projects / deliverable / linux.git / blob
? search:


e532b2ab5b0f387c368d00e39a937dcecfdb83fe
[deliverable/linux.git] / drivers / net / vxlan.c
1 /*
2 * VXLAN: Virtual eXtensible Local Area Network
3 *
4 * Copyright (c) 2012 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 * - use IANA UDP port number (when defined)
12 * - IPv6 (not in RFC)
13 */
14
15 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16
17 #include <linux/kernel.h>
18 #include <linux/types.h>
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/skbuff.h>
23 #include <linux/rculist.h>
24 #include <linux/netdevice.h>
25 #include <linux/in.h>
26 #include <linux/ip.h>
27 #include <linux/udp.h>
28 #include <linux/igmp.h>
29 #include <linux/etherdevice.h>
30 #include <linux/if_ether.h>
31 #include <linux/hash.h>
32 #include <linux/ethtool.h>
33 #include <net/arp.h>
34 #include <net/ndisc.h>
35 #include <net/ip.h>
36 #include <net/ip_tunnels.h>
37 #include <net/icmp.h>
38 #include <net/udp.h>
39 #include <net/rtnetlink.h>
40 #include <net/route.h>
41 #include <net/dsfield.h>
42 #include <net/inet_ecn.h>
43 #include <net/net_namespace.h>
44 #include <net/netns/generic.h>
45
46 #define VXLAN_VERSION "0.1"
47
48 #define VNI_HASH_BITS 10
49 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
50 #define FDB_HASH_BITS 8
51 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
52 #define FDB_AGE_DEFAULT 300 /* 5 min */
53 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
54
55 #define VXLAN_N_VID (1u << 24)
56 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
57 /* IP header + UDP + VXLAN + Ethernet header */
58 #define VXLAN_HEADROOM (20 + 8 + 8 + 14)
59
60 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
61
62 /* VXLAN protocol header */
63 struct vxlanhdr {
64 __be32 vx_flags;
65 __be32 vx_vni;
66 };
67
68 /* UDP port for VXLAN traffic. */
69 static unsigned int vxlan_port __read_mostly = 8472;
70 module_param_named(udp_port, vxlan_port, uint, 0444);
71 MODULE_PARM_DESC(udp_port, "Destination UDP port");
72
73 static bool log_ecn_error = true;
74 module_param(log_ecn_error, bool, 0644);
75 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
76
77 /* per-net private data for this module */
78 static unsigned int vxlan_net_id;
79 struct vxlan_net {
80 struct socket *sock; /* UDP encap socket */
81 struct hlist_head vni_list[VNI_HASH_SIZE];
82 };
83
84 struct vxlan_rdst {
85 struct rcu_head rcu;
86 __be32 remote_ip;
87 __be16 remote_port;
88 u32 remote_vni;
89 u32 remote_ifindex;
90 struct vxlan_rdst *remote_next;
91 };
92
93 /* Forwarding table entry */
94 struct vxlan_fdb {
95 struct hlist_node hlist; /* linked list of entries */
96 struct rcu_head rcu;
97 unsigned long updated; /* jiffies */
98 unsigned long used;
99 struct vxlan_rdst remote;
100 u16 state; /* see ndm_state */
101 u8 eth_addr[ETH_ALEN];
102 };
103
104 /* Pseudo network device */
105 struct vxlan_dev {
106 struct hlist_node hlist;
107 struct net_device *dev;
108 __u32 vni; /* virtual network id */
109 __be32 gaddr; /* multicast group */
110 __be32 saddr; /* source address */
111 unsigned int link; /* link to multicast over */
112 __u16 port_min; /* source port range */
113 __u16 port_max;
114 __u8 tos; /* TOS override */
115 __u8 ttl;
116 u32 flags; /* VXLAN_F_* below */
117
118 unsigned long age_interval;
119 struct timer_list age_timer;
120 spinlock_t hash_lock;
121 unsigned int addrcnt;
122 unsigned int addrmax;
123
124 struct hlist_head fdb_head[FDB_HASH_SIZE];
125 };
126
127 #define VXLAN_F_LEARN 0x01
128 #define VXLAN_F_PROXY 0x02
129 #define VXLAN_F_RSC 0x04
130 #define VXLAN_F_L2MISS 0x08
131 #define VXLAN_F_L3MISS 0x10
132
133 /* salt for hash table */
134 static u32 vxlan_salt __read_mostly;
135
136 static inline struct hlist_head *vni_head(struct net *net, u32 id)
137 {
138 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
139
140 return &vn->vni_list[hash_32(id, VNI_HASH_BITS)];
141 }
142
143 /* Look up VNI in a per net namespace table */
144 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
145 {
146 struct vxlan_dev *vxlan;
147
148 hlist_for_each_entry_rcu(vxlan, vni_head(net, id), hlist) {
149 if (vxlan->vni == id)
150 return vxlan;
151 }
152
153 return NULL;
154 }
155
156 /* Fill in neighbour message in skbuff. */
157 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
158 const struct vxlan_fdb *fdb,
159 u32 portid, u32 seq, int type, unsigned int flags,
160 const struct vxlan_rdst *rdst)
161 {
162 unsigned long now = jiffies;
163 struct nda_cacheinfo ci;
164 struct nlmsghdr *nlh;
165 struct ndmsg *ndm;
166 bool send_ip, send_eth;
167
168 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
169 if (nlh == NULL)
170 return -EMSGSIZE;
171
172 ndm = nlmsg_data(nlh);
173 memset(ndm, 0, sizeof(*ndm));
174
175 send_eth = send_ip = true;
176
177 if (type == RTM_GETNEIGH) {
178 ndm->ndm_family = AF_INET;
179 send_ip = rdst->remote_ip != htonl(INADDR_ANY);
180 send_eth = !is_zero_ether_addr(fdb->eth_addr);
181 } else
182 ndm->ndm_family = AF_BRIDGE;
183 ndm->ndm_state = fdb->state;
184 ndm->ndm_ifindex = vxlan->dev->ifindex;
185 ndm->ndm_flags = NTF_SELF;
186 ndm->ndm_type = NDA_DST;
187
188 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
189 goto nla_put_failure;
190
191 if (send_ip && nla_put_be32(skb, NDA_DST, rdst->remote_ip))
192 goto nla_put_failure;
193
194 if (rdst->remote_port && rdst->remote_port != vxlan_port &&
195 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
196 goto nla_put_failure;
197 if (rdst->remote_vni != vxlan->vni &&
198 nla_put_be32(skb, NDA_VNI, rdst->remote_vni))
199 goto nla_put_failure;
200 if (rdst->remote_ifindex &&
201 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
202 goto nla_put_failure;
203
204 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
205 ci.ndm_confirmed = 0;
206 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
207 ci.ndm_refcnt = 0;
208
209 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
210 goto nla_put_failure;
211
212 return nlmsg_end(skb, nlh);
213
214 nla_put_failure:
215 nlmsg_cancel(skb, nlh);
216 return -EMSGSIZE;
217 }
218
219 static inline size_t vxlan_nlmsg_size(void)
220 {
221 return NLMSG_ALIGN(sizeof(struct ndmsg))
222 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
223 + nla_total_size(sizeof(__be32)) /* NDA_DST */
224 + nla_total_size(sizeof(__be32)) /* NDA_PORT */
225 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
226 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
227 + nla_total_size(sizeof(struct nda_cacheinfo));
228 }
229
230 static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
231 const struct vxlan_fdb *fdb, int type)
232 {
233 struct net *net = dev_net(vxlan->dev);
234 struct sk_buff *skb;
235 int err = -ENOBUFS;
236
237 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
238 if (skb == NULL)
239 goto errout;
240
241 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, &fdb->remote);
242 if (err < 0) {
243 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
244 WARN_ON(err == -EMSGSIZE);
245 kfree_skb(skb);
246 goto errout;
247 }
248
249 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
250 return;
251 errout:
252 if (err < 0)
253 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
254 }
255
256 static void vxlan_ip_miss(struct net_device *dev, __be32 ipa)
257 {
258 struct vxlan_dev *vxlan = netdev_priv(dev);
259 struct vxlan_fdb f;
260
261 memset(&f, 0, sizeof f);
262 f.state = NUD_STALE;
263 f.remote.remote_ip = ipa; /* goes to NDA_DST */
264 f.remote.remote_vni = VXLAN_N_VID;
265
266 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
267 }
268
269 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
270 {
271 struct vxlan_fdb f;
272
273 memset(&f, 0, sizeof f);
274 f.state = NUD_STALE;
275 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
276
277 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
278 }
279
280 /* Hash Ethernet address */
281 static u32 eth_hash(const unsigned char *addr)
282 {
283 u64 value = get_unaligned((u64 *)addr);
284
285 /* only want 6 bytes */
286 #ifdef __BIG_ENDIAN
287 value >>= 16;
288 #else
289 value <<= 16;
290 #endif
291 return hash_64(value, FDB_HASH_BITS);
292 }
293
294 /* Hash chain to use given mac address */
295 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
296 const u8 *mac)
297 {
298 return &vxlan->fdb_head[eth_hash(mac)];
299 }
300
301 /* Look up Ethernet address in forwarding table */
302 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
303 const u8 *mac)
304
305 {
306 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
307 struct vxlan_fdb *f;
308
309 hlist_for_each_entry_rcu(f, head, hlist) {
310 if (compare_ether_addr(mac, f->eth_addr) == 0)
311 return f;
312 }
313
314 return NULL;
315 }
316
317 /* Add/update destinations for multicast */
318 static int vxlan_fdb_append(struct vxlan_fdb *f,
319 __be32 ip, __u32 port, __u32 vni, __u32 ifindex)
320 {
321 struct vxlan_rdst *rd_prev, *rd;
322
323 rd_prev = NULL;
324 for (rd = &f->remote; rd; rd = rd->remote_next) {
325 if (rd->remote_ip == ip &&
326 rd->remote_port == port &&
327 rd->remote_vni == vni &&
328 rd->remote_ifindex == ifindex)
329 return 0;
330 rd_prev = rd;
331 }
332 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
333 if (rd == NULL)
334 return -ENOBUFS;
335 rd->remote_ip = ip;
336 rd->remote_port = port;
337 rd->remote_vni = vni;
338 rd->remote_ifindex = ifindex;
339 rd->remote_next = NULL;
340 rd_prev->remote_next = rd;
341 return 1;
342 }
343
344 /* Add new entry to forwarding table -- assumes lock held */
345 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
346 const u8 *mac, __be32 ip,
347 __u16 state, __u16 flags,
348 __u32 port, __u32 vni, __u32 ifindex)
349 {
350 struct vxlan_fdb *f;
351 int notify = 0;
352
353 f = vxlan_find_mac(vxlan, mac);
354 if (f) {
355 if (flags & NLM_F_EXCL) {
356 netdev_dbg(vxlan->dev,
357 "lost race to create %pM\n", mac);
358 return -EEXIST;
359 }
360 if (f->state != state) {
361 f->state = state;
362 f->updated = jiffies;
363 notify = 1;
364 }
365 if ((flags & NLM_F_APPEND) &&
366 is_multicast_ether_addr(f->eth_addr)) {
367 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex);
368
369 if (rc < 0)
370 return rc;
371 notify |= rc;
372 }
373 } else {
374 if (!(flags & NLM_F_CREATE))
375 return -ENOENT;
376
377 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
378 return -ENOSPC;
379
380 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
381 f = kmalloc(sizeof(*f), GFP_ATOMIC);
382 if (!f)
383 return -ENOMEM;
384
385 notify = 1;
386 f->remote.remote_ip = ip;
387 f->remote.remote_port = port;
388 f->remote.remote_vni = vni;
389 f->remote.remote_ifindex = ifindex;
390 f->remote.remote_next = NULL;
391 f->state = state;
392 f->updated = f->used = jiffies;
393 memcpy(f->eth_addr, mac, ETH_ALEN);
394
395 ++vxlan->addrcnt;
396 hlist_add_head_rcu(&f->hlist,
397 vxlan_fdb_head(vxlan, mac));
398 }
399
400 if (notify)
401 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
402
403 return 0;
404 }
405
406 void vxlan_fdb_free(struct rcu_head *head)
407 {
408 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
409
410 while (f->remote.remote_next) {
411 struct vxlan_rdst *rd = f->remote.remote_next;
412
413 f->remote.remote_next = rd->remote_next;
414 kfree(rd);
415 }
416 kfree(f);
417 }
418
419 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
420 {
421 netdev_dbg(vxlan->dev,
422 "delete %pM\n", f->eth_addr);
423
424 --vxlan->addrcnt;
425 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
426
427 hlist_del_rcu(&f->hlist);
428 call_rcu(&f->rcu, vxlan_fdb_free);
429 }
430
431 /* Add static entry (via netlink) */
432 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
433 struct net_device *dev,
434 const unsigned char *addr, u16 flags)
435 {
436 struct vxlan_dev *vxlan = netdev_priv(dev);
437 struct net *net = dev_net(vxlan->dev);
438 __be32 ip;
439 u32 port, vni, ifindex;
440 int err;
441
442 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
443 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
444 ndm->ndm_state);
445 return -EINVAL;
446 }
447
448 if (tb[NDA_DST] == NULL)
449 return -EINVAL;
450
451 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
452 return -EAFNOSUPPORT;
453
454 ip = nla_get_be32(tb[NDA_DST]);
455
456 if (tb[NDA_PORT]) {
457 if (nla_len(tb[NDA_PORT]) != sizeof(u32))
458 return -EINVAL;
459 port = nla_get_u32(tb[NDA_PORT]);
460 } else
461 port = vxlan_port;
462
463 if (tb[NDA_VNI]) {
464 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
465 return -EINVAL;
466 vni = nla_get_u32(tb[NDA_VNI]);
467 } else
468 vni = vxlan->vni;
469
470 if (tb[NDA_IFINDEX]) {
471 struct net_device *dev;
472
473 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
474 return -EINVAL;
475 ifindex = nla_get_u32(tb[NDA_IFINDEX]);
476 dev = dev_get_by_index(net, ifindex);
477 if (!dev)
478 return -EADDRNOTAVAIL;
479 dev_put(dev);
480 } else
481 ifindex = 0;
482
483 spin_lock_bh(&vxlan->hash_lock);
484 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags, port,
485 vni, ifindex);
486 spin_unlock_bh(&vxlan->hash_lock);
487
488 return err;
489 }
490
491 /* Delete entry (via netlink) */
492 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
493 struct net_device *dev,
494 const unsigned char *addr)
495 {
496 struct vxlan_dev *vxlan = netdev_priv(dev);
497 struct vxlan_fdb *f;
498 int err = -ENOENT;
499
500 spin_lock_bh(&vxlan->hash_lock);
501 f = vxlan_find_mac(vxlan, addr);
502 if (f) {
503 vxlan_fdb_destroy(vxlan, f);
504 err = 0;
505 }
506 spin_unlock_bh(&vxlan->hash_lock);
507
508 return err;
509 }
510
511 /* Dump forwarding table */
512 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
513 struct net_device *dev, int idx)
514 {
515 struct vxlan_dev *vxlan = netdev_priv(dev);
516 unsigned int h;
517
518 for (h = 0; h < FDB_HASH_SIZE; ++h) {
519 struct vxlan_fdb *f;
520 int err;
521
522 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
523 struct vxlan_rdst *rd;
524 for (rd = &f->remote; rd; rd = rd->remote_next) {
525 if (idx < cb->args[0])
526 goto skip;
527
528 err = vxlan_fdb_info(skb, vxlan, f,
529 NETLINK_CB(cb->skb).portid,
530 cb->nlh->nlmsg_seq,
531 RTM_NEWNEIGH,
532 NLM_F_MULTI, rd);
533 if (err < 0)
534 break;
535 skip:
536 ++idx;
537 }
538 }
539 }
540
541 return idx;
542 }
543
544 /* Watch incoming packets to learn mapping between Ethernet address
545 * and Tunnel endpoint.
546 */
547 static void vxlan_snoop(struct net_device *dev,
548 __be32 src_ip, const u8 *src_mac)
549 {
550 struct vxlan_dev *vxlan = netdev_priv(dev);
551 struct vxlan_fdb *f;
552 int err;
553
554 f = vxlan_find_mac(vxlan, src_mac);
555 if (likely(f)) {
556 f->used = jiffies;
557 if (likely(f->remote.remote_ip == src_ip))
558 return;
559
560 if (net_ratelimit())
561 netdev_info(dev,
562 "%pM migrated from %pI4 to %pI4\n",
563 src_mac, &f->remote.remote_ip, &src_ip);
564
565 f->remote.remote_ip = src_ip;
566 f->updated = jiffies;
567 } else {
568 /* learned new entry */
569 spin_lock(&vxlan->hash_lock);
570 err = vxlan_fdb_create(vxlan, src_mac, src_ip,
571 NUD_REACHABLE,
572 NLM_F_EXCL|NLM_F_CREATE,
573 vxlan_port, vxlan->vni, 0);
574 spin_unlock(&vxlan->hash_lock);
575 }
576 }
577
578
579 /* See if multicast group is already in use by other ID */
580 static bool vxlan_group_used(struct vxlan_net *vn,
581 const struct vxlan_dev *this)
582 {
583 const struct vxlan_dev *vxlan;
584 unsigned h;
585
586 for (h = 0; h < VNI_HASH_SIZE; ++h)
587 hlist_for_each_entry(vxlan, &vn->vni_list[h], hlist) {
588 if (vxlan == this)
589 continue;
590
591 if (!netif_running(vxlan->dev))
592 continue;
593
594 if (vxlan->gaddr == this->gaddr)
595 return true;
596 }
597
598 return false;
599 }
600
601 /* kernel equivalent to IP_ADD_MEMBERSHIP */
602 static int vxlan_join_group(struct net_device *dev)
603 {
604 struct vxlan_dev *vxlan = netdev_priv(dev);
605 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
606 struct sock *sk = vn->sock->sk;
607 struct ip_mreqn mreq = {
608 .imr_multiaddr.s_addr = vxlan->gaddr,
609 .imr_ifindex = vxlan->link,
610 };
611 int err;
612
613 /* Already a member of group */
614 if (vxlan_group_used(vn, vxlan))
615 return 0;
616
617 /* Need to drop RTNL to call multicast join */
618 rtnl_unlock();
619 lock_sock(sk);
620 err = ip_mc_join_group(sk, &mreq);
621 release_sock(sk);
622 rtnl_lock();
623
624 return err;
625 }
626
627
628 /* kernel equivalent to IP_DROP_MEMBERSHIP */
629 static int vxlan_leave_group(struct net_device *dev)
630 {
631 struct vxlan_dev *vxlan = netdev_priv(dev);
632 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
633 int err = 0;
634 struct sock *sk = vn->sock->sk;
635 struct ip_mreqn mreq = {
636 .imr_multiaddr.s_addr = vxlan->gaddr,
637 .imr_ifindex = vxlan->link,
638 };
639
640 /* Only leave group when last vxlan is done. */
641 if (vxlan_group_used(vn, vxlan))
642 return 0;
643
644 /* Need to drop RTNL to call multicast leave */
645 rtnl_unlock();
646 lock_sock(sk);
647 err = ip_mc_leave_group(sk, &mreq);
648 release_sock(sk);
649 rtnl_lock();
650
651 return err;
652 }
653
654 /* Callback from net/ipv4/udp.c to receive packets */
655 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
656 {
657 struct iphdr *oip;
658 struct vxlanhdr *vxh;
659 struct vxlan_dev *vxlan;
660 struct pcpu_tstats *stats;
661 __u32 vni;
662 int err;
663
664 /* pop off outer UDP header */
665 __skb_pull(skb, sizeof(struct udphdr));
666
667 /* Need Vxlan and inner Ethernet header to be present */
668 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
669 goto error;
670
671 /* Drop packets with reserved bits set */
672 vxh = (struct vxlanhdr *) skb->data;
673 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
674 (vxh->vx_vni & htonl(0xff))) {
675 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
676 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
677 goto error;
678 }
679
680 __skb_pull(skb, sizeof(struct vxlanhdr));
681
682 /* Is this VNI defined? */
683 vni = ntohl(vxh->vx_vni) >> 8;
684 vxlan = vxlan_find_vni(sock_net(sk), vni);
685 if (!vxlan) {
686 netdev_dbg(skb->dev, "unknown vni %d\n", vni);
687 goto drop;
688 }
689
690 if (!pskb_may_pull(skb, ETH_HLEN)) {
691 vxlan->dev->stats.rx_length_errors++;
692 vxlan->dev->stats.rx_errors++;
693 goto drop;
694 }
695
696 skb_reset_mac_header(skb);
697
698 /* Re-examine inner Ethernet packet */
699 oip = ip_hdr(skb);
700 skb->protocol = eth_type_trans(skb, vxlan->dev);
701
702 /* Ignore packet loops (and multicast echo) */
703 if (compare_ether_addr(eth_hdr(skb)->h_source,
704 vxlan->dev->dev_addr) == 0)
705 goto drop;
706
707 if (vxlan->flags & VXLAN_F_LEARN)
708 vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source);
709
710 __skb_tunnel_rx(skb, vxlan->dev);
711 skb_reset_network_header(skb);
712
713 /* If the NIC driver gave us an encapsulated packet with
714 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
715 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
716 * for us. Otherwise force the upper layers to verify it.
717 */
718 if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
719 !(vxlan->dev->features & NETIF_F_RXCSUM))
720 skb->ip_summed = CHECKSUM_NONE;
721
722 skb->encapsulation = 0;
723
724 err = IP_ECN_decapsulate(oip, skb);
725 if (unlikely(err)) {
726 if (log_ecn_error)
727 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
728 &oip->saddr, oip->tos);
729 if (err > 1) {
730 ++vxlan->dev->stats.rx_frame_errors;
731 ++vxlan->dev->stats.rx_errors;
732 goto drop;
733 }
734 }
735
736 stats = this_cpu_ptr(vxlan->dev->tstats);
737 u64_stats_update_begin(&stats->syncp);
738 stats->rx_packets++;
739 stats->rx_bytes += skb->len;
740 u64_stats_update_end(&stats->syncp);
741
742 netif_rx(skb);
743
744 return 0;
745 error:
746 /* Put UDP header back */
747 __skb_push(skb, sizeof(struct udphdr));
748
749 return 1;
750 drop:
751 /* Consume bad packet */
752 kfree_skb(skb);
753 return 0;
754 }
755
756 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
757 {
758 struct vxlan_dev *vxlan = netdev_priv(dev);
759 struct arphdr *parp;
760 u8 *arpptr, *sha;
761 __be32 sip, tip;
762 struct neighbour *n;
763
764 if (dev->flags & IFF_NOARP)
765 goto out;
766
767 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
768 dev->stats.tx_dropped++;
769 goto out;
770 }
771 parp = arp_hdr(skb);
772
773 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
774 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
775 parp->ar_pro != htons(ETH_P_IP) ||
776 parp->ar_op != htons(ARPOP_REQUEST) ||
777 parp->ar_hln != dev->addr_len ||
778 parp->ar_pln != 4)
779 goto out;
780 arpptr = (u8 *)parp + sizeof(struct arphdr);
781 sha = arpptr;
782 arpptr += dev->addr_len; /* sha */
783 memcpy(&sip, arpptr, sizeof(sip));
784 arpptr += sizeof(sip);
785 arpptr += dev->addr_len; /* tha */
786 memcpy(&tip, arpptr, sizeof(tip));
787
788 if (ipv4_is_loopback(tip) ||
789 ipv4_is_multicast(tip))
790 goto out;
791
792 n = neigh_lookup(&arp_tbl, &tip, dev);
793
794 if (n) {
795 struct vxlan_dev *vxlan = netdev_priv(dev);
796 struct vxlan_fdb *f;
797 struct sk_buff *reply;
798
799 if (!(n->nud_state & NUD_CONNECTED)) {
800 neigh_release(n);
801 goto out;
802 }
803
804 f = vxlan_find_mac(vxlan, n->ha);
805 if (f && f->remote.remote_ip == htonl(INADDR_ANY)) {
806 /* bridge-local neighbor */
807 neigh_release(n);
808 goto out;
809 }
810
811 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
812 n->ha, sha);
813
814 neigh_release(n);
815
816 skb_reset_mac_header(reply);
817 __skb_pull(reply, skb_network_offset(reply));
818 reply->ip_summed = CHECKSUM_UNNECESSARY;
819 reply->pkt_type = PACKET_HOST;
820
821 if (netif_rx_ni(reply) == NET_RX_DROP)
822 dev->stats.rx_dropped++;
823 } else if (vxlan->flags & VXLAN_F_L3MISS)
824 vxlan_ip_miss(dev, tip);
825 out:
826 consume_skb(skb);
827 return NETDEV_TX_OK;
828 }
829
830 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
831 {
832 struct vxlan_dev *vxlan = netdev_priv(dev);
833 struct neighbour *n;
834 struct iphdr *pip;
835
836 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
837 return false;
838
839 n = NULL;
840 switch (ntohs(eth_hdr(skb)->h_proto)) {
841 case ETH_P_IP:
842 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
843 return false;
844 pip = ip_hdr(skb);
845 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
846 break;
847 default:
848 return false;
849 }
850
851 if (n) {
852 bool diff;
853
854 diff = compare_ether_addr(eth_hdr(skb)->h_dest, n->ha) != 0;
855 if (diff) {
856 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
857 dev->addr_len);
858 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
859 }
860 neigh_release(n);
861 return diff;
862 } else if (vxlan->flags & VXLAN_F_L3MISS)
863 vxlan_ip_miss(dev, pip->daddr);
864 return false;
865 }
866
867 /* Extract dsfield from inner protocol */
868 static inline u8 vxlan_get_dsfield(const struct iphdr *iph,
869 const struct sk_buff *skb)
870 {
871 if (skb->protocol == htons(ETH_P_IP))
872 return iph->tos;
873 else if (skb->protocol == htons(ETH_P_IPV6))
874 return ipv6_get_dsfield((const struct ipv6hdr *)iph);
875 else
876 return 0;
877 }
878
879 /* Propogate ECN bits out */
880 static inline u8 vxlan_ecn_encap(u8 tos,
881 const struct iphdr *iph,
882 const struct sk_buff *skb)
883 {
884 u8 inner = vxlan_get_dsfield(iph, skb);
885
886 return INET_ECN_encapsulate(tos, inner);
887 }
888
889 static void vxlan_sock_free(struct sk_buff *skb)
890 {
891 sock_put(skb->sk);
892 }
893
894 /* On transmit, associate with the tunnel socket */
895 static void vxlan_set_owner(struct net_device *dev, struct sk_buff *skb)
896 {
897 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
898 struct sock *sk = vn->sock->sk;
899
900 skb_orphan(skb);
901 sock_hold(sk);
902 skb->sk = sk;
903 skb->destructor = vxlan_sock_free;
904 }
905
906 /* Compute source port for outgoing packet
907 * first choice to use L4 flow hash since it will spread
908 * better and maybe available from hardware
909 * secondary choice is to use jhash on the Ethernet header
910 */
911 static u16 vxlan_src_port(const struct vxlan_dev *vxlan, struct sk_buff *skb)
912 {
913 unsigned int range = (vxlan->port_max - vxlan->port_min) + 1;
914 u32 hash;
915
916 hash = skb_get_rxhash(skb);
917 if (!hash)
918 hash = jhash(skb->data, 2 * ETH_ALEN,
919 (__force u32) skb->protocol);
920
921 return (((u64) hash * range) >> 32) + vxlan->port_min;
922 }
923
924 static int handle_offloads(struct sk_buff *skb)
925 {
926 if (skb_is_gso(skb)) {
927 int err = skb_unclone(skb, GFP_ATOMIC);
928 if (unlikely(err))
929 return err;
930
931 skb_shinfo(skb)->gso_type |= (SKB_GSO_UDP_TUNNEL | SKB_GSO_UDP);
932 } else if (skb->ip_summed != CHECKSUM_PARTIAL)
933 skb->ip_summed = CHECKSUM_NONE;
934
935 return 0;
936 }
937
938 static netdev_tx_t vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
939 struct vxlan_rdst *rdst, bool did_rsc)
940 {
941 struct vxlan_dev *vxlan = netdev_priv(dev);
942 struct rtable *rt;
943 const struct iphdr *old_iph;
944 struct iphdr *iph;
945 struct vxlanhdr *vxh;
946 struct udphdr *uh;
947 struct flowi4 fl4;
948 unsigned int pkt_len = skb->len;
949 __be32 dst;
950 __u16 src_port, dst_port;
951 u32 vni;
952 __be16 df = 0;
953 __u8 tos, ttl;
954
955 dst_port = rdst->remote_port ? rdst->remote_port : vxlan_port;
956 vni = rdst->remote_vni;
957 dst = rdst->remote_ip;
958
959 if (!dst) {
960 if (did_rsc) {
961 __skb_pull(skb, skb_network_offset(skb));
962 skb->ip_summed = CHECKSUM_NONE;
963 skb->pkt_type = PACKET_HOST;
964
965 /* short-circuited back to local bridge */
966 if (netif_rx(skb) == NET_RX_SUCCESS) {
967 struct pcpu_tstats *stats = this_cpu_ptr(dev->tstats);
968
969 u64_stats_update_begin(&stats->syncp);
970 stats->tx_packets++;
971 stats->tx_bytes += pkt_len;
972 u64_stats_update_end(&stats->syncp);
973 } else {
974 dev->stats.tx_errors++;
975 dev->stats.tx_aborted_errors++;
976 }
977 return NETDEV_TX_OK;
978 }
979 goto drop;
980 }
981
982 if (!skb->encapsulation) {
983 skb_reset_inner_headers(skb);
984 skb->encapsulation = 1;
985 }
986
987 /* Need space for new headers (invalidates iph ptr) */
988 if (skb_cow_head(skb, VXLAN_HEADROOM))
989 goto drop;
990
991 old_iph = ip_hdr(skb);
992
993 ttl = vxlan->ttl;
994 if (!ttl && IN_MULTICAST(ntohl(dst)))
995 ttl = 1;
996
997 tos = vxlan->tos;
998 if (tos == 1)
999 tos = vxlan_get_dsfield(old_iph, skb);
1000
1001 src_port = vxlan_src_port(vxlan, skb);
1002
1003 memset(&fl4, 0, sizeof(fl4));
1004 fl4.flowi4_oif = rdst->remote_ifindex;
1005 fl4.flowi4_tos = RT_TOS(tos);
1006 fl4.daddr = dst;
1007 fl4.saddr = vxlan->saddr;
1008
1009 rt = ip_route_output_key(dev_net(dev), &fl4);
1010 if (IS_ERR(rt)) {
1011 netdev_dbg(dev, "no route to %pI4\n", &dst);
1012 dev->stats.tx_carrier_errors++;
1013 goto tx_error;
1014 }
1015
1016 if (rt->dst.dev == dev) {
1017 netdev_dbg(dev, "circular route to %pI4\n", &dst);
1018 ip_rt_put(rt);
1019 dev->stats.collisions++;
1020 goto tx_error;
1021 }
1022
1023 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1024 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1025 IPSKB_REROUTED);
1026 skb_dst_drop(skb);
1027 skb_dst_set(skb, &rt->dst);
1028
1029 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1030 vxh->vx_flags = htonl(VXLAN_FLAGS);
1031 vxh->vx_vni = htonl(vni << 8);
1032
1033 __skb_push(skb, sizeof(*uh));
1034 skb_reset_transport_header(skb);
1035 uh = udp_hdr(skb);
1036
1037 uh->dest = htons(dst_port);
1038 uh->source = htons(src_port);
1039
1040 uh->len = htons(skb->len);
1041 uh->check = 0;
1042
1043 __skb_push(skb, sizeof(*iph));
1044 skb_reset_network_header(skb);
1045 iph = ip_hdr(skb);
1046 iph->version = 4;
1047 iph->ihl = sizeof(struct iphdr) >> 2;
1048 iph->frag_off = df;
1049 iph->protocol = IPPROTO_UDP;
1050 iph->tos = vxlan_ecn_encap(tos, old_iph, skb);
1051 iph->daddr = dst;
1052 iph->saddr = fl4.saddr;
1053 iph->ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1054 tunnel_ip_select_ident(skb, old_iph, &rt->dst);
1055
1056 nf_reset(skb);
1057
1058 vxlan_set_owner(dev, skb);
1059
1060 if (handle_offloads(skb))
1061 goto drop;
1062
1063 iptunnel_xmit(skb, dev);
1064 return NETDEV_TX_OK;
1065
1066 drop:
1067 dev->stats.tx_dropped++;
1068 goto tx_free;
1069
1070 tx_error:
1071 dev->stats.tx_errors++;
1072 tx_free:
1073 dev_kfree_skb(skb);
1074 return NETDEV_TX_OK;
1075 }
1076
1077 /* Transmit local packets over Vxlan
1078 *
1079 * Outer IP header inherits ECN and DF from inner header.
1080 * Outer UDP destination is the VXLAN assigned port.
1081 * source port is based on hash of flow
1082 */
1083 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1084 {
1085 struct vxlan_dev *vxlan = netdev_priv(dev);
1086 struct ethhdr *eth;
1087 bool did_rsc = false;
1088 struct vxlan_rdst group, *rdst0, *rdst;
1089 struct vxlan_fdb *f;
1090 int rc1, rc;
1091
1092 skb_reset_mac_header(skb);
1093 eth = eth_hdr(skb);
1094
1095 if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP)
1096 return arp_reduce(dev, skb);
1097 else if ((vxlan->flags&VXLAN_F_RSC) && ntohs(eth->h_proto) == ETH_P_IP)
1098 did_rsc = route_shortcircuit(dev, skb);
1099
1100 f = vxlan_find_mac(vxlan, eth->h_dest);
1101 if (f == NULL) {
1102 did_rsc = false;
1103 group.remote_port = vxlan_port;
1104 group.remote_vni = vxlan->vni;
1105 group.remote_ip = vxlan->gaddr;
1106 group.remote_ifindex = vxlan->link;
1107 group.remote_next = 0;
1108 rdst0 = &group;
1109
1110 if (group.remote_ip == htonl(INADDR_ANY) &&
1111 (vxlan->flags & VXLAN_F_L2MISS) &&
1112 !is_multicast_ether_addr(eth->h_dest))
1113 vxlan_fdb_miss(vxlan, eth->h_dest);
1114 } else
1115 rdst0 = &f->remote;
1116
1117 rc = NETDEV_TX_OK;
1118
1119 /* if there are multiple destinations, send copies */
1120 for (rdst = rdst0->remote_next; rdst; rdst = rdst->remote_next) {
1121 struct sk_buff *skb1;
1122
1123 skb1 = skb_clone(skb, GFP_ATOMIC);
1124 rc1 = vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1125 if (rc == NETDEV_TX_OK)
1126 rc = rc1;
1127 }
1128
1129 rc1 = vxlan_xmit_one(skb, dev, rdst0, did_rsc);
1130 if (rc == NETDEV_TX_OK)
1131 rc = rc1;
1132 return rc;
1133 }
1134
1135 /* Walk the forwarding table and purge stale entries */
1136 static void vxlan_cleanup(unsigned long arg)
1137 {
1138 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1139 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1140 unsigned int h;
1141
1142 if (!netif_running(vxlan->dev))
1143 return;
1144
1145 spin_lock_bh(&vxlan->hash_lock);
1146 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1147 struct hlist_node *p, *n;
1148 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1149 struct vxlan_fdb *f
1150 = container_of(p, struct vxlan_fdb, hlist);
1151 unsigned long timeout;
1152
1153 if (f->state & NUD_PERMANENT)
1154 continue;
1155
1156 timeout = f->used + vxlan->age_interval * HZ;
1157 if (time_before_eq(timeout, jiffies)) {
1158 netdev_dbg(vxlan->dev,
1159 "garbage collect %pM\n",
1160 f->eth_addr);
1161 f->state = NUD_STALE;
1162 vxlan_fdb_destroy(vxlan, f);
1163 } else if (time_before(timeout, next_timer))
1164 next_timer = timeout;
1165 }
1166 }
1167 spin_unlock_bh(&vxlan->hash_lock);
1168
1169 mod_timer(&vxlan->age_timer, next_timer);
1170 }
1171
1172 /* Setup stats when device is created */
1173 static int vxlan_init(struct net_device *dev)
1174 {
1175 dev->tstats = alloc_percpu(struct pcpu_tstats);
1176 if (!dev->tstats)
1177 return -ENOMEM;
1178
1179 return 0;
1180 }
1181
1182 /* Start ageing timer and join group when device is brought up */
1183 static int vxlan_open(struct net_device *dev)
1184 {
1185 struct vxlan_dev *vxlan = netdev_priv(dev);
1186 int err;
1187
1188 if (vxlan->gaddr) {
1189 err = vxlan_join_group(dev);
1190 if (err)
1191 return err;
1192 }
1193
1194 if (vxlan->age_interval)
1195 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
1196
1197 return 0;
1198 }
1199
1200 /* Purge the forwarding table */
1201 static void vxlan_flush(struct vxlan_dev *vxlan)
1202 {
1203 unsigned h;
1204
1205 spin_lock_bh(&vxlan->hash_lock);
1206 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1207 struct hlist_node *p, *n;
1208 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1209 struct vxlan_fdb *f
1210 = container_of(p, struct vxlan_fdb, hlist);
1211 vxlan_fdb_destroy(vxlan, f);
1212 }
1213 }
1214 spin_unlock_bh(&vxlan->hash_lock);
1215 }
1216
1217 /* Cleanup timer and forwarding table on shutdown */
1218 static int vxlan_stop(struct net_device *dev)
1219 {
1220 struct vxlan_dev *vxlan = netdev_priv(dev);
1221
1222 if (vxlan->gaddr)
1223 vxlan_leave_group(dev);
1224
1225 del_timer_sync(&vxlan->age_timer);
1226
1227 vxlan_flush(vxlan);
1228
1229 return 0;
1230 }
1231
1232 /* Stub, nothing needs to be done. */
1233 static void vxlan_set_multicast_list(struct net_device *dev)
1234 {
1235 }
1236
1237 static const struct net_device_ops vxlan_netdev_ops = {
1238 .ndo_init = vxlan_init,
1239 .ndo_open = vxlan_open,
1240 .ndo_stop = vxlan_stop,
1241 .ndo_start_xmit = vxlan_xmit,
1242 .ndo_get_stats64 = ip_tunnel_get_stats64,
1243 .ndo_set_rx_mode = vxlan_set_multicast_list,
1244 .ndo_change_mtu = eth_change_mtu,
1245 .ndo_validate_addr = eth_validate_addr,
1246 .ndo_set_mac_address = eth_mac_addr,
1247 .ndo_fdb_add = vxlan_fdb_add,
1248 .ndo_fdb_del = vxlan_fdb_delete,
1249 .ndo_fdb_dump = vxlan_fdb_dump,
1250 };
1251
1252 /* Info for udev, that this is a virtual tunnel endpoint */
1253 static struct device_type vxlan_type = {
1254 .name = "vxlan",
1255 };
1256
1257 static void vxlan_free(struct net_device *dev)
1258 {
1259 free_percpu(dev->tstats);
1260 free_netdev(dev);
1261 }
1262
1263 /* Initialize the device structure. */
1264 static void vxlan_setup(struct net_device *dev)
1265 {
1266 struct vxlan_dev *vxlan = netdev_priv(dev);
1267 unsigned h;
1268 int low, high;
1269
1270 eth_hw_addr_random(dev);
1271 ether_setup(dev);
1272 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
1273
1274 dev->netdev_ops = &vxlan_netdev_ops;
1275 dev->destructor = vxlan_free;
1276 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
1277
1278 dev->tx_queue_len = 0;
1279 dev->features |= NETIF_F_LLTX;
1280 dev->features |= NETIF_F_NETNS_LOCAL;
1281 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1282 dev->features |= NETIF_F_RXCSUM;
1283 dev->features |= NETIF_F_GSO_SOFTWARE;
1284
1285 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1286 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1287 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1288 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1289
1290 spin_lock_init(&vxlan->hash_lock);
1291
1292 init_timer_deferrable(&vxlan->age_timer);
1293 vxlan->age_timer.function = vxlan_cleanup;
1294 vxlan->age_timer.data = (unsigned long) vxlan;
1295
1296 inet_get_local_port_range(&low, &high);
1297 vxlan->port_min = low;
1298 vxlan->port_max = high;
1299
1300 vxlan->dev = dev;
1301
1302 for (h = 0; h < FDB_HASH_SIZE; ++h)
1303 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
1304 }
1305
1306 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
1307 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
1308 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1309 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
1310 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1311 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
1312 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
1313 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
1314 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
1315 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
1316 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
1317 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
1318 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
1319 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
1320 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
1321 };
1322
1323 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
1324 {
1325 if (tb[IFLA_ADDRESS]) {
1326 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1327 pr_debug("invalid link address (not ethernet)\n");
1328 return -EINVAL;
1329 }
1330
1331 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1332 pr_debug("invalid all zero ethernet address\n");
1333 return -EADDRNOTAVAIL;
1334 }
1335 }
1336
1337 if (!data)
1338 return -EINVAL;
1339
1340 if (data[IFLA_VXLAN_ID]) {
1341 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
1342 if (id >= VXLAN_VID_MASK)
1343 return -ERANGE;
1344 }
1345
1346 if (data[IFLA_VXLAN_GROUP]) {
1347 __be32 gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1348 if (!IN_MULTICAST(ntohl(gaddr))) {
1349 pr_debug("group address is not IPv4 multicast\n");
1350 return -EADDRNOTAVAIL;
1351 }
1352 }
1353
1354 if (data[IFLA_VXLAN_PORT_RANGE]) {
1355 const struct ifla_vxlan_port_range *p
1356 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1357
1358 if (ntohs(p->high) < ntohs(p->low)) {
1359 pr_debug("port range %u .. %u not valid\n",
1360 ntohs(p->low), ntohs(p->high));
1361 return -EINVAL;
1362 }
1363 }
1364
1365 return 0;
1366 }
1367
1368 static void vxlan_get_drvinfo(struct net_device *netdev,
1369 struct ethtool_drvinfo *drvinfo)
1370 {
1371 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
1372 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
1373 }
1374
1375 static const struct ethtool_ops vxlan_ethtool_ops = {
1376 .get_drvinfo = vxlan_get_drvinfo,
1377 .get_link = ethtool_op_get_link,
1378 };
1379
1380 static int vxlan_newlink(struct net *net, struct net_device *dev,
1381 struct nlattr *tb[], struct nlattr *data[])
1382 {
1383 struct vxlan_dev *vxlan = netdev_priv(dev);
1384 __u32 vni;
1385 int err;
1386
1387 if (!data[IFLA_VXLAN_ID])
1388 return -EINVAL;
1389
1390 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1391 if (vxlan_find_vni(net, vni)) {
1392 pr_info("duplicate VNI %u\n", vni);
1393 return -EEXIST;
1394 }
1395 vxlan->vni = vni;
1396
1397 if (data[IFLA_VXLAN_GROUP])
1398 vxlan->gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1399
1400 if (data[IFLA_VXLAN_LOCAL])
1401 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1402
1403 if (data[IFLA_VXLAN_LINK] &&
1404 (vxlan->link = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
1405 struct net_device *lowerdev
1406 = __dev_get_by_index(net, vxlan->link);
1407
1408 if (!lowerdev) {
1409 pr_info("ifindex %d does not exist\n", vxlan->link);
1410 return -ENODEV;
1411 }
1412
1413 if (!tb[IFLA_MTU])
1414 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
1415
1416 /* update header length based on lower device */
1417 dev->hard_header_len = lowerdev->hard_header_len +
1418 VXLAN_HEADROOM;
1419 }
1420
1421 if (data[IFLA_VXLAN_TOS])
1422 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1423
1424 if (data[IFLA_VXLAN_TTL])
1425 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
1426
1427 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
1428 vxlan->flags |= VXLAN_F_LEARN;
1429
1430 if (data[IFLA_VXLAN_AGEING])
1431 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1432 else
1433 vxlan->age_interval = FDB_AGE_DEFAULT;
1434
1435 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
1436 vxlan->flags |= VXLAN_F_PROXY;
1437
1438 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
1439 vxlan->flags |= VXLAN_F_RSC;
1440
1441 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
1442 vxlan->flags |= VXLAN_F_L2MISS;
1443
1444 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
1445 vxlan->flags |= VXLAN_F_L3MISS;
1446
1447 if (data[IFLA_VXLAN_LIMIT])
1448 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1449
1450 if (data[IFLA_VXLAN_PORT_RANGE]) {
1451 const struct ifla_vxlan_port_range *p
1452 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1453 vxlan->port_min = ntohs(p->low);
1454 vxlan->port_max = ntohs(p->high);
1455 }
1456
1457 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
1458
1459 err = register_netdevice(dev);
1460 if (!err)
1461 hlist_add_head_rcu(&vxlan->hlist, vni_head(net, vxlan->vni));
1462
1463 return err;
1464 }
1465
1466 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1467 {
1468 struct vxlan_dev *vxlan = netdev_priv(dev);
1469
1470 hlist_del_rcu(&vxlan->hlist);
1471
1472 unregister_netdevice_queue(dev, head);
1473 }
1474
1475 static size_t vxlan_get_size(const struct net_device *dev)
1476 {
1477
1478 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1479 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1480 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1481 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1482 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1483 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1484 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
1485 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
1486 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
1487 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
1488 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
1489 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1490 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
1491 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
1492 0;
1493 }
1494
1495 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1496 {
1497 const struct vxlan_dev *vxlan = netdev_priv(dev);
1498 struct ifla_vxlan_port_range ports = {
1499 .low = htons(vxlan->port_min),
1500 .high = htons(vxlan->port_max),
1501 };
1502
1503 if (nla_put_u32(skb, IFLA_VXLAN_ID, vxlan->vni))
1504 goto nla_put_failure;
1505
1506 if (vxlan->gaddr && nla_put_be32(skb, IFLA_VXLAN_GROUP, vxlan->gaddr))
1507 goto nla_put_failure;
1508
1509 if (vxlan->link && nla_put_u32(skb, IFLA_VXLAN_LINK, vxlan->link))
1510 goto nla_put_failure;
1511
1512 if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
1513 goto nla_put_failure;
1514
1515 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1516 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
1517 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
1518 !!(vxlan->flags & VXLAN_F_LEARN)) ||
1519 nla_put_u8(skb, IFLA_VXLAN_PROXY,
1520 !!(vxlan->flags & VXLAN_F_PROXY)) ||
1521 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
1522 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
1523 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
1524 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
1525 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
1526 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1527 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax))
1528 goto nla_put_failure;
1529
1530 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
1531 goto nla_put_failure;
1532
1533 return 0;
1534
1535 nla_put_failure:
1536 return -EMSGSIZE;
1537 }
1538
1539 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1540 .kind = "vxlan",
1541 .maxtype = IFLA_VXLAN_MAX,
1542 .policy = vxlan_policy,
1543 .priv_size = sizeof(struct vxlan_dev),
1544 .setup = vxlan_setup,
1545 .validate = vxlan_validate,
1546 .newlink = vxlan_newlink,
1547 .dellink = vxlan_dellink,
1548 .get_size = vxlan_get_size,
1549 .fill_info = vxlan_fill_info,
1550 };
1551
1552 static __net_init int vxlan_init_net(struct net *net)
1553 {
1554 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1555 struct sock *sk;
1556 struct sockaddr_in vxlan_addr = {
1557 .sin_family = AF_INET,
1558 .sin_addr.s_addr = htonl(INADDR_ANY),
1559 };
1560 int rc;
1561 unsigned h;
1562
1563 /* Create UDP socket for encapsulation receive. */
1564 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vn->sock);
1565 if (rc < 0) {
1566 pr_debug("UDP socket create failed\n");
1567 return rc;
1568 }
1569 /* Put in proper namespace */
1570 sk = vn->sock->sk;
1571 sk_change_net(sk, net);
1572
1573 vxlan_addr.sin_port = htons(vxlan_port);
1574
1575 rc = kernel_bind(vn->sock, (struct sockaddr *) &vxlan_addr,
1576 sizeof(vxlan_addr));
1577 if (rc < 0) {
1578 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1579 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
1580 sk_release_kernel(sk);
1581 vn->sock = NULL;
1582 return rc;
1583 }
1584
1585 /* Disable multicast loopback */
1586 inet_sk(sk)->mc_loop = 0;
1587
1588 /* Mark socket as an encapsulation socket. */
1589 udp_sk(sk)->encap_type = 1;
1590 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1591 udp_encap_enable();
1592
1593 for (h = 0; h < VNI_HASH_SIZE; ++h)
1594 INIT_HLIST_HEAD(&vn->vni_list[h]);
1595
1596 return 0;
1597 }
1598
1599 static __net_exit void vxlan_exit_net(struct net *net)
1600 {
1601 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1602 struct vxlan_dev *vxlan;
1603 unsigned h;
1604
1605 rtnl_lock();
1606 for (h = 0; h < VNI_HASH_SIZE; ++h)
1607 hlist_for_each_entry(vxlan, &vn->vni_list[h], hlist)
1608 dev_close(vxlan->dev);
1609 rtnl_unlock();
1610
1611 if (vn->sock) {
1612 sk_release_kernel(vn->sock->sk);
1613 vn->sock = NULL;
1614 }
1615 }
1616
1617 static struct pernet_operations vxlan_net_ops = {
1618 .init = vxlan_init_net,
1619 .exit = vxlan_exit_net,
1620 .id = &vxlan_net_id,
1621 .size = sizeof(struct vxlan_net),
1622 };
1623
1624 static int __init vxlan_init_module(void)
1625 {
1626 int rc;
1627
1628 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1629
1630 rc = register_pernet_device(&vxlan_net_ops);
1631 if (rc)
1632 goto out1;
1633
1634 rc = rtnl_link_register(&vxlan_link_ops);
1635 if (rc)
1636 goto out2;
1637
1638 return 0;
1639
1640 out2:
1641 unregister_pernet_device(&vxlan_net_ops);
1642 out1:
1643 return rc;
1644 }
1645 module_init(vxlan_init_module);
1646
1647 static void __exit vxlan_cleanup_module(void)
1648 {
1649 rtnl_link_unregister(&vxlan_link_ops);
1650 unregister_pernet_device(&vxlan_net_ops);
1651 rcu_barrier();
1652 }
1653 module_exit(vxlan_cleanup_module);
1654
1655 MODULE_LICENSE("GPL");
1656 MODULE_VERSION(VXLAN_VERSION);
1657 MODULE_AUTHOR("Stephen Hemminger <shemminger@vyatta.com>");
1658 MODULE_ALIAS_RTNL_LINK("vxlan");
Linux kernel - Deliverables
This page took 0.062152 seconds and 4 git commands to generate.