projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
vxlan: compute source port in network byte order
[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 VNI_HASH_BITS 10
48 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
49 #define FDB_HASH_BITS 8
50 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
51 #define FDB_AGE_DEFAULT 300 /* 5 min */
52 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
53
54 #define VXLAN_N_VID (1u << 24)
55 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
56 /* IP header + UDP + VXLAN + Ethernet header */
57 #define VXLAN_HEADROOM (20 + 8 + 8 + 14)
58
59 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
60
61 /* VXLAN protocol header */
62 struct vxlanhdr {
63 __be32 vx_flags;
64 __be32 vx_vni;
65 };
66
67 /* UDP port for VXLAN traffic.
68 * The IANA assigned port is 4789, but the Linux default is 8472
69 * for compatability with early adopters.
70 */
71 static unsigned int vxlan_port __read_mostly = 8472;
72 module_param_named(udp_port, vxlan_port, uint, 0444);
73 MODULE_PARM_DESC(udp_port, "Destination UDP port");
74
75 static bool log_ecn_error = true;
76 module_param(log_ecn_error, bool, 0644);
77 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
78
79 /* per-net private data for this module */
80 static unsigned int vxlan_net_id;
81 struct vxlan_net {
82 struct socket *sock; /* UDP encap socket */
83 struct hlist_head vni_list[VNI_HASH_SIZE];
84 };
85
86 struct vxlan_rdst {
87 struct rcu_head rcu;
88 __be32 remote_ip;
89 __be16 remote_port;
90 u32 remote_vni;
91 u32 remote_ifindex;
92 struct vxlan_rdst *remote_next;
93 };
94
95 /* Forwarding table entry */
96 struct vxlan_fdb {
97 struct hlist_node hlist; /* linked list of entries */
98 struct rcu_head rcu;
99 unsigned long updated; /* jiffies */
100 unsigned long used;
101 struct vxlan_rdst remote;
102 u16 state; /* see ndm_state */
103 u8 flags; /* see ndm_flags */
104 u8 eth_addr[ETH_ALEN];
105 };
106
107 /* Pseudo network device */
108 struct vxlan_dev {
109 struct hlist_node hlist;
110 struct net_device *dev;
111 struct vxlan_rdst default_dst; /* default destination */
112 __be32 saddr; /* source address */
113 __u16 port_min; /* source port range */
114 __u16 port_max;
115 __u8 tos; /* TOS override */
116 __u8 ttl;
117 u32 flags; /* VXLAN_F_* below */
118
119 unsigned long age_interval;
120 struct timer_list age_timer;
121 spinlock_t hash_lock;
122 unsigned int addrcnt;
123 unsigned int addrmax;
124
125 struct hlist_head fdb_head[FDB_HASH_SIZE];
126 };
127
128 #define VXLAN_F_LEARN 0x01
129 #define VXLAN_F_PROXY 0x02
130 #define VXLAN_F_RSC 0x04
131 #define VXLAN_F_L2MISS 0x08
132 #define VXLAN_F_L3MISS 0x10
133
134 /* salt for hash table */
135 static u32 vxlan_salt __read_mostly;
136
137 static inline struct hlist_head *vni_head(struct net *net, u32 id)
138 {
139 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
140
141 return &vn->vni_list[hash_32(id, VNI_HASH_BITS)];
142 }
143
144 /* Look up VNI in a per net namespace table */
145 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
146 {
147 struct vxlan_dev *vxlan;
148
149 hlist_for_each_entry_rcu(vxlan, vni_head(net, id), hlist) {
150 if (vxlan->default_dst.remote_vni == id)
151 return vxlan;
152 }
153
154 return NULL;
155 }
156
157 /* Fill in neighbour message in skbuff. */
158 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
159 const struct vxlan_fdb *fdb,
160 u32 portid, u32 seq, int type, unsigned int flags,
161 const struct vxlan_rdst *rdst)
162 {
163 unsigned long now = jiffies;
164 struct nda_cacheinfo ci;
165 struct nlmsghdr *nlh;
166 struct ndmsg *ndm;
167 bool send_ip, send_eth;
168
169 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
170 if (nlh == NULL)
171 return -EMSGSIZE;
172
173 ndm = nlmsg_data(nlh);
174 memset(ndm, 0, sizeof(*ndm));
175
176 send_eth = send_ip = true;
177
178 if (type == RTM_GETNEIGH) {
179 ndm->ndm_family = AF_INET;
180 send_ip = rdst->remote_ip != htonl(INADDR_ANY);
181 send_eth = !is_zero_ether_addr(fdb->eth_addr);
182 } else
183 ndm->ndm_family = AF_BRIDGE;
184 ndm->ndm_state = fdb->state;
185 ndm->ndm_ifindex = vxlan->dev->ifindex;
186 ndm->ndm_flags = fdb->flags;
187 ndm->ndm_type = NDA_DST;
188
189 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
190 goto nla_put_failure;
191
192 if (send_ip && nla_put_be32(skb, NDA_DST, rdst->remote_ip))
193 goto nla_put_failure;
194
195 if (rdst->remote_port && rdst->remote_port != htons(vxlan_port) &&
196 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
197 goto nla_put_failure;
198 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
199 nla_put_be32(skb, NDA_VNI, rdst->remote_vni))
200 goto nla_put_failure;
201 if (rdst->remote_ifindex &&
202 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
203 goto nla_put_failure;
204
205 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
206 ci.ndm_confirmed = 0;
207 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
208 ci.ndm_refcnt = 0;
209
210 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
211 goto nla_put_failure;
212
213 return nlmsg_end(skb, nlh);
214
215 nla_put_failure:
216 nlmsg_cancel(skb, nlh);
217 return -EMSGSIZE;
218 }
219
220 static inline size_t vxlan_nlmsg_size(void)
221 {
222 return NLMSG_ALIGN(sizeof(struct ndmsg))
223 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
224 + nla_total_size(sizeof(__be32)) /* NDA_DST */
225 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
226 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
227 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
228 + nla_total_size(sizeof(struct nda_cacheinfo));
229 }
230
231 static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
232 const struct vxlan_fdb *fdb, int type)
233 {
234 struct net *net = dev_net(vxlan->dev);
235 struct sk_buff *skb;
236 int err = -ENOBUFS;
237
238 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
239 if (skb == NULL)
240 goto errout;
241
242 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, &fdb->remote);
243 if (err < 0) {
244 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
245 WARN_ON(err == -EMSGSIZE);
246 kfree_skb(skb);
247 goto errout;
248 }
249
250 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
251 return;
252 errout:
253 if (err < 0)
254 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
255 }
256
257 static void vxlan_ip_miss(struct net_device *dev, __be32 ipa)
258 {
259 struct vxlan_dev *vxlan = netdev_priv(dev);
260 struct vxlan_fdb f;
261
262 memset(&f, 0, sizeof f);
263 f.state = NUD_STALE;
264 f.remote.remote_ip = ipa; /* goes to NDA_DST */
265 f.remote.remote_vni = VXLAN_N_VID;
266
267 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
268 }
269
270 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
271 {
272 struct vxlan_fdb f;
273
274 memset(&f, 0, sizeof f);
275 f.state = NUD_STALE;
276 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
277
278 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
279 }
280
281 /* Hash Ethernet address */
282 static u32 eth_hash(const unsigned char *addr)
283 {
284 u64 value = get_unaligned((u64 *)addr);
285
286 /* only want 6 bytes */
287 #ifdef __BIG_ENDIAN
288 value >>= 16;
289 #else
290 value <<= 16;
291 #endif
292 return hash_64(value, FDB_HASH_BITS);
293 }
294
295 /* Hash chain to use given mac address */
296 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
297 const u8 *mac)
298 {
299 return &vxlan->fdb_head[eth_hash(mac)];
300 }
301
302 /* Look up Ethernet address in forwarding table */
303 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
304 const u8 *mac)
305
306 {
307 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
308 struct vxlan_fdb *f;
309
310 hlist_for_each_entry_rcu(f, head, hlist) {
311 if (compare_ether_addr(mac, f->eth_addr) == 0)
312 return f;
313 }
314
315 return NULL;
316 }
317
318 /* Add/update destinations for multicast */
319 static int vxlan_fdb_append(struct vxlan_fdb *f,
320 __be32 ip, __be16 port, __u32 vni, __u32 ifindex)
321 {
322 struct vxlan_rdst *rd_prev, *rd;
323
324 rd_prev = NULL;
325 for (rd = &f->remote; rd; rd = rd->remote_next) {
326 if (rd->remote_ip == ip &&
327 rd->remote_port == port &&
328 rd->remote_vni == vni &&
329 rd->remote_ifindex == ifindex)
330 return 0;
331 rd_prev = rd;
332 }
333 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
334 if (rd == NULL)
335 return -ENOBUFS;
336 rd->remote_ip = ip;
337 rd->remote_port = port;
338 rd->remote_vni = vni;
339 rd->remote_ifindex = ifindex;
340 rd->remote_next = NULL;
341 rd_prev->remote_next = rd;
342 return 1;
343 }
344
345 /* Add new entry to forwarding table -- assumes lock held */
346 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
347 const u8 *mac, __be32 ip,
348 __u16 state, __u16 flags,
349 __be16 port, __u32 vni, __u32 ifindex,
350 __u8 ndm_flags)
351 {
352 struct vxlan_fdb *f;
353 int notify = 0;
354
355 f = vxlan_find_mac(vxlan, mac);
356 if (f) {
357 if (flags & NLM_F_EXCL) {
358 netdev_dbg(vxlan->dev,
359 "lost race to create %pM\n", mac);
360 return -EEXIST;
361 }
362 if (f->state != state) {
363 f->state = state;
364 f->updated = jiffies;
365 notify = 1;
366 }
367 if (f->flags != ndm_flags) {
368 f->flags = ndm_flags;
369 f->updated = jiffies;
370 notify = 1;
371 }
372 if ((flags & NLM_F_APPEND) &&
373 is_multicast_ether_addr(f->eth_addr)) {
374 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex);
375
376 if (rc < 0)
377 return rc;
378 notify |= rc;
379 }
380 } else {
381 if (!(flags & NLM_F_CREATE))
382 return -ENOENT;
383
384 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
385 return -ENOSPC;
386
387 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
388 f = kmalloc(sizeof(*f), GFP_ATOMIC);
389 if (!f)
390 return -ENOMEM;
391
392 notify = 1;
393 f->remote.remote_ip = ip;
394 f->remote.remote_port = port;
395 f->remote.remote_vni = vni;
396 f->remote.remote_ifindex = ifindex;
397 f->remote.remote_next = NULL;
398 f->state = state;
399 f->flags = ndm_flags;
400 f->updated = f->used = jiffies;
401 memcpy(f->eth_addr, mac, ETH_ALEN);
402
403 ++vxlan->addrcnt;
404 hlist_add_head_rcu(&f->hlist,
405 vxlan_fdb_head(vxlan, mac));
406 }
407
408 if (notify)
409 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
410
411 return 0;
412 }
413
414 static void vxlan_fdb_free(struct rcu_head *head)
415 {
416 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
417
418 while (f->remote.remote_next) {
419 struct vxlan_rdst *rd = f->remote.remote_next;
420
421 f->remote.remote_next = rd->remote_next;
422 kfree(rd);
423 }
424 kfree(f);
425 }
426
427 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
428 {
429 netdev_dbg(vxlan->dev,
430 "delete %pM\n", f->eth_addr);
431
432 --vxlan->addrcnt;
433 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
434
435 hlist_del_rcu(&f->hlist);
436 call_rcu(&f->rcu, vxlan_fdb_free);
437 }
438
439 /* Add static entry (via netlink) */
440 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
441 struct net_device *dev,
442 const unsigned char *addr, u16 flags)
443 {
444 struct vxlan_dev *vxlan = netdev_priv(dev);
445 struct net *net = dev_net(vxlan->dev);
446 __be32 ip;
447 __be16 port;
448 u32 vni, ifindex;
449 int err;
450
451 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
452 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
453 ndm->ndm_state);
454 return -EINVAL;
455 }
456
457 if (tb[NDA_DST] == NULL)
458 return -EINVAL;
459
460 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
461 return -EAFNOSUPPORT;
462
463 ip = nla_get_be32(tb[NDA_DST]);
464
465 if (tb[NDA_PORT]) {
466 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
467 return -EINVAL;
468 port = nla_get_be16(tb[NDA_PORT]);
469 } else
470 port = htons(vxlan_port);
471
472 if (tb[NDA_VNI]) {
473 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
474 return -EINVAL;
475 vni = nla_get_u32(tb[NDA_VNI]);
476 } else
477 vni = vxlan->default_dst.remote_vni;
478
479 if (tb[NDA_IFINDEX]) {
480 struct net_device *tdev;
481
482 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
483 return -EINVAL;
484 ifindex = nla_get_u32(tb[NDA_IFINDEX]);
485 tdev = dev_get_by_index(net, ifindex);
486 if (!tdev)
487 return -EADDRNOTAVAIL;
488 dev_put(tdev);
489 } else
490 ifindex = 0;
491
492 spin_lock_bh(&vxlan->hash_lock);
493 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags,
494 port, vni, ifindex, ndm->ndm_flags);
495 spin_unlock_bh(&vxlan->hash_lock);
496
497 return err;
498 }
499
500 /* Delete entry (via netlink) */
501 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
502 struct net_device *dev,
503 const unsigned char *addr)
504 {
505 struct vxlan_dev *vxlan = netdev_priv(dev);
506 struct vxlan_fdb *f;
507 int err = -ENOENT;
508
509 spin_lock_bh(&vxlan->hash_lock);
510 f = vxlan_find_mac(vxlan, addr);
511 if (f) {
512 vxlan_fdb_destroy(vxlan, f);
513 err = 0;
514 }
515 spin_unlock_bh(&vxlan->hash_lock);
516
517 return err;
518 }
519
520 /* Dump forwarding table */
521 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
522 struct net_device *dev, int idx)
523 {
524 struct vxlan_dev *vxlan = netdev_priv(dev);
525 unsigned int h;
526
527 for (h = 0; h < FDB_HASH_SIZE; ++h) {
528 struct vxlan_fdb *f;
529 int err;
530
531 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
532 struct vxlan_rdst *rd;
533 for (rd = &f->remote; rd; rd = rd->remote_next) {
534 if (idx < cb->args[0])
535 goto skip;
536
537 err = vxlan_fdb_info(skb, vxlan, f,
538 NETLINK_CB(cb->skb).portid,
539 cb->nlh->nlmsg_seq,
540 RTM_NEWNEIGH,
541 NLM_F_MULTI, rd);
542 if (err < 0)
543 break;
544 skip:
545 ++idx;
546 }
547 }
548 }
549
550 return idx;
551 }
552
553 /* Watch incoming packets to learn mapping between Ethernet address
554 * and Tunnel endpoint.
555 */
556 static void vxlan_snoop(struct net_device *dev,
557 __be32 src_ip, const u8 *src_mac)
558 {
559 struct vxlan_dev *vxlan = netdev_priv(dev);
560 struct vxlan_fdb *f;
561 int err;
562
563 f = vxlan_find_mac(vxlan, src_mac);
564 if (likely(f)) {
565 f->used = jiffies;
566 if (likely(f->remote.remote_ip == src_ip))
567 return;
568
569 if (net_ratelimit())
570 netdev_info(dev,
571 "%pM migrated from %pI4 to %pI4\n",
572 src_mac, &f->remote.remote_ip, &src_ip);
573
574 f->remote.remote_ip = src_ip;
575 f->updated = jiffies;
576 } else {
577 /* learned new entry */
578 spin_lock(&vxlan->hash_lock);
579 err = vxlan_fdb_create(vxlan, src_mac, src_ip,
580 NUD_REACHABLE,
581 NLM_F_EXCL|NLM_F_CREATE,
582 vxlan_port,
583 vxlan->default_dst.remote_vni,
584 0, NTF_SELF);
585 spin_unlock(&vxlan->hash_lock);
586 }
587 }
588
589
590 /* See if multicast group is already in use by other ID */
591 static bool vxlan_group_used(struct vxlan_net *vn,
592 const struct vxlan_dev *this)
593 {
594 const struct vxlan_dev *vxlan;
595 unsigned h;
596
597 for (h = 0; h < VNI_HASH_SIZE; ++h)
598 hlist_for_each_entry(vxlan, &vn->vni_list[h], hlist) {
599 if (vxlan == this)
600 continue;
601
602 if (!netif_running(vxlan->dev))
603 continue;
604
605 if (vxlan->default_dst.remote_ip == this->default_dst.remote_ip)
606 return true;
607 }
608
609 return false;
610 }
611
612 /* kernel equivalent to IP_ADD_MEMBERSHIP */
613 static int vxlan_join_group(struct net_device *dev)
614 {
615 struct vxlan_dev *vxlan = netdev_priv(dev);
616 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
617 struct sock *sk = vn->sock->sk;
618 struct ip_mreqn mreq = {
619 .imr_multiaddr.s_addr = vxlan->default_dst.remote_ip,
620 .imr_ifindex = vxlan->default_dst.remote_ifindex,
621 };
622 int err;
623
624 /* Already a member of group */
625 if (vxlan_group_used(vn, vxlan))
626 return 0;
627
628 /* Need to drop RTNL to call multicast join */
629 rtnl_unlock();
630 lock_sock(sk);
631 err = ip_mc_join_group(sk, &mreq);
632 release_sock(sk);
633 rtnl_lock();
634
635 return err;
636 }
637
638
639 /* kernel equivalent to IP_DROP_MEMBERSHIP */
640 static int vxlan_leave_group(struct net_device *dev)
641 {
642 struct vxlan_dev *vxlan = netdev_priv(dev);
643 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
644 int err = 0;
645 struct sock *sk = vn->sock->sk;
646 struct ip_mreqn mreq = {
647 .imr_multiaddr.s_addr = vxlan->default_dst.remote_ip,
648 .imr_ifindex = vxlan->default_dst.remote_ifindex,
649 };
650
651 /* Only leave group when last vxlan is done. */
652 if (vxlan_group_used(vn, vxlan))
653 return 0;
654
655 /* Need to drop RTNL to call multicast leave */
656 rtnl_unlock();
657 lock_sock(sk);
658 err = ip_mc_leave_group(sk, &mreq);
659 release_sock(sk);
660 rtnl_lock();
661
662 return err;
663 }
664
665 /* Callback from net/ipv4/udp.c to receive packets */
666 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
667 {
668 struct iphdr *oip;
669 struct vxlanhdr *vxh;
670 struct vxlan_dev *vxlan;
671 struct pcpu_tstats *stats;
672 __u32 vni;
673 int err;
674
675 /* pop off outer UDP header */
676 __skb_pull(skb, sizeof(struct udphdr));
677
678 /* Need Vxlan and inner Ethernet header to be present */
679 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
680 goto error;
681
682 /* Drop packets with reserved bits set */
683 vxh = (struct vxlanhdr *) skb->data;
684 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
685 (vxh->vx_vni & htonl(0xff))) {
686 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
687 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
688 goto error;
689 }
690
691 __skb_pull(skb, sizeof(struct vxlanhdr));
692
693 /* Is this VNI defined? */
694 vni = ntohl(vxh->vx_vni) >> 8;
695 vxlan = vxlan_find_vni(sock_net(sk), vni);
696 if (!vxlan) {
697 netdev_dbg(skb->dev, "unknown vni %d\n", vni);
698 goto drop;
699 }
700
701 if (!pskb_may_pull(skb, ETH_HLEN)) {
702 vxlan->dev->stats.rx_length_errors++;
703 vxlan->dev->stats.rx_errors++;
704 goto drop;
705 }
706
707 skb_reset_mac_header(skb);
708
709 /* Re-examine inner Ethernet packet */
710 oip = ip_hdr(skb);
711 skb->protocol = eth_type_trans(skb, vxlan->dev);
712
713 /* Ignore packet loops (and multicast echo) */
714 if (compare_ether_addr(eth_hdr(skb)->h_source,
715 vxlan->dev->dev_addr) == 0)
716 goto drop;
717
718 if (vxlan->flags & VXLAN_F_LEARN)
719 vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source);
720
721 __skb_tunnel_rx(skb, vxlan->dev);
722 skb_reset_network_header(skb);
723
724 /* If the NIC driver gave us an encapsulated packet with
725 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
726 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
727 * for us. Otherwise force the upper layers to verify it.
728 */
729 if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
730 !(vxlan->dev->features & NETIF_F_RXCSUM))
731 skb->ip_summed = CHECKSUM_NONE;
732
733 skb->encapsulation = 0;
734
735 err = IP_ECN_decapsulate(oip, skb);
736 if (unlikely(err)) {
737 if (log_ecn_error)
738 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
739 &oip->saddr, oip->tos);
740 if (err > 1) {
741 ++vxlan->dev->stats.rx_frame_errors;
742 ++vxlan->dev->stats.rx_errors;
743 goto drop;
744 }
745 }
746
747 stats = this_cpu_ptr(vxlan->dev->tstats);
748 u64_stats_update_begin(&stats->syncp);
749 stats->rx_packets++;
750 stats->rx_bytes += skb->len;
751 u64_stats_update_end(&stats->syncp);
752
753 netif_rx(skb);
754
755 return 0;
756 error:
757 /* Put UDP header back */
758 __skb_push(skb, sizeof(struct udphdr));
759
760 return 1;
761 drop:
762 /* Consume bad packet */
763 kfree_skb(skb);
764 return 0;
765 }
766
767 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
768 {
769 struct vxlan_dev *vxlan = netdev_priv(dev);
770 struct arphdr *parp;
771 u8 *arpptr, *sha;
772 __be32 sip, tip;
773 struct neighbour *n;
774
775 if (dev->flags & IFF_NOARP)
776 goto out;
777
778 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
779 dev->stats.tx_dropped++;
780 goto out;
781 }
782 parp = arp_hdr(skb);
783
784 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
785 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
786 parp->ar_pro != htons(ETH_P_IP) ||
787 parp->ar_op != htons(ARPOP_REQUEST) ||
788 parp->ar_hln != dev->addr_len ||
789 parp->ar_pln != 4)
790 goto out;
791 arpptr = (u8 *)parp + sizeof(struct arphdr);
792 sha = arpptr;
793 arpptr += dev->addr_len; /* sha */
794 memcpy(&sip, arpptr, sizeof(sip));
795 arpptr += sizeof(sip);
796 arpptr += dev->addr_len; /* tha */
797 memcpy(&tip, arpptr, sizeof(tip));
798
799 if (ipv4_is_loopback(tip) ||
800 ipv4_is_multicast(tip))
801 goto out;
802
803 n = neigh_lookup(&arp_tbl, &tip, dev);
804
805 if (n) {
806 struct vxlan_fdb *f;
807 struct sk_buff *reply;
808
809 if (!(n->nud_state & NUD_CONNECTED)) {
810 neigh_release(n);
811 goto out;
812 }
813
814 f = vxlan_find_mac(vxlan, n->ha);
815 if (f && f->remote.remote_ip == htonl(INADDR_ANY)) {
816 /* bridge-local neighbor */
817 neigh_release(n);
818 goto out;
819 }
820
821 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
822 n->ha, sha);
823
824 neigh_release(n);
825
826 skb_reset_mac_header(reply);
827 __skb_pull(reply, skb_network_offset(reply));
828 reply->ip_summed = CHECKSUM_UNNECESSARY;
829 reply->pkt_type = PACKET_HOST;
830
831 if (netif_rx_ni(reply) == NET_RX_DROP)
832 dev->stats.rx_dropped++;
833 } else if (vxlan->flags & VXLAN_F_L3MISS)
834 vxlan_ip_miss(dev, tip);
835 out:
836 consume_skb(skb);
837 return NETDEV_TX_OK;
838 }
839
840 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
841 {
842 struct vxlan_dev *vxlan = netdev_priv(dev);
843 struct neighbour *n;
844 struct iphdr *pip;
845
846 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
847 return false;
848
849 n = NULL;
850 switch (ntohs(eth_hdr(skb)->h_proto)) {
851 case ETH_P_IP:
852 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
853 return false;
854 pip = ip_hdr(skb);
855 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
856 break;
857 default:
858 return false;
859 }
860
861 if (n) {
862 bool diff;
863
864 diff = compare_ether_addr(eth_hdr(skb)->h_dest, n->ha) != 0;
865 if (diff) {
866 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
867 dev->addr_len);
868 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
869 }
870 neigh_release(n);
871 return diff;
872 } else if (vxlan->flags & VXLAN_F_L3MISS)
873 vxlan_ip_miss(dev, pip->daddr);
874 return false;
875 }
876
877 static void vxlan_sock_free(struct sk_buff *skb)
878 {
879 sock_put(skb->sk);
880 }
881
882 /* On transmit, associate with the tunnel socket */
883 static void vxlan_set_owner(struct net_device *dev, struct sk_buff *skb)
884 {
885 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
886 struct sock *sk = vn->sock->sk;
887
888 skb_orphan(skb);
889 sock_hold(sk);
890 skb->sk = sk;
891 skb->destructor = vxlan_sock_free;
892 }
893
894 /* Compute source port for outgoing packet
895 * first choice to use L4 flow hash since it will spread
896 * better and maybe available from hardware
897 * secondary choice is to use jhash on the Ethernet header
898 */
899 static __be16 vxlan_src_port(const struct vxlan_dev *vxlan, struct sk_buff *skb)
900 {
901 unsigned int range = (vxlan->port_max - vxlan->port_min) + 1;
902 u32 hash;
903
904 hash = skb_get_rxhash(skb);
905 if (!hash)
906 hash = jhash(skb->data, 2 * ETH_ALEN,
907 (__force u32) skb->protocol);
908
909 return htons((((u64) hash * range) >> 32) + vxlan->port_min);
910 }
911
912 static int handle_offloads(struct sk_buff *skb)
913 {
914 if (skb_is_gso(skb)) {
915 int err = skb_unclone(skb, GFP_ATOMIC);
916 if (unlikely(err))
917 return err;
918
919 skb_shinfo(skb)->gso_type |= (SKB_GSO_UDP_TUNNEL | SKB_GSO_UDP);
920 } else if (skb->ip_summed != CHECKSUM_PARTIAL)
921 skb->ip_summed = CHECKSUM_NONE;
922
923 return 0;
924 }
925
926 /* Bypass encapsulation if the destination is local */
927 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
928 struct vxlan_dev *dst_vxlan)
929 {
930 struct pcpu_tstats *tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
931 struct pcpu_tstats *rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
932
933 skb->pkt_type = PACKET_HOST;
934 skb->encapsulation = 0;
935 skb->dev = dst_vxlan->dev;
936 __skb_pull(skb, skb_network_offset(skb));
937
938 if (dst_vxlan->flags & VXLAN_F_LEARN)
939 vxlan_snoop(skb->dev, htonl(INADDR_LOOPBACK),
940 eth_hdr(skb)->h_source);
941
942 u64_stats_update_begin(&tx_stats->syncp);
943 tx_stats->tx_packets++;
944 tx_stats->tx_bytes += skb->len;
945 u64_stats_update_end(&tx_stats->syncp);
946
947 if (netif_rx(skb) == NET_RX_SUCCESS) {
948 u64_stats_update_begin(&rx_stats->syncp);
949 rx_stats->rx_packets++;
950 rx_stats->rx_bytes += skb->len;
951 u64_stats_update_end(&rx_stats->syncp);
952 } else {
953 skb->dev->stats.rx_dropped++;
954 }
955 }
956
957 static netdev_tx_t vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
958 struct vxlan_rdst *rdst, bool did_rsc)
959 {
960 struct vxlan_dev *vxlan = netdev_priv(dev);
961 struct rtable *rt;
962 const struct iphdr *old_iph;
963 struct iphdr *iph;
964 struct vxlanhdr *vxh;
965 struct udphdr *uh;
966 struct flowi4 fl4;
967 __be32 dst;
968 __be16 src_port, dst_port;
969 u32 vni;
970 __be16 df = 0;
971 __u8 tos, ttl;
972
973 dst_port = rdst->remote_port ? rdst->remote_port : htons(vxlan_port);
974 vni = rdst->remote_vni;
975 dst = rdst->remote_ip;
976
977 if (!dst) {
978 if (did_rsc) {
979 /* short-circuited back to local bridge */
980 vxlan_encap_bypass(skb, vxlan, vxlan);
981 return NETDEV_TX_OK;
982 }
983 goto drop;
984 }
985
986 if (!skb->encapsulation) {
987 skb_reset_inner_headers(skb);
988 skb->encapsulation = 1;
989 }
990
991 /* Need space for new headers (invalidates iph ptr) */
992 if (skb_cow_head(skb, VXLAN_HEADROOM))
993 goto drop;
994
995 old_iph = ip_hdr(skb);
996
997 ttl = vxlan->ttl;
998 if (!ttl && IN_MULTICAST(ntohl(dst)))
999 ttl = 1;
1000
1001 tos = vxlan->tos;
1002 if (tos == 1)
1003 tos = ip_tunnel_get_dsfield(old_iph, skb);
1004
1005 src_port = vxlan_src_port(vxlan, skb);
1006
1007 memset(&fl4, 0, sizeof(fl4));
1008 fl4.flowi4_oif = rdst->remote_ifindex;
1009 fl4.flowi4_tos = RT_TOS(tos);
1010 fl4.daddr = dst;
1011 fl4.saddr = vxlan->saddr;
1012
1013 rt = ip_route_output_key(dev_net(dev), &fl4);
1014 if (IS_ERR(rt)) {
1015 netdev_dbg(dev, "no route to %pI4\n", &dst);
1016 dev->stats.tx_carrier_errors++;
1017 goto tx_error;
1018 }
1019
1020 if (rt->dst.dev == dev) {
1021 netdev_dbg(dev, "circular route to %pI4\n", &dst);
1022 ip_rt_put(rt);
1023 dev->stats.collisions++;
1024 goto tx_error;
1025 }
1026
1027 /* Bypass encapsulation if the destination is local */
1028 if (rt->rt_flags & RTCF_LOCAL &&
1029 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1030 struct vxlan_dev *dst_vxlan;
1031
1032 ip_rt_put(rt);
1033 dst_vxlan = vxlan_find_vni(dev_net(dev), vni);
1034 if (!dst_vxlan)
1035 goto tx_error;
1036 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1037 return NETDEV_TX_OK;
1038 }
1039
1040 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1041 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1042 IPSKB_REROUTED);
1043 skb_dst_drop(skb);
1044 skb_dst_set(skb, &rt->dst);
1045
1046 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1047 vxh->vx_flags = htonl(VXLAN_FLAGS);
1048 vxh->vx_vni = htonl(vni << 8);
1049
1050 __skb_push(skb, sizeof(*uh));
1051 skb_reset_transport_header(skb);
1052 uh = udp_hdr(skb);
1053
1054 uh->dest = dst_port;
1055 uh->source = src_port;
1056
1057 uh->len = htons(skb->len);
1058 uh->check = 0;
1059
1060 __skb_push(skb, sizeof(*iph));
1061 skb_reset_network_header(skb);
1062 iph = ip_hdr(skb);
1063 iph->version = 4;
1064 iph->ihl = sizeof(struct iphdr) >> 2;
1065 iph->frag_off = df;
1066 iph->protocol = IPPROTO_UDP;
1067 iph->tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1068 iph->daddr = dst;
1069 iph->saddr = fl4.saddr;
1070 iph->ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1071 tunnel_ip_select_ident(skb, old_iph, &rt->dst);
1072
1073 nf_reset(skb);
1074
1075 vxlan_set_owner(dev, skb);
1076
1077 if (handle_offloads(skb))
1078 goto drop;
1079
1080 iptunnel_xmit(skb, dev);
1081 return NETDEV_TX_OK;
1082
1083 drop:
1084 dev->stats.tx_dropped++;
1085 goto tx_free;
1086
1087 tx_error:
1088 dev->stats.tx_errors++;
1089 tx_free:
1090 dev_kfree_skb(skb);
1091 return NETDEV_TX_OK;
1092 }
1093
1094 /* Transmit local packets over Vxlan
1095 *
1096 * Outer IP header inherits ECN and DF from inner header.
1097 * Outer UDP destination is the VXLAN assigned port.
1098 * source port is based on hash of flow
1099 */
1100 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1101 {
1102 struct vxlan_dev *vxlan = netdev_priv(dev);
1103 struct ethhdr *eth;
1104 bool did_rsc = false;
1105 struct vxlan_rdst *rdst0, *rdst;
1106 struct vxlan_fdb *f;
1107 int rc1, rc;
1108
1109 skb_reset_mac_header(skb);
1110 eth = eth_hdr(skb);
1111
1112 if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP)
1113 return arp_reduce(dev, skb);
1114
1115 f = vxlan_find_mac(vxlan, eth->h_dest);
1116 did_rsc = false;
1117
1118 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1119 ntohs(eth->h_proto) == ETH_P_IP) {
1120 did_rsc = route_shortcircuit(dev, skb);
1121 if (did_rsc)
1122 f = vxlan_find_mac(vxlan, eth->h_dest);
1123 }
1124
1125 if (f == NULL) {
1126 rdst0 = &vxlan->default_dst;
1127
1128 if (rdst0->remote_ip == htonl(INADDR_ANY) &&
1129 (vxlan->flags & VXLAN_F_L2MISS) &&
1130 !is_multicast_ether_addr(eth->h_dest))
1131 vxlan_fdb_miss(vxlan, eth->h_dest);
1132 } else
1133 rdst0 = &f->remote;
1134
1135 rc = NETDEV_TX_OK;
1136
1137 /* if there are multiple destinations, send copies */
1138 for (rdst = rdst0->remote_next; rdst; rdst = rdst->remote_next) {
1139 struct sk_buff *skb1;
1140
1141 skb1 = skb_clone(skb, GFP_ATOMIC);
1142 rc1 = vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1143 if (rc == NETDEV_TX_OK)
1144 rc = rc1;
1145 }
1146
1147 rc1 = vxlan_xmit_one(skb, dev, rdst0, did_rsc);
1148 if (rc == NETDEV_TX_OK)
1149 rc = rc1;
1150 return rc;
1151 }
1152
1153 /* Walk the forwarding table and purge stale entries */
1154 static void vxlan_cleanup(unsigned long arg)
1155 {
1156 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1157 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1158 unsigned int h;
1159
1160 if (!netif_running(vxlan->dev))
1161 return;
1162
1163 spin_lock_bh(&vxlan->hash_lock);
1164 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1165 struct hlist_node *p, *n;
1166 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1167 struct vxlan_fdb *f
1168 = container_of(p, struct vxlan_fdb, hlist);
1169 unsigned long timeout;
1170
1171 if (f->state & NUD_PERMANENT)
1172 continue;
1173
1174 timeout = f->used + vxlan->age_interval * HZ;
1175 if (time_before_eq(timeout, jiffies)) {
1176 netdev_dbg(vxlan->dev,
1177 "garbage collect %pM\n",
1178 f->eth_addr);
1179 f->state = NUD_STALE;
1180 vxlan_fdb_destroy(vxlan, f);
1181 } else if (time_before(timeout, next_timer))
1182 next_timer = timeout;
1183 }
1184 }
1185 spin_unlock_bh(&vxlan->hash_lock);
1186
1187 mod_timer(&vxlan->age_timer, next_timer);
1188 }
1189
1190 /* Setup stats when device is created */
1191 static int vxlan_init(struct net_device *dev)
1192 {
1193 dev->tstats = alloc_percpu(struct pcpu_tstats);
1194 if (!dev->tstats)
1195 return -ENOMEM;
1196
1197 return 0;
1198 }
1199
1200 /* Start ageing timer and join group when device is brought up */
1201 static int vxlan_open(struct net_device *dev)
1202 {
1203 struct vxlan_dev *vxlan = netdev_priv(dev);
1204 int err;
1205
1206 if (IN_MULTICAST(ntohl(vxlan->default_dst.remote_ip))) {
1207 err = vxlan_join_group(dev);
1208 if (err)
1209 return err;
1210 }
1211
1212 if (vxlan->age_interval)
1213 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
1214
1215 return 0;
1216 }
1217
1218 /* Purge the forwarding table */
1219 static void vxlan_flush(struct vxlan_dev *vxlan)
1220 {
1221 unsigned h;
1222
1223 spin_lock_bh(&vxlan->hash_lock);
1224 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1225 struct hlist_node *p, *n;
1226 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1227 struct vxlan_fdb *f
1228 = container_of(p, struct vxlan_fdb, hlist);
1229 vxlan_fdb_destroy(vxlan, f);
1230 }
1231 }
1232 spin_unlock_bh(&vxlan->hash_lock);
1233 }
1234
1235 /* Cleanup timer and forwarding table on shutdown */
1236 static int vxlan_stop(struct net_device *dev)
1237 {
1238 struct vxlan_dev *vxlan = netdev_priv(dev);
1239
1240 if (IN_MULTICAST(ntohl(vxlan->default_dst.remote_ip)))
1241 vxlan_leave_group(dev);
1242
1243 del_timer_sync(&vxlan->age_timer);
1244
1245 vxlan_flush(vxlan);
1246
1247 return 0;
1248 }
1249
1250 /* Stub, nothing needs to be done. */
1251 static void vxlan_set_multicast_list(struct net_device *dev)
1252 {
1253 }
1254
1255 static const struct net_device_ops vxlan_netdev_ops = {
1256 .ndo_init = vxlan_init,
1257 .ndo_open = vxlan_open,
1258 .ndo_stop = vxlan_stop,
1259 .ndo_start_xmit = vxlan_xmit,
1260 .ndo_get_stats64 = ip_tunnel_get_stats64,
1261 .ndo_set_rx_mode = vxlan_set_multicast_list,
1262 .ndo_change_mtu = eth_change_mtu,
1263 .ndo_validate_addr = eth_validate_addr,
1264 .ndo_set_mac_address = eth_mac_addr,
1265 .ndo_fdb_add = vxlan_fdb_add,
1266 .ndo_fdb_del = vxlan_fdb_delete,
1267 .ndo_fdb_dump = vxlan_fdb_dump,
1268 };
1269
1270 /* Info for udev, that this is a virtual tunnel endpoint */
1271 static struct device_type vxlan_type = {
1272 .name = "vxlan",
1273 };
1274
1275 static void vxlan_free(struct net_device *dev)
1276 {
1277 free_percpu(dev->tstats);
1278 free_netdev(dev);
1279 }
1280
1281 /* Initialize the device structure. */
1282 static void vxlan_setup(struct net_device *dev)
1283 {
1284 struct vxlan_dev *vxlan = netdev_priv(dev);
1285 unsigned h;
1286 int low, high;
1287
1288 eth_hw_addr_random(dev);
1289 ether_setup(dev);
1290 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
1291
1292 dev->netdev_ops = &vxlan_netdev_ops;
1293 dev->destructor = vxlan_free;
1294 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
1295
1296 dev->tx_queue_len = 0;
1297 dev->features |= NETIF_F_LLTX;
1298 dev->features |= NETIF_F_NETNS_LOCAL;
1299 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1300 dev->features |= NETIF_F_RXCSUM;
1301 dev->features |= NETIF_F_GSO_SOFTWARE;
1302
1303 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1304 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1305 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1306 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1307
1308 spin_lock_init(&vxlan->hash_lock);
1309
1310 init_timer_deferrable(&vxlan->age_timer);
1311 vxlan->age_timer.function = vxlan_cleanup;
1312 vxlan->age_timer.data = (unsigned long) vxlan;
1313
1314 inet_get_local_port_range(&low, &high);
1315 vxlan->port_min = low;
1316 vxlan->port_max = high;
1317
1318 vxlan->dev = dev;
1319
1320 for (h = 0; h < FDB_HASH_SIZE; ++h)
1321 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
1322 }
1323
1324 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
1325 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
1326 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1327 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
1328 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1329 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
1330 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
1331 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
1332 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
1333 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
1334 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
1335 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
1336 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
1337 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
1338 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
1339 };
1340
1341 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
1342 {
1343 if (tb[IFLA_ADDRESS]) {
1344 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1345 pr_debug("invalid link address (not ethernet)\n");
1346 return -EINVAL;
1347 }
1348
1349 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1350 pr_debug("invalid all zero ethernet address\n");
1351 return -EADDRNOTAVAIL;
1352 }
1353 }
1354
1355 if (!data)
1356 return -EINVAL;
1357
1358 if (data[IFLA_VXLAN_ID]) {
1359 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
1360 if (id >= VXLAN_VID_MASK)
1361 return -ERANGE;
1362 }
1363
1364 if (data[IFLA_VXLAN_PORT_RANGE]) {
1365 const struct ifla_vxlan_port_range *p
1366 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1367
1368 if (ntohs(p->high) < ntohs(p->low)) {
1369 pr_debug("port range %u .. %u not valid\n",
1370 ntohs(p->low), ntohs(p->high));
1371 return -EINVAL;
1372 }
1373 }
1374
1375 return 0;
1376 }
1377
1378 static void vxlan_get_drvinfo(struct net_device *netdev,
1379 struct ethtool_drvinfo *drvinfo)
1380 {
1381 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
1382 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
1383 }
1384
1385 static const struct ethtool_ops vxlan_ethtool_ops = {
1386 .get_drvinfo = vxlan_get_drvinfo,
1387 .get_link = ethtool_op_get_link,
1388 };
1389
1390 static int vxlan_newlink(struct net *net, struct net_device *dev,
1391 struct nlattr *tb[], struct nlattr *data[])
1392 {
1393 struct vxlan_dev *vxlan = netdev_priv(dev);
1394 struct vxlan_rdst *dst = &vxlan->default_dst;
1395 __u32 vni;
1396 int err;
1397
1398 if (!data[IFLA_VXLAN_ID])
1399 return -EINVAL;
1400
1401 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1402 if (vxlan_find_vni(net, vni)) {
1403 pr_info("duplicate VNI %u\n", vni);
1404 return -EEXIST;
1405 }
1406 dst->remote_vni = vni;
1407
1408 if (data[IFLA_VXLAN_GROUP])
1409 dst->remote_ip = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1410
1411 if (data[IFLA_VXLAN_LOCAL])
1412 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1413
1414 if (data[IFLA_VXLAN_LINK] &&
1415 (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
1416 struct net_device *lowerdev
1417 = __dev_get_by_index(net, dst->remote_ifindex);
1418
1419 if (!lowerdev) {
1420 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
1421 return -ENODEV;
1422 }
1423
1424 if (!tb[IFLA_MTU])
1425 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
1426
1427 /* update header length based on lower device */
1428 dev->hard_header_len = lowerdev->hard_header_len +
1429 VXLAN_HEADROOM;
1430 }
1431
1432 if (data[IFLA_VXLAN_TOS])
1433 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1434
1435 if (data[IFLA_VXLAN_TTL])
1436 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
1437
1438 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
1439 vxlan->flags |= VXLAN_F_LEARN;
1440
1441 if (data[IFLA_VXLAN_AGEING])
1442 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1443 else
1444 vxlan->age_interval = FDB_AGE_DEFAULT;
1445
1446 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
1447 vxlan->flags |= VXLAN_F_PROXY;
1448
1449 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
1450 vxlan->flags |= VXLAN_F_RSC;
1451
1452 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
1453 vxlan->flags |= VXLAN_F_L2MISS;
1454
1455 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
1456 vxlan->flags |= VXLAN_F_L3MISS;
1457
1458 if (data[IFLA_VXLAN_LIMIT])
1459 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1460
1461 if (data[IFLA_VXLAN_PORT_RANGE]) {
1462 const struct ifla_vxlan_port_range *p
1463 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1464 vxlan->port_min = ntohs(p->low);
1465 vxlan->port_max = ntohs(p->high);
1466 }
1467
1468 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
1469
1470 err = register_netdevice(dev);
1471 if (!err)
1472 hlist_add_head_rcu(&vxlan->hlist, vni_head(net, dst->remote_vni));
1473
1474 return err;
1475 }
1476
1477 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1478 {
1479 struct vxlan_dev *vxlan = netdev_priv(dev);
1480
1481 hlist_del_rcu(&vxlan->hlist);
1482
1483 unregister_netdevice_queue(dev, head);
1484 }
1485
1486 static size_t vxlan_get_size(const struct net_device *dev)
1487 {
1488
1489 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1490 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1491 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1492 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1493 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1494 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1495 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
1496 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
1497 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
1498 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
1499 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
1500 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1501 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
1502 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
1503 0;
1504 }
1505
1506 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1507 {
1508 const struct vxlan_dev *vxlan = netdev_priv(dev);
1509 const struct vxlan_rdst *dst = &vxlan->default_dst;
1510 struct ifla_vxlan_port_range ports = {
1511 .low = htons(vxlan->port_min),
1512 .high = htons(vxlan->port_max),
1513 };
1514
1515 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
1516 goto nla_put_failure;
1517
1518 if (dst->remote_ip && nla_put_be32(skb, IFLA_VXLAN_GROUP, dst->remote_ip))
1519 goto nla_put_failure;
1520
1521 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
1522 goto nla_put_failure;
1523
1524 if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
1525 goto nla_put_failure;
1526
1527 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1528 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
1529 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
1530 !!(vxlan->flags & VXLAN_F_LEARN)) ||
1531 nla_put_u8(skb, IFLA_VXLAN_PROXY,
1532 !!(vxlan->flags & VXLAN_F_PROXY)) ||
1533 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
1534 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
1535 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
1536 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
1537 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
1538 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1539 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax))
1540 goto nla_put_failure;
1541
1542 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
1543 goto nla_put_failure;
1544
1545 return 0;
1546
1547 nla_put_failure:
1548 return -EMSGSIZE;
1549 }
1550
1551 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1552 .kind = "vxlan",
1553 .maxtype = IFLA_VXLAN_MAX,
1554 .policy = vxlan_policy,
1555 .priv_size = sizeof(struct vxlan_dev),
1556 .setup = vxlan_setup,
1557 .validate = vxlan_validate,
1558 .newlink = vxlan_newlink,
1559 .dellink = vxlan_dellink,
1560 .get_size = vxlan_get_size,
1561 .fill_info = vxlan_fill_info,
1562 };
1563
1564 static __net_init int vxlan_init_net(struct net *net)
1565 {
1566 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1567 struct sock *sk;
1568 struct sockaddr_in vxlan_addr = {
1569 .sin_family = AF_INET,
1570 .sin_addr.s_addr = htonl(INADDR_ANY),
1571 };
1572 int rc;
1573 unsigned h;
1574
1575 /* Create UDP socket for encapsulation receive. */
1576 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vn->sock);
1577 if (rc < 0) {
1578 pr_debug("UDP socket create failed\n");
1579 return rc;
1580 }
1581 /* Put in proper namespace */
1582 sk = vn->sock->sk;
1583 sk_change_net(sk, net);
1584
1585 vxlan_addr.sin_port = htons(vxlan_port);
1586
1587 rc = kernel_bind(vn->sock, (struct sockaddr *) &vxlan_addr,
1588 sizeof(vxlan_addr));
1589 if (rc < 0) {
1590 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1591 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
1592 sk_release_kernel(sk);
1593 vn->sock = NULL;
1594 return rc;
1595 }
1596
1597 /* Disable multicast loopback */
1598 inet_sk(sk)->mc_loop = 0;
1599
1600 /* Mark socket as an encapsulation socket. */
1601 udp_sk(sk)->encap_type = 1;
1602 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1603 udp_encap_enable();
1604
1605 for (h = 0; h < VNI_HASH_SIZE; ++h)
1606 INIT_HLIST_HEAD(&vn->vni_list[h]);
1607
1608 return 0;
1609 }
1610
1611 static __net_exit void vxlan_exit_net(struct net *net)
1612 {
1613 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1614 struct vxlan_dev *vxlan;
1615 unsigned h;
1616
1617 rtnl_lock();
1618 for (h = 0; h < VNI_HASH_SIZE; ++h)
1619 hlist_for_each_entry(vxlan, &vn->vni_list[h], hlist)
1620 dev_close(vxlan->dev);
1621 rtnl_unlock();
1622
1623 if (vn->sock) {
1624 sk_release_kernel(vn->sock->sk);
1625 vn->sock = NULL;
1626 }
1627 }
1628
1629 static struct pernet_operations vxlan_net_ops = {
1630 .init = vxlan_init_net,
1631 .exit = vxlan_exit_net,
1632 .id = &vxlan_net_id,
1633 .size = sizeof(struct vxlan_net),
1634 };
1635
1636 static int __init vxlan_init_module(void)
1637 {
1638 int rc;
1639
1640 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1641
1642 rc = register_pernet_device(&vxlan_net_ops);
1643 if (rc)
1644 goto out1;
1645
1646 rc = rtnl_link_register(&vxlan_link_ops);
1647 if (rc)
1648 goto out2;
1649
1650 return 0;
1651
1652 out2:
1653 unregister_pernet_device(&vxlan_net_ops);
1654 out1:
1655 return rc;
1656 }
1657 module_init(vxlan_init_module);
1658
1659 static void __exit vxlan_cleanup_module(void)
1660 {
1661 rtnl_link_unregister(&vxlan_link_ops);
1662 unregister_pernet_device(&vxlan_net_ops);
1663 rcu_barrier();
1664 }
1665 module_exit(vxlan_cleanup_module);
1666
1667 MODULE_LICENSE("GPL");
1668 MODULE_VERSION(VXLAN_VERSION);
1669 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
1670 MODULE_ALIAS_RTNL_LINK("vxlan");
Linux kernel - Deliverables
This page took 0.070669 seconds and 5 git commands to generate.