield(old_iph, skb);
1065
1066 src_port = vxlan_src_port(vxlan, skb);
1067
1068 memset(&fl4, 0, sizeof(fl4));
1069 fl4.flowi4_oif = rdst->remote_ifindex;
1070 fl4.flowi4_tos = RT_TOS(tos);
1071 fl4.daddr = dst;
1072 fl4.saddr = vxlan->saddr;
1073
1074 rt = ip_route_output_key(dev_net(dev), &fl4);
1075 if (IS_ERR(rt)) {
1076 netdev_dbg(dev, "no route to %pI4\n", &dst);
1077 dev->stats.tx_carrier_errors++;
1078 goto tx_error;
1079 }
1080
1081 if (rt->dst.dev == dev) {
1082 netdev_dbg(dev, "circular route to %pI4\n", &dst);
1083 ip_rt_put(rt);
1084 dev->stats.collisions++;
1085 goto tx_error;
1086 }
1087
1088 /* Bypass encapsulation if the destination is local */
1089 if (rt->rt_flags & RTCF_LOCAL &&
1090 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1091 struct vxlan_dev *dst_vxlan;
1092
1093 ip_rt_put(rt);
1094 dst_vxlan = vxlan_find_vni(dev_net(dev), vni, dst_port);
1095 if (!dst_vxlan)
1096 goto tx_error;
1097 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1098 return NETDEV_TX_OK;
1099 }
1100 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1101 vxh->vx_flags = htonl(VXLAN_FLAGS);
1102 vxh->vx_vni = htonl(vni << 8);
1103
1104 __skb_push(skb, sizeof(*uh));
1105 skb_reset_transport_header(skb);
1106 uh = udp_hdr(skb);
1107
1108 uh->dest = dst_port;
1109 uh->source = src_port;
1110
1111 uh->len = htons(skb->len);
1112 uh->check = 0;
1113
1114 vxlan_set_owner(dev, skb);
1115
1116 if (handle_offloads(skb))
1117 goto drop;
1118
1119 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1120 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1121
1122 err = iptunnel_xmit(dev_net(dev), rt, skb, fl4.saddr, dst,
1123 IPPROTO_UDP, tos, ttl, df);
1124 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1125
1126 return NETDEV_TX_OK;
1127
1128 drop:
1129 dev->stats.tx_dropped++;
1130 goto tx_free;
1131
1132 tx_error:
1133 dev->stats.tx_errors++;
1134 tx_free:
1135 dev_kfree_skb(skb);
1136 return NETDEV_TX_OK;
1137 }
1138
1139 /* Transmit local packets over Vxlan
1140 *
1141 * Outer IP header inherits ECN and DF from inner header.
1142 * Outer UDP destination is the VXLAN assigned port.
1143 * source port is based on hash of flow
1144 */
1145 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1146 {
1147 struct vxlan_dev *vxlan = netdev_priv(dev);
1148 struct ethhdr *eth;
1149 bool did_rsc = false;
1150 struct vxlan_rdst *rdst0, *rdst;
1151 struct vxlan_fdb *f;
1152 int rc1, rc;
1153
1154 skb_reset_mac_header(skb);
1155 eth = eth_hdr(skb);
1156
1157 if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP)
1158 return arp_reduce(dev, skb);
1159
1160 f = vxlan_find_mac(vxlan, eth->h_dest);
1161 did_rsc = false;
1162
1163 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1164 ntohs(eth->h_proto) == ETH_P_IP) {
1165 did_rsc = route_shortcircuit(dev, skb);
1166 if (did_rsc)
1167 f = vxlan_find_mac(vxlan, eth->h_dest);
1168 }
1169
1170 if (f == NULL) {
1171 rdst0 = &vxlan->default_dst;
1172
1173 if (rdst0->remote_ip == htonl(INADDR_ANY) &&
1174 (vxlan->flags & VXLAN_F_L2MISS) &&
1175 !is_multicast_ether_addr(eth->h_dest))
1176 vxlan_fdb_miss(vxlan, eth->h_dest);
1177 } else
1178 rdst0 = &f->remote;
1179
1180 rc = NETDEV_TX_OK;
1181
1182 /* if there are multiple destinations, send copies */
1183 for (rdst = rdst0->remote_next; rdst; rdst = rdst->remote_next) {
1184 struct sk_buff *skb1;
1185
1186 skb1 = skb_clone(skb, GFP_ATOMIC);
1187 if (skb1) {
1188 rc1 = vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1189 if (rc == NETDEV_TX_OK)
1190 rc = rc1;
1191 }
1192 }
1193
1194 rc1 = vxlan_xmit_one(skb, dev, rdst0, did_rsc);
1195 if (rc == NETDEV_TX_OK)
1196 rc = rc1;
1197 return rc;
1198 }
1199
1200 /* Walk the forwarding table and purge stale entries */
1201 static void vxlan_cleanup(unsigned long arg)
1202 {
1203 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1204 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1205 unsigned int h;
1206
1207 if (!netif_running(vxlan->dev))
1208 return;
1209
1210 spin_lock_bh(&vxlan->hash_lock);
1211 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1212 struct hlist_node *p, *n;
1213 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1214 struct vxlan_fdb *f
1215 = container_of(p, struct vxlan_fdb, hlist);
1216 unsigned long timeout;
1217
1218 if (f->state & NUD_PERMANENT)
1219 continue;
1220
1221 timeout = f->used + vxlan->age_interval * HZ;
1222 if (time_before_eq(timeout, jiffies)) {
1223 netdev_dbg(vxlan->dev,
1224 "garbage collect %pM\n",
1225 f->eth_addr);
1226 f->state = NUD_STALE;
1227 vxlan_fdb_destroy(vxlan, f);
1228 } else if (time_before(timeout, next_timer))
1229 next_timer = timeout;
1230 }
1231 }
1232 spin_unlock_bh(&vxlan->hash_lock);
1233
1234 mod_timer(&vxlan->age_timer, next_timer);
1235 }
1236
1237 /* Setup stats when device is created */
1238 static int vxlan_init(struct net_device *dev)
1239 {
1240 dev->tstats = alloc_percpu(struct pcpu_tstats);
1241 if (!dev->tstats)
1242 return -ENOMEM;
1243
1244 return 0;
1245 }
1246
1247 /* Start ageing timer and join group when device is brought up */
1248 static int vxlan_open(struct net_device *dev)
1249 {
1250 struct vxlan_dev *vxlan = netdev_priv(dev);
1251 int err;
1252
1253 if (IN_MULTICAST(ntohl(vxlan->default_dst.remote_ip))) {
1254 err = vxlan_join_group(dev);
1255 if (err)
1256 return err;
1257 }
1258
1259 if (vxlan->age_interval)
1260 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
1261
1262 return 0;
1263 }
1264
1265 /* Purge the forwarding table */
1266 static void vxlan_flush(struct vxlan_dev *vxlan)
1267 {
1268 unsigned int h;
1269
1270 spin_lock_bh(&vxlan->hash_lock);
1271 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1272 struct hlist_node *p, *n;
1273 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1274 struct vxlan_fdb *f
1275 = container_of(p, struct vxlan_fdb, hlist);
1276 vxlan_fdb_destroy(vxlan, f);
1277 }
1278 }
1279 spin_unlock_bh(&vxlan->hash_lock);
1280 }
1281
1282 /* Cleanup timer and forwarding table on shutdown */
1283 static int vxlan_stop(struct net_device *dev)
1284 {
1285 struct vxlan_dev *vxlan = netdev_priv(dev);
1286
1287 if (IN_MULTICAST(ntohl(vxlan->default_dst.remote_ip)))
1288 vxlan_leave_group(dev);
1289
1290 del_timer_sync(&vxlan->age_timer);
1291
1292 vxlan_flush(vxlan);
1293
1294 return 0;
1295 }
1296
1297 /* Stub, nothing needs to be done. */
1298 static void vxlan_set_multicast_list(struct net_device *dev)
1299 {
1300 }
1301
1302 static const struct net_device_ops vxlan_netdev_ops = {
1303 .ndo_init = vxlan_init,
1304 .ndo_open = vxlan_open,
1305 .ndo_stop = vxlan_stop,
1306 .ndo_start_xmit = vxlan_xmit,
1307 .ndo_get_stats64 = ip_tunnel_get_stats64,
1308 .ndo_set_rx_mode = vxlan_set_multicast_list,
1309 .ndo_change_mtu = eth_change_mtu,
1310 .ndo_validate_addr = eth_validate_addr,
1311 .ndo_set_mac_address = eth_mac_addr,
1312 .ndo_fdb_add = vxlan_fdb_add,
1313 .ndo_fdb_del = vxlan_fdb_delete,
1314 .ndo_fdb_dump = vxlan_fdb_dump,
1315 };
1316
1317 /* Info for udev, that this is a virtual tunnel endpoint */
1318 static struct device_type vxlan_type = {
1319 .name = "vxlan",
1320 };
1321
1322 static void vxlan_free(struct net_device *dev)
1323 {
1324 free_percpu(dev->tstats);
1325 free_netdev(dev);
1326 }
1327
1328 /* Initialize the device structure. */
1329 static void vxlan_setup(struct net_device *dev)
1330 {
1331 struct vxlan_dev *vxlan = netdev_priv(dev);
1332 unsigned int h;
1333 int low, high;
1334
1335 eth_hw_addr_random(dev);
1336 ether_setup(dev);
1337 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
1338
1339 dev->netdev_ops = &vxlan_netdev_ops;
1340 dev->destructor = vxlan_free;
1341 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
1342
1343 dev->tx_queue_len = 0;
1344 dev->features |= NETIF_F_LLTX;
1345 dev->features |= NETIF_F_NETNS_LOCAL;
1346 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1347 dev->features |= NETIF_F_RXCSUM;
1348 dev->features |= NETIF_F_GSO_SOFTWARE;
1349
1350 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1351 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1352 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1353 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1354
1355 INIT_LIST_HEAD(&vxlan->next);
1356 spin_lock_init(&vxlan->hash_lock);
1357
1358 init_timer_deferrable(&vxlan->age_timer);
1359 vxlan->age_timer.function = vxlan_cleanup;
1360 vxlan->age_timer.data = (unsigned long) vxlan;
1361
1362 inet_get_local_port_range(&low, &high);
1363 vxlan->port_min = low;
1364 vxlan->port_max = high;
1365 vxlan->dst_port = htons(vxlan_port);
1366
1367 vxlan->dev = dev;
1368
1369 for (h = 0; h < FDB_HASH_SIZE; ++h)
1370 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
1371 }
1372
1373 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
1374 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
1375 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1376 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
1377 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1378 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
1379 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
1380 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
1381 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
1382 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
1383 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
1384 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
1385 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
1386 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
1387 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
1388 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
1389 };
1390
1391 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
1392 {
1393 if (tb[IFLA_ADDRESS]) {
1394 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1395 pr_debug("invalid link address (not ethernet)\n");
1396 return -EINVAL;
1397 }
1398
1399 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1400 pr_debug("invalid all zero ethernet address\n");
1401 return -EADDRNOTAVAIL;
1402 }
1403 }
1404
1405 if (!data)
1406 return -EINVAL;
1407
1408 if (data[IFLA_VXLAN_ID]) {
1409 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
1410 if (id >= VXLAN_VID_MASK)
1411 return -ERANGE;
1412 }
1413
1414 if (data[IFLA_VXLAN_PORT_RANGE]) {
1415 const struct ifla_vxlan_port_range *p
1416 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1417
1418 if (ntohs(p->high) < ntohs(p->low)) {
1419 pr_debug("port range %u .. %u not valid\n",
1420 ntohs(p->low), ntohs(p->high));
1421 return -EINVAL;
1422 }
1423 }
1424
1425 return 0;
1426 }
1427
1428 static void vxlan_get_drvinfo(struct net_device *netdev,
1429 struct ethtool_drvinfo *drvinfo)
1430 {
1431 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
1432 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
1433 }
1434
1435 static const struct ethtool_ops vxlan_ethtool_ops = {
1436 .get_drvinfo = vxlan_get_drvinfo,
1437 .get_link = ethtool_op_get_link,
1438 };
1439
1440 static void vxlan_del_work(struct work_struct *work)
1441 {
1442 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
1443
1444 sk_release_kernel(vs->sock->sk);
1445 kfree_rcu(vs, rcu);
1446 }
1447
1448 /* Create new listen socket if needed */
1449 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port)
1450 {
1451 struct vxlan_sock *vs;
1452 struct sock *sk;
1453 struct sockaddr_in vxlan_addr = {
1454 .sin_family = AF_INET,
1455 .sin_addr.s_addr = htonl(INADDR_ANY),
1456 };
1457 int rc;
1458 unsigned int h;
1459
1460 vs = kmalloc(sizeof(*vs), GFP_KERNEL);
1461 if (!vs)
1462 return ERR_PTR(-ENOMEM);
1463
1464 for (h = 0; h < VNI_HASH_SIZE; ++h)
1465 INIT_HLIST_HEAD(&vs->vni_list[h]);
1466
1467 INIT_WORK(&vs->del_work, vxlan_del_work);
1468
1469 /* Create UDP socket for encapsulation receive. */
1470 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vs->sock);
1471 if (rc < 0) {
1472 pr_debug("UDP socket create failed\n");
1473 kfree(vs);
1474 return ERR_PTR(rc);
1475 }
1476
1477 /* Put in proper namespace */
1478 sk = vs->sock->sk;
1479 sk_change_net(sk, net);
1480
1481 vxlan_addr.sin_port = port;
1482
1483 rc = kernel_bind(vs->sock, (struct sockaddr *) &vxlan_addr,
1484 sizeof(vxlan_addr));
1485 if (rc < 0) {
1486 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1487 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
1488 sk_release_kernel(sk);
1489 kfree(vs);
1490 return ERR_PTR(rc);
1491 }
1492
1493 /* Disable multicast loopback */
1494 inet_sk(sk)->mc_loop = 0;
1495
1496 /* Mark socket as an encapsulation socket. */
1497 udp_sk(sk)->encap_type = 1;
1498 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1499 udp_encap_enable();
1500
1501 vs->refcnt = 1;
1502 return vs;
1503 }
1504
1505 static int vxlan_newlink(struct net *net, struct net_device *dev,
1506 struct nlattr *tb[], struct nlattr *data[])
1507 {
1508 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1509 struct vxlan_dev *vxlan = netdev_priv(dev);
1510 struct vxlan_rdst *dst = &vxlan->default_dst;
1511 struct vxlan_sock *vs;
1512 __u32 vni;
1513 int err;
1514
1515 if (!data[IFLA_VXLAN_ID])
1516 return -EINVAL;
1517
1518 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1519 dst->remote_vni = vni;
1520
1521 if (data[IFLA_VXLAN_GROUP])
1522 dst->remote_ip = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1523
1524 if (data[IFLA_VXLAN_LOCAL])
1525 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1526
1527 if (data[IFLA_VXLAN_LINK] &&
1528 (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
1529 struct net_device *lowerdev
1530 = __dev_get_by_index(net, dst->remote_ifindex);
1531
1532 if (!lowerdev) {
1533 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
1534 return -ENODEV;
1535 }
1536
1537 if (!tb[IFLA_MTU])
1538 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
1539
1540 /* update header length based on lower device */
1541 dev->hard_header_len = lowerdev->hard_header_len +
1542 VXLAN_HEADROOM;
1543 }
1544
1545 if (data[IFLA_VXLAN_TOS])
1546 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1547
1548 if (data[IFLA_VXLAN_TTL])
1549 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
1550
1551 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
1552 vxlan->flags |= VXLAN_F_LEARN;
1553
1554 if (data[IFLA_VXLAN_AGEING])
1555 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1556 else
1557 vxlan->age_interval = FDB_AGE_DEFAULT;
1558
1559 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
1560 vxlan->flags |= VXLAN_F_PROXY;
1561
1562 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
1563 vxlan->flags |= VXLAN_F_RSC;
1564
1565 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
1566 vxlan->flags |= VXLAN_F_L2MISS;
1567
1568 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
1569 vxlan->flags |= VXLAN_F_L3MISS;
1570
1571 if (data[IFLA_VXLAN_LIMIT])
1572 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1573
1574 if (data[IFLA_VXLAN_PORT_RANGE]) {
1575 const struct ifla_vxlan_port_range *p
1576 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1577 vxlan->port_min = ntohs(p->low);
1578 vxlan->port_max = ntohs(p->high);
1579 }
1580
1581 if (data[IFLA_VXLAN_PORT])
1582 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
1583
1584 if (vxlan_find_vni(net, vni, vxlan->dst_port)) {
1585 pr_info("duplicate VNI %u\n", vni);
1586 return -EEXIST;
1587 }
1588
1589 vs = vxlan_find_port(net, vxlan->dst_port);
1590 if (vs)
1591 ++vs->refcnt;
1592 else {
1593 /* Drop lock because socket create acquires RTNL lock */
1594 rtnl_unlock();
1595 vs = vxlan_socket_create(net, vxlan->dst_port);
1596 rtnl_lock();
1597 if (IS_ERR(vs))
1598 return PTR_ERR(vs);
1599
1600 hlist_add_head_rcu(&vs->hlist, vs_head(net, vxlan->dst_port));
1601 }
1602 vxlan->vn_sock = vs;
1603
1604 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
1605
1606 err = register_netdevice(dev);
1607 if (err) {
1608 if (--vs->refcnt == 0) {
1609 rtnl_unlock();
1610 sk_release_kernel(vs->sock->sk);
1611 kfree(vs);
1612 rtnl_lock();
1613 }
1614 return err;
1615 }
1616
1617 list_add(&vxlan->next, &vn->vxlan_list);
1618 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
1619
1620 return 0;
1621 }
1622
1623 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1624 {
1625 struct vxlan_dev *vxlan = netdev_priv(dev);
1626 struct vxlan_sock *vs = vxlan->vn_sock;
1627
1628 hlist_del_rcu(&vxlan->hlist);
1629 list_del(&vxlan->next);
1630 unregister_netdevice_queue(dev, head);
1631
1632 if (--vs->refcnt == 0) {
1633 hlist_del_rcu(&vs->hlist);
1634 schedule_work(&vs->del_work);
1635 }
1636 }
1637
1638 static size_t vxlan_get_size(const struct net_device *dev)
1639 {
1640
1641 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1642 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1643 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1644 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1645 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1646 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1647 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
1648 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
1649 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
1650 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
1651 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
1652 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1653 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
1654 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
1655 nla_total_size(sizeof(__be16))+ /* IFLA_VXLAN_PORT */
1656 0;
1657 }
1658
1659 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1660 {
1661 const struct vxlan_dev *vxlan = netdev_priv(dev);
1662 const struct vxlan_rdst *dst = &vxlan->default_dst;
1663 struct ifla_vxlan_port_range ports = {
1664 .low = htons(vxlan->port_min),
1665 .high = htons(vxlan->port_max),
1666 };
1667
1668 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
1669 goto nla_put_failure;
1670
1671 if (dst->remote_ip && nla_put_be32(skb, IFLA_VXLAN_GROUP, dst->remote_ip))
1672 goto nla_put_failure;
1673
1674 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
1675 goto nla_put_failure;
1676
1677 if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
1678 goto nla_put_failure;
1679
1680 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1681 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
1682 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
1683 !!(vxlan->flags & VXLAN_F_LEARN)) ||
1684 nla_put_u8(skb, IFLA_VXLAN_PROXY,
1685 !!(vxlan->flags & VXLAN_F_PROXY)) ||
1686 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
1687 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
1688 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
1689 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
1690 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
1691 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1692 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
1693 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port))
1694 goto nla_put_failure;
1695
1696 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
1697 goto nla_put_failure;
1698
1699 return 0;
1700
1701 nla_put_failure:
1702 return -EMSGSIZE;
1703 }
1704
1705 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1706 .kind = "vxlan",
1707 .maxtype = IFLA_VXLAN_MAX,
1708 .policy = vxlan_policy,
1709 .priv_size = sizeof(struct vxlan_dev),
1710 .setup = vxlan_setup,
1711 .validate = vxlan_validate,
1712 .newlink = vxlan_newlink,
1713 .dellink = vxlan_dellink,
1714 .get_size = vxlan_get_size,
1715 .fill_info = vxlan_fill_info,
1716 };
1717
1718 static __net_init int vxlan_init_net(struct net *net)
1719 {
1720 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1721 unsigned int h;
1722
1723 INIT_LIST_HEAD(&vn->vxlan_list);
1724
1725 for (h = 0; h < PORT_HASH_SIZE; ++h)
1726 INIT_HLIST_HEAD(&vn->sock_list[h]);
1727
1728 return 0;
1729 }
1730
1731 static __net_exit void vxlan_exit_net(struct net *net)
1732 {
1733 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1734 struct vxlan_dev *vxlan;
1735
1736 rtnl_lock();
1737 list_for_each_entry(vxlan, &vn->vxlan_list, next)
1738 dev_close(vxlan->dev);
1739 rtnl_unlock();
1740 }
1741
1742 static struct pernet_operations vxlan_net_ops = {
1743 .init = vxlan_init_net,
1744 .exit = vxlan_exit_net,
1745 .id = &vxlan_net_id,
1746 .size = sizeof(struct vxlan_net),
1747 };
1748
1749 static int __init vxlan_init_module(void)
1750 {
1751 int rc;
1752
1753 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1754
1755 rc = register_pernet_device(&vxlan_net_ops);
1756 if (rc)
1757 goto out1;
1758
1759 rc = rtnl_link_register(&vxlan_link_ops);
1760 if (rc)
1761 goto out2;
1762
1763 return 0;
1764
1765 out2:
1766 unregister_pernet_device(&vxlan_net_ops);
1767 out1:
1768 return rc;
1769 }
1770 late_initcall(vxlan_init_module);
1771
1772 static void __exit vxlan_cleanup_module(void)
1773 {
1774 rtnl_link_unregister(&vxlan_link_ops);
1775 unregister_pernet_device(&vxlan_net_ops);
1776 rcu_barrier();
1777 }
1778 module_exit(vxlan_cleanup_module);
1779
1780 MODULE_LICENSE("GPL");
1781 MODULE_VERSION(VXLAN_VERSION);
1782 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
1783 MODULE_ALIAS_RTNL_LINK("vxlan");
Linux kernel - Deliverables
This page took 0.09761 seconds and 6 git commands to generate.