projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
vxlan: cosmetic cleanup's
[deliverable/linux.git] / drivers / net / vxlan.c
1 /*
2 * VXLAN: Virtual eXtensible Local Area Network
3 *
4 * Copyright (c) 2012-2013 Vyatta Inc.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 *
10 * TODO
11 * - IPv6 (not in RFC)
12 */
13
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15
16 #include <linux/kernel.h>
17 #include <linux/types.h>
18 #include <linux/module.h>
19 #include <linux/errno.h>
20 #include <linux/slab.h>
21 #include <linux/skbuff.h>
22 #include <linux/rculist.h>
23 #include <linux/netdevice.h>
24 #include <linux/in.h>
25 #include <linux/ip.h>
26 #include <linux/udp.h>
27 #include <linux/igmp.h>
28 #include <linux/etherdevice.h>
29 #include <linux/if_ether.h>
30 #include <linux/hash.h>
31 #include <linux/ethtool.h>
32 #include <net/arp.h>
33 #include <net/ndisc.h>
34 #include <net/ip.h>
35 #include <net/ip_tunnels.h>
36 #include <net/icmp.h>
37 #include <net/udp.h>
38 #include <net/rtnetlink.h>
39 #include <net/route.h>
40 #include <net/dsfield.h>
41 #include <net/inet_ecn.h>
42 #include <net/net_namespace.h>
43 #include <net/netns/generic.h>
44
45 #define VXLAN_VERSION "0.1"
46
47 #define PORT_HASH_BITS 8
48 #define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
49 #define VNI_HASH_BITS 10
50 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
51 #define FDB_HASH_BITS 8
52 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
53 #define FDB_AGE_DEFAULT 300 /* 5 min */
54 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
55
56 #define VXLAN_N_VID (1u << 24)
57 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
58 /* IP header + UDP + VXLAN + Ethernet header */
59 #define VXLAN_HEADROOM (20 + 8 + 8 + 14)
60
61 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
62
63 /* VXLAN protocol header */
64 struct vxlanhdr {
65 __be32 vx_flags;
66 __be32 vx_vni;
67 };
68
69 /* UDP port for VXLAN traffic.
70 * The IANA assigned port is 4789, but the Linux default is 8472
71 * for compatibility with early adopters.
72 */
73 static unsigned short vxlan_port __read_mostly = 8472;
74 module_param_named(udp_port, vxlan_port, ushort, 0444);
75 MODULE_PARM_DESC(udp_port, "Destination UDP port");
76
77 static bool log_ecn_error = true;
78 module_param(log_ecn_error, bool, 0644);
79 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
80
81 static unsigned int vxlan_net_id;
82
83 /* per UDP socket information */
84 struct vxlan_sock {
85 struct hlist_node hlist;
86 struct rcu_head rcu;
87 struct work_struct del_work;
88 atomic_t refcnt;
89 struct socket *sock;
90 struct hlist_head vni_list[VNI_HASH_SIZE];
91 };
92
93 /* per-network namespace private data for this module */
94 struct vxlan_net {
95 struct list_head vxlan_list;
96 struct hlist_head sock_list[PORT_HASH_SIZE];
97 spinlock_t sock_lock;
98 };
99
100 struct vxlan_rdst {
101 __be32 remote_ip;
102 __be16 remote_port;
103 u32 remote_vni;
104 u32 remote_ifindex;
105 struct list_head list;
106 };
107
108 /* Forwarding table entry */
109 struct vxlan_fdb {
110 struct hlist_node hlist; /* linked list of entries */
111 struct rcu_head rcu;
112 unsigned long updated; /* jiffies */
113 unsigned long used;
114 struct list_head remotes;
115 u16 state; /* see ndm_state */
116 u8 flags; /* see ndm_flags */
117 u8 eth_addr[ETH_ALEN];
118 };
119
120 /* Pseudo network device */
121 struct vxlan_dev {
122 struct hlist_node hlist; /* vni hash table */
123 struct list_head next; /* vxlan's per namespace list */
124 struct vxlan_sock *vn_sock; /* listening socket */
125 struct net_device *dev;
126 struct vxlan_rdst default_dst; /* default destination */
127 __be32 saddr; /* source address */
128 __be16 dst_port;
129 __u16 port_min; /* source port range */
130 __u16 port_max;
131 __u8 tos; /* TOS override */
132 __u8 ttl;
133 u32 flags; /* VXLAN_F_* below */
134
135 struct work_struct sock_work;
136 struct work_struct igmp_work;
137
138 unsigned long age_interval;
139 struct timer_list age_timer;
140 spinlock_t hash_lock;
141 unsigned int addrcnt;
142 unsigned int addrmax;
143
144 struct hlist_head fdb_head[FDB_HASH_SIZE];
145 };
146
147 #define VXLAN_F_LEARN 0x01
148 #define VXLAN_F_PROXY 0x02
149 #define VXLAN_F_RSC 0x04
150 #define VXLAN_F_L2MISS 0x08
151 #define VXLAN_F_L3MISS 0x10
152
153 /* salt for hash table */
154 static u32 vxlan_salt __read_mostly;
155 static struct workqueue_struct *vxlan_wq;
156
157 static void vxlan_sock_work(struct work_struct *work);
158
159 /* Virtual Network hash table head */
160 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
161 {
162 return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
163 }
164
165 /* Socket hash table head */
166 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
167 {
168 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
169
170 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
171 }
172
173 /* First remote destination for a forwarding entry.
174 * Guaranteed to be non-NULL because remotes are never deleted.
175 */
176 static inline struct vxlan_rdst *first_remote(struct vxlan_fdb *fdb)
177 {
178 return list_first_or_null_rcu(&fdb->remotes, struct vxlan_rdst, list);
179 }
180
181 /* Find VXLAN socket based on network namespace and UDP port */
182 static struct vxlan_sock *vxlan_find_port(struct net *net, __be16 port)
183 {
184 struct vxlan_sock *vs;
185
186 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
187 if (inet_sk(vs->sock->sk)->inet_sport == port)
188 return vs;
189 }
190 return NULL;
191 }
192
193 /* Look up VNI in a per net namespace table */
194 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id, __be16 port)
195 {
196 struct vxlan_sock *vs;
197 struct vxlan_dev *vxlan;
198
199 vs = vxlan_find_port(net, port);
200 if (!vs)
201 return NULL;
202
203 hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
204 if (vxlan->default_dst.remote_vni == id)
205 return vxlan;
206 }
207
208 return NULL;
209 }
210
211 /* Fill in neighbour message in skbuff. */
212 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
213 const struct vxlan_fdb *fdb,
214 u32 portid, u32 seq, int type, unsigned int flags,
215 const struct vxlan_rdst *rdst)
216 {
217 unsigned long now = jiffies;
218 struct nda_cacheinfo ci;
219 struct nlmsghdr *nlh;
220 struct ndmsg *ndm;
221 bool send_ip, send_eth;
222
223 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
224 if (nlh == NULL)
225 return -EMSGSIZE;
226
227 ndm = nlmsg_data(nlh);
228 memset(ndm, 0, sizeof(*ndm));
229
230 send_eth = send_ip = true;
231
232 if (type == RTM_GETNEIGH) {
233 ndm->ndm_family = AF_INET;
234 send_ip = rdst->remote_ip != htonl(INADDR_ANY);
235 send_eth = !is_zero_ether_addr(fdb->eth_addr);
236 } else
237 ndm->ndm_family = AF_BRIDGE;
238 ndm->ndm_state = fdb->state;
239 ndm->ndm_ifindex = vxlan->dev->ifindex;
240 ndm->ndm_flags = fdb->flags;
241 ndm->ndm_type = NDA_DST;
242
243 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
244 goto nla_put_failure;
245
246 if (send_ip && nla_put_be32(skb, NDA_DST, rdst->remote_ip))
247 goto nla_put_failure;
248
249 if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
250 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
251 goto nla_put_failure;
252 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
253 nla_put_be32(skb, NDA_VNI, rdst->remote_vni))
254 goto nla_put_failure;
255 if (rdst->remote_ifindex &&
256 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
257 goto nla_put_failure;
258
259 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
260 ci.ndm_confirmed = 0;
261 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
262 ci.ndm_refcnt = 0;
263
264 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
265 goto nla_put_failure;
266
267 return nlmsg_end(skb, nlh);
268
269 nla_put_failure:
270 nlmsg_cancel(skb, nlh);
271 return -EMSGSIZE;
272 }
273
274 static inline size_t vxlan_nlmsg_size(void)
275 {
276 return NLMSG_ALIGN(sizeof(struct ndmsg))
277 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
278 + nla_total_size(sizeof(__be32)) /* NDA_DST */
279 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
280 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
281 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
282 + nla_total_size(sizeof(struct nda_cacheinfo));
283 }
284
285 static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
286 struct vxlan_fdb *fdb, int type)
287 {
288 struct net *net = dev_net(vxlan->dev);
289 struct sk_buff *skb;
290 int err = -ENOBUFS;
291
292 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
293 if (skb == NULL)
294 goto errout;
295
296 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, first_remote(fdb));
297 if (err < 0) {
298 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
299 WARN_ON(err == -EMSGSIZE);
300 kfree_skb(skb);
301 goto errout;
302 }
303
304 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
305 return;
306 errout:
307 if (err < 0)
308 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
309 }
310
311 static void vxlan_ip_miss(struct net_device *dev, __be32 ipa)
312 {
313 struct vxlan_dev *vxlan = netdev_priv(dev);
314 struct vxlan_fdb f = {
315 .state = NUD_STALE,
316 };
317 struct vxlan_rdst remote = {
318 .remote_ip = ipa, /* goes to NDA_DST */
319 .remote_vni = VXLAN_N_VID,
320 };
321
322 INIT_LIST_HEAD(&f.remotes);
323 list_add_rcu(&remote.list, &f.remotes);
324
325 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
326 }
327
328 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
329 {
330 struct vxlan_fdb f = {
331 .state = NUD_STALE,
332 };
333
334 INIT_LIST_HEAD(&f.remotes);
335 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
336
337 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
338 }
339
340 /* Hash Ethernet address */
341 static u32 eth_hash(const unsigned char *addr)
342 {
343 u64 value = get_unaligned((u64 *)addr);
344
345 /* only want 6 bytes */
346 #ifdef __BIG_ENDIAN
347 value >>= 16;
348 #else
349 value <<= 16;
350 #endif
351 return hash_64(value, FDB_HASH_BITS);
352 }
353
354 /* Hash chain to use given mac address */
355 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
356 const u8 *mac)
357 {
358 return &vxlan->fdb_head[eth_hash(mac)];
359 }
360
361 /* Look up Ethernet address in forwarding table */
362 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
363 const u8 *mac)
364
365 {
366 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
367 struct vxlan_fdb *f;
368
369 hlist_for_each_entry_rcu(f, head, hlist) {
370 if (compare_ether_addr(mac, f->eth_addr) == 0)
371 return f;
372 }
373
374 return NULL;
375 }
376
377 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
378 const u8 *mac)
379 {
380 struct vxlan_fdb *f;
381
382 f = __vxlan_find_mac(vxlan, mac);
383 if (f)
384 f->used = jiffies;
385
386 return f;
387 }
388
389 /* Add/update destinations for multicast */
390 static int vxlan_fdb_append(struct vxlan_fdb *f,
391 __be32 ip, __be16 port, __u32 vni, __u32 ifindex)
392 {
393 struct vxlan_rdst *rd;
394
395 /* protected by vxlan->hash_lock */
396 list_for_each_entry(rd, &f->remotes, list) {
397 if (rd->remote_ip == ip &&
398 rd->remote_port == port &&
399 rd->remote_vni == vni &&
400 rd->remote_ifindex == ifindex)
401 return 0;
402 }
403
404 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
405 if (rd == NULL)
406 return -ENOBUFS;
407 rd->remote_ip = ip;
408 rd->remote_port = port;
409 rd->remote_vni = vni;
410 rd->remote_ifindex = ifindex;
411
412 list_add_tail_rcu(&rd->list, &f->remotes);
413
414 return 1;
415 }
416
417 /* Add new entry to forwarding table -- assumes lock held */
418 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
419 const u8 *mac, __be32 ip,
420 __u16 state, __u16 flags,
421 __be16 port, __u32 vni, __u32 ifindex,
422 __u8 ndm_flags)
423 {
424 struct vxlan_fdb *f;
425 int notify = 0;
426
427 f = __vxlan_find_mac(vxlan, mac);
428 if (f) {
429 if (flags & NLM_F_EXCL) {
430 netdev_dbg(vxlan->dev,
431 "lost race to create %pM\n", mac);
432 return -EEXIST;
433 }
434 if (f->state != state) {
435 f->state = state;
436 f->updated = jiffies;
437 notify = 1;
438 }
439 if (f->flags != ndm_flags) {
440 f->flags = ndm_flags;
441 f->updated = jiffies;
442 notify = 1;
443 }
444 if ((flags & NLM_F_APPEND) &&
445 is_multicast_ether_addr(f->eth_addr)) {
446 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex);
447
448 if (rc < 0)
449 return rc;
450 notify |= rc;
451 }
452 } else {
453 if (!(flags & NLM_F_CREATE))
454 return -ENOENT;
455
456 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
457 return -ENOSPC;
458
459 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
460 f = kmalloc(sizeof(*f), GFP_ATOMIC);
461 if (!f)
462 return -ENOMEM;
463
464 notify = 1;
465 f->state = state;
466 f->flags = ndm_flags;
467 f->updated = f->used = jiffies;
468 INIT_LIST_HEAD(&f->remotes);
469 memcpy(f->eth_addr, mac, ETH_ALEN);
470
471 vxlan_fdb_append(f, ip, port, vni, ifindex);
472
473 ++vxlan->addrcnt;
474 hlist_add_head_rcu(&f->hlist,
475 vxlan_fdb_head(vxlan, mac));
476 }
477
478 if (notify)
479 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
480
481 return 0;
482 }
483
484 static void vxlan_fdb_free(struct rcu_head *head)
485 {
486 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
487 struct vxlan_rdst *rd, *nd;
488
489 list_for_each_entry_safe(rd, nd, &f->remotes, list)
490 kfree(rd);
491 kfree(f);
492 }
493
494 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
495 {
496 netdev_dbg(vxlan->dev,
497 "delete %pM\n", f->eth_addr);
498
499 --vxlan->addrcnt;
500 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
501
502 hlist_del_rcu(&f->hlist);
503 call_rcu(&f->rcu, vxlan_fdb_free);
504 }
505
506 /* Add static entry (via netlink) */
507 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
508 struct net_device *dev,
509 const unsigned char *addr, u16 flags)
510 {
511 struct vxlan_dev *vxlan = netdev_priv(dev);
512 struct net *net = dev_net(vxlan->dev);
513 __be32 ip;
514 __be16 port;
515 u32 vni, ifindex;
516 int err;
517
518 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
519 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
520 ndm->ndm_state);
521 return -EINVAL;
522 }
523
524 if (tb[NDA_DST] == NULL)
525 return -EINVAL;
526
527 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
528 return -EAFNOSUPPORT;
529
530 ip = nla_get_be32(tb[NDA_DST]);
531
532 if (tb[NDA_PORT]) {
533 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
534 return -EINVAL;
535 port = nla_get_be16(tb[NDA_PORT]);
536 } else
537 port = vxlan->dst_port;
538
539 if (tb[NDA_VNI]) {
540 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
541 return -EINVAL;
542 vni = nla_get_u32(tb[NDA_VNI]);
543 } else
544 vni = vxlan->default_dst.remote_vni;
545
546 if (tb[NDA_IFINDEX]) {
547 struct net_device *tdev;
548
549 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
550 return -EINVAL;
551 ifindex = nla_get_u32(tb[NDA_IFINDEX]);
552 tdev = dev_get_by_index(net, ifindex);
553 if (!tdev)
554 return -EADDRNOTAVAIL;
555 dev_put(tdev);
556 } else
557 ifindex = 0;
558
559 spin_lock_bh(&vxlan->hash_lock);
560 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags,
561 port, vni, ifindex, ndm->ndm_flags);
562 spin_unlock_bh(&vxlan->hash_lock);
563
564 return err;
565 }
566
567 /* Delete entry (via netlink) */
568 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
569 struct net_device *dev,
570 const unsigned char *addr)
571 {
572 struct vxlan_dev *vxlan = netdev_priv(dev);
573 struct vxlan_fdb *f;
574 int err = -ENOENT;
575
576 spin_lock_bh(&vxlan->hash_lock);
577 f = vxlan_find_mac(vxlan, addr);
578 if (f) {
579 vxlan_fdb_destroy(vxlan, f);
580 err = 0;
581 }
582 spin_unlock_bh(&vxlan->hash_lock);
583
584 return err;
585 }
586
587 /* Dump forwarding table */
588 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
589 struct net_device *dev, int idx)
590 {
591 struct vxlan_dev *vxlan = netdev_priv(dev);
592 unsigned int h;
593
594 for (h = 0; h < FDB_HASH_SIZE; ++h) {
595 struct vxlan_fdb *f;
596 int err;
597
598 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
599 struct vxlan_rdst *rd;
600
601 if (idx < cb->args[0])
602 goto skip;
603
604 list_for_each_entry_rcu(rd, &f->remotes, list) {
605 err = vxlan_fdb_info(skb, vxlan, f,
606 NETLINK_CB(cb->skb).portid,
607 cb->nlh->nlmsg_seq,
608 RTM_NEWNEIGH,
609 NLM_F_MULTI, rd);
610 if (err < 0)
611 goto out;
612 }
613 skip:
614 ++idx;
615 }
616 }
617 out:
618 return idx;
619 }
620
621 /* Watch incoming packets to learn mapping between Ethernet address
622 * and Tunnel endpoint.
623 * Return true if packet is bogus and should be droppped.
624 */
625 static bool vxlan_snoop(struct net_device *dev,
626 __be32 src_ip, const u8 *src_mac)
627 {
628 struct vxlan_dev *vxlan = netdev_priv(dev);
629 struct vxlan_fdb *f;
630
631 f = vxlan_find_mac(vxlan, src_mac);
632 if (likely(f)) {
633 struct vxlan_rdst *rdst = first_remote(f);
634
635 if (likely(rdst->remote_ip == src_ip))
636 return false;
637
638 /* Don't migrate static entries, drop packets */
639 if (f->state & NUD_NOARP)
640 return true;
641
642 if (net_ratelimit())
643 netdev_info(dev,
644 "%pM migrated from %pI4 to %pI4\n",
645 src_mac, &rdst->remote_ip, &src_ip);
646
647 rdst->remote_ip = src_ip;
648 f->updated = jiffies;
649 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
650 } else {
651 /* learned new entry */
652 spin_lock(&vxlan->hash_lock);
653
654 /* close off race between vxlan_flush and incoming packets */
655 if (netif_running(dev))
656 vxlan_fdb_create(vxlan, src_mac, src_ip,
657 NUD_REACHABLE,
658 NLM_F_EXCL|NLM_F_CREATE,
659 vxlan->dst_port,
660 vxlan->default_dst.remote_vni,
661 0, NTF_SELF);
662 spin_unlock(&vxlan->hash_lock);
663 }
664
665 return false;
666 }
667
668
669 /* See if multicast group is already in use by other ID */
670 static bool vxlan_group_used(struct vxlan_net *vn, __be32 remote_ip)
671 {
672 struct vxlan_dev *vxlan;
673
674 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
675 if (!netif_running(vxlan->dev))
676 continue;
677
678 if (vxlan->default_dst.remote_ip == remote_ip)
679 return true;
680 }
681
682 return false;
683 }
684
685 static void vxlan_sock_hold(struct vxlan_sock *vs)
686 {
687 atomic_inc(&vs->refcnt);
688 }
689
690 static void vxlan_sock_release(struct vxlan_net *vn, struct vxlan_sock *vs)
691 {
692 if (!atomic_dec_and_test(&vs->refcnt))
693 return;
694
695 spin_lock(&vn->sock_lock);
696 hlist_del_rcu(&vs->hlist);
697 spin_unlock(&vn->sock_lock);
698
699 queue_work(vxlan_wq, &vs->del_work);
700 }
701
702 /* Callback to update multicast group membership.
703 * Scheduled when vxlan goes up/down.
704 */
705 static void vxlan_igmp_work(struct work_struct *work)
706 {
707 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_work);
708 struct vxlan_net *vn = net_generic(dev_net(vxlan->dev), vxlan_net_id);
709 struct vxlan_sock *vs = vxlan->vn_sock;
710 struct sock *sk = vs->sock->sk;
711 struct ip_mreqn mreq = {
712 .imr_multiaddr.s_addr = vxlan->default_dst.remote_ip,
713 .imr_ifindex = vxlan->default_dst.remote_ifindex,
714 };
715
716 lock_sock(sk);
717 if (vxlan_group_used(vn, vxlan->default_dst.remote_ip))
718 ip_mc_join_group(sk, &mreq);
719 else
720 ip_mc_leave_group(sk, &mreq);
721 release_sock(sk);
722
723 vxlan_sock_release(vn, vs);
724 dev_put(vxlan->dev);
725 }
726
727 /* Callback from net/ipv4/udp.c to receive packets */
728 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
729 {
730 struct iphdr *oip;
731 struct vxlanhdr *vxh;
732 struct vxlan_dev *vxlan;
733 struct pcpu_tstats *stats;
734 __be16 port;
735 __u32 vni;
736 int err;
737
738 /* pop off outer UDP header */
739 __skb_pull(skb, sizeof(struct udphdr));
740
741 /* Need Vxlan and inner Ethernet header to be present */
742 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
743 goto error;
744
745 /* Drop packets with reserved bits set */
746 vxh = (struct vxlanhdr *) skb->data;
747 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
748 (vxh->vx_vni & htonl(0xff))) {
749 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
750 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
751 goto error;
752 }
753
754 __skb_pull(skb, sizeof(struct vxlanhdr));
755
756 /* Is this VNI defined? */
757 vni = ntohl(vxh->vx_vni) >> 8;
758 port = inet_sk(sk)->inet_sport;
759 vxlan = vxlan_find_vni(sock_net(sk), vni, port);
760 if (!vxlan) {
761 netdev_dbg(skb->dev, "unknown vni %d port %u\n",
762 vni, ntohs(port));
763 goto drop;
764 }
765
766 if (!pskb_may_pull(skb, ETH_HLEN)) {
767 vxlan->dev->stats.rx_length_errors++;
768 vxlan->dev->stats.rx_errors++;
769 goto drop;
770 }
771
772 skb_reset_mac_header(skb);
773
774 /* Re-examine inner Ethernet packet */
775 oip = ip_hdr(skb);
776 skb->protocol = eth_type_trans(skb, vxlan->dev);
777
778 /* Ignore packet loops (and multicast echo) */
779 if (compare_ether_addr(eth_hdr(skb)->h_source,
780 vxlan->dev->dev_addr) == 0)
781 goto drop;
782
783 if ((vxlan->flags & VXLAN_F_LEARN) &&
784 vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source))
785 goto drop;
786
787 __skb_tunnel_rx(skb, vxlan->dev);
788 skb_reset_network_header(skb);
789
790 /* If the NIC driver gave us an encapsulated packet with
791 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
792 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
793 * for us. Otherwise force the upper layers to verify it.
794 */
795 if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
796 !(vxlan->dev->features & NETIF_F_RXCSUM))
797 skb->ip_summed = CHECKSUM_NONE;
798
799 skb->encapsulation = 0;
800
801 err = IP_ECN_decapsulate(oip, skb);
802 if (unlikely(err)) {
803 if (log_ecn_error)
804 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
805 &oip->saddr, oip->tos);
806 if (err > 1) {
807 ++vxlan->dev->stats.rx_frame_errors;
808 ++vxlan->dev->stats.rx_errors;
809 goto drop;
810 }
811 }
812
813 stats = this_cpu_ptr(vxlan->dev->tstats);
814 u64_stats_update_begin(&stats->syncp);
815 stats->rx_packets++;
816 stats->rx_bytes += skb->len;
817 u64_stats_update_end(&stats->syncp);
818
819 netif_rx(skb);
820
821 return 0;
822 error:
823 /* Put UDP header back */
824 __skb_push(skb, sizeof(struct udphdr));
825
826 return 1;
827 drop:
828 /* Consume bad packet */
829 kfree_skb(skb);
830 return 0;
831 }
832
833 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
834 {
835 struct vxlan_dev *vxlan = netdev_priv(dev);
836 struct arphdr *parp;
837 u8 *arpptr, *sha;
838 __be32 sip, tip;
839 struct neighbour *n;
840
841 if (dev->flags & IFF_NOARP)
842 goto out;
843
844 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
845 dev->stats.tx_dropped++;
846 goto out;
847 }
848 parp = arp_hdr(skb);
849
850 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
851 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
852 parp->ar_pro != htons(ETH_P_IP) ||
853 parp->ar_op != htons(ARPOP_REQUEST) ||
854 parp->ar_hln != dev->addr_len ||
855 parp->ar_pln != 4)
856 goto out;
857 arpptr = (u8 *)parp + sizeof(struct arphdr);
858 sha = arpptr;
859 arpptr += dev->addr_len; /* sha */
860 memcpy(&sip, arpptr, sizeof(sip));
861 arpptr += sizeof(sip);
862 arpptr += dev->addr_len; /* tha */
863 memcpy(&tip, arpptr, sizeof(tip));
864
865 if (ipv4_is_loopback(tip) ||
866 ipv4_is_multicast(tip))
867 goto out;
868
869 n = neigh_lookup(&arp_tbl, &tip, dev);
870
871 if (n) {
872 struct vxlan_fdb *f;
873 struct sk_buff *reply;
874
875 if (!(n->nud_state & NUD_CONNECTED)) {
876 neigh_release(n);
877 goto out;
878 }
879
880 f = vxlan_find_mac(vxlan, n->ha);
881 if (f && first_remote(f)->remote_ip == htonl(INADDR_ANY)) {
882 /* bridge-local neighbor */
883 neigh_release(n);
884 goto out;
885 }
886
887 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
888 n->ha, sha);
889
890 neigh_release(n);
891
892 skb_reset_mac_header(reply);
893 __skb_pull(reply, skb_network_offset(reply));
894 reply->ip_summed = CHECKSUM_UNNECESSARY;
895 reply->pkt_type = PACKET_HOST;
896
897 if (netif_rx_ni(reply) == NET_RX_DROP)
898 dev->stats.rx_dropped++;
899 } else if (vxlan->flags & VXLAN_F_L3MISS)
900 vxlan_ip_miss(dev, tip);
901 out:
902 consume_skb(skb);
903 return NETDEV_TX_OK;
904 }
905
906 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
907 {
908 struct vxlan_dev *vxlan = netdev_priv(dev);
909 struct neighbour *n;
910 struct iphdr *pip;
911
912 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
913 return false;
914
915 n = NULL;
916 switch (ntohs(eth_hdr(skb)->h_proto)) {
917 case ETH_P_IP:
918 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
919 return false;
920 pip = ip_hdr(skb);
921 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
922 break;
923 default:
924 return false;
925 }
926
927 if (n) {
928 bool diff;
929
930 diff = compare_ether_addr(eth_hdr(skb)->h_dest, n->ha) != 0;
931 if (diff) {
932 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
933 dev->addr_len);
934 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
935 }
936 neigh_release(n);
937 return diff;
938 } else if (vxlan->flags & VXLAN_F_L3MISS)
939 vxlan_ip_miss(dev, pip->daddr);
940 return false;
941 }
942
943 static void vxlan_sock_put(struct sk_buff *skb)
944 {
945 sock_put(skb->sk);
946 }
947
948 /* On transmit, associate with the tunnel socket */
949 static void vxlan_set_owner(struct net_device *dev, struct sk_buff *skb)
950 {
951 struct vxlan_dev *vxlan = netdev_priv(dev);
952 struct sock *sk = vxlan->vn_sock->sock->sk;
953
954 skb_orphan(skb);
955 sock_hold(sk);
956 skb->sk = sk;
957 skb->destructor = vxlan_sock_put;
958 }
959
960 /* Compute source port for outgoing packet
961 * first choice to use L4 flow hash since it will spread
962 * better and maybe available from hardware
963 * secondary choice is to use jhash on the Ethernet header
964 */
965 static __be16 vxlan_src_port(const struct vxlan_dev *vxlan, struct sk_buff *skb)
966 {
967 unsigned int range = (vxlan->port_max - vxlan->port_min) + 1;
968 u32 hash;
969
970 hash = skb_get_rxhash(skb);
971 if (!hash)
972 hash = jhash(skb->data, 2 * ETH_ALEN,
973 (__force u32) skb->protocol);
974
975 return htons((((u64) hash * range) >> 32) + vxlan->port_min);
976 }
977
978 static int handle_offloads(struct sk_buff *skb)
979 {
980 if (skb_is_gso(skb)) {
981 int err = skb_unclone(skb, GFP_ATOMIC);
982 if (unlikely(err))
983 return err;
984
985 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
986 } else if (skb->ip_summed != CHECKSUM_PARTIAL)
987 skb->ip_summed = CHECKSUM_NONE;
988
989 return 0;
990 }
991
992 /* Bypass encapsulation if the destination is local */
993 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
994 struct vxlan_dev *dst_vxlan)
995 {
996 struct pcpu_tstats *tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
997 struct pcpu_tstats *rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
998
999 skb->pkt_type = PACKET_HOST;
1000 skb->encapsulation = 0;
1001 skb->dev = dst_vxlan->dev;
1002 __skb_pull(skb, skb_network_offset(skb));
1003
1004 if (dst_vxlan->flags & VXLAN_F_LEARN)
1005 vxlan_snoop(skb->dev, htonl(INADDR_LOOPBACK),
1006 eth_hdr(skb)->h_source);
1007
1008 u64_stats_update_begin(&tx_stats->syncp);
1009 tx_stats->tx_packets++;
1010 tx_stats->tx_bytes += skb->len;
1011 u64_stats_update_end(&tx_stats->syncp);
1012
1013 if (netif_rx(skb) == NET_RX_SUCCESS) {
1014 u64_stats_update_begin(&rx_stats->syncp);
1015 rx_stats->rx_packets++;
1016 rx_stats->rx_bytes += skb->len;
1017 u64_stats_update_end(&rx_stats->syncp);
1018 } else {
1019 skb->dev->stats.rx_dropped++;
1020 }
1021 }
1022
1023 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1024 struct vxlan_rdst *rdst, bool did_rsc)
1025 {
1026 struct vxlan_dev *vxlan = netdev_priv(dev);
1027 struct rtable *rt;
1028 const struct iphdr *old_iph;
1029 struct vxlanhdr *vxh;
1030 struct udphdr *uh;
1031 struct flowi4 fl4;
1032 __be32 dst;
1033 __be16 src_port, dst_port;
1034 u32 vni;
1035 __be16 df = 0;
1036 __u8 tos, ttl;
1037 int err;
1038
1039 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1040 vni = rdst->remote_vni;
1041 dst = rdst->remote_ip;
1042
1043 if (!dst) {
1044 if (did_rsc) {
1045 /* short-circuited back to local bridge */
1046 vxlan_encap_bypass(skb, vxlan, vxlan);
1047 return;
1048 }
1049 goto drop;
1050 }
1051
1052 if (!skb->encapsulation) {
1053 skb_reset_inner_headers(skb);
1054 skb->encapsulation = 1;
1055 }
1056
1057 /* Need space for new headers (invalidates iph ptr) */
1058 if (skb_cow_head(skb, VXLAN_HEADROOM))
1059 goto drop;
1060
1061 old_iph = ip_hdr(skb);
1062
1063 ttl = vxlan->ttl;
1064 if (!ttl && IN_MULTICAST(ntohl(dst)))
1065 ttl = 1;
1066
1067 tos = vxlan->tos;
1068 if (tos == 1)
1069 tos = ip_tunnel_get_dsfield(old_iph, skb);
1070
1071 src_port = vxlan_src_port(vxlan, skb);
1072
1073 memset(&fl4, 0, sizeof(fl4));
1074 fl4.flowi4_oif = rdst->remote_ifindex;
1075 fl4.flowi4_tos = RT_TOS(tos);
1076 fl4.daddr = dst;
1077 fl4.saddr = vxlan->saddr;
1078
1079 rt = ip_route_output_key(dev_net(dev), &fl4);
1080 if (IS_ERR(rt)) {
1081 netdev_dbg(dev, "no route to %pI4\n", &dst);
1082 dev->stats.tx_carrier_errors++;
1083 goto tx_error;
1084 }
1085
1086 if (rt->dst.dev == dev) {
1087 netdev_dbg(dev, "circular route to %pI4\n", &dst);
1088 ip_rt_put(rt);
1089 dev->stats.collisions++;
1090 goto tx_error;
1091 }
1092
1093 /* Bypass encapsulation if the destination is local */
1094 if (rt->rt_flags & RTCF_LOCAL &&
1095 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1096 struct vxlan_dev *dst_vxlan;
1097
1098 ip_rt_put(rt);
1099 dst_vxlan = vxlan_find_vni(dev_net(dev), vni, dst_port);
1100 if (!dst_vxlan)
1101 goto tx_error;
1102 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1103 return;
1104 }
1105 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1106 vxh->vx_flags = htonl(VXLAN_FLAGS);
1107 vxh->vx_vni = htonl(vni << 8);
1108
1109 __skb_push(skb, sizeof(*uh));
1110 skb_reset_transport_header(skb);
1111 uh = udp_hdr(skb);
1112
1113 uh->dest = dst_port;
1114 uh->source = src_port;
1115
1116 uh->len = htons(skb->len);
1117 uh->check = 0;
1118
1119 vxlan_set_owner(dev, skb);
1120
1121 if (handle_offloads(skb))
1122 goto drop;
1123
1124 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1125 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1126
1127 err = iptunnel_xmit(dev_net(dev), rt, skb, fl4.saddr, dst,
1128 IPPROTO_UDP, tos, ttl, df);
1129 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1130
1131 return;
1132
1133 drop:
1134 dev->stats.tx_dropped++;
1135 goto tx_free;
1136
1137 tx_error:
1138 dev->stats.tx_errors++;
1139 tx_free:
1140 dev_kfree_skb(skb);
1141 }
1142
1143 /* Transmit local packets over Vxlan
1144 *
1145 * Outer IP header inherits ECN and DF from inner header.
1146 * Outer UDP destination is the VXLAN assigned port.
1147 * source port is based on hash of flow
1148 */
1149 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1150 {
1151 struct vxlan_dev *vxlan = netdev_priv(dev);
1152 struct ethhdr *eth;
1153 bool did_rsc = false;
1154 struct vxlan_rdst *rdst0, *rdst;
1155 struct vxlan_fdb *f;
1156
1157 skb_reset_mac_header(skb);
1158 eth = eth_hdr(skb);
1159
1160 if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP)
1161 return arp_reduce(dev, skb);
1162
1163 f = vxlan_find_mac(vxlan, eth->h_dest);
1164 did_rsc = false;
1165
1166 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1167 ntohs(eth->h_proto) == ETH_P_IP) {
1168 did_rsc = route_shortcircuit(dev, skb);
1169 if (did_rsc)
1170 f = vxlan_find_mac(vxlan, eth->h_dest);
1171 }
1172
1173 if (f == NULL) {
1174 rdst0 = &vxlan->default_dst;
1175
1176 if (rdst0->remote_ip == htonl(INADDR_ANY) &&
1177 (vxlan->flags & VXLAN_F_L2MISS) &&
1178 !is_multicast_ether_addr(eth->h_dest))
1179 vxlan_fdb_miss(vxlan, eth->h_dest);
1180 } else {
1181 rdst = rdst0 = first_remote(f);
1182
1183 /* if there are multiple destinations, send copies */
1184 list_for_each_entry_continue_rcu(rdst, &f->remotes, list) {
1185 struct sk_buff *skb1;
1186
1187 skb1 = skb_clone(skb, GFP_ATOMIC);
1188 if (skb1)
1189 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1190 }
1191 }
1192
1193 vxlan_xmit_one(skb, dev, rdst0, did_rsc);
1194 return NETDEV_TX_OK;
1195 }
1196
1197 /* Walk the forwarding table and purge stale entries */
1198 static void vxlan_cleanup(unsigned long arg)
1199 {
1200 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1201 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1202 unsigned int h;
1203
1204 if (!netif_running(vxlan->dev))
1205 return;
1206
1207 spin_lock_bh(&vxlan->hash_lock);
1208 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1209 struct hlist_node *p, *n;
1210 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1211 struct vxlan_fdb *f
1212 = container_of(p, struct vxlan_fdb, hlist);
1213 unsigned long timeout;
1214
1215 if (f->state & NUD_PERMANENT)
1216 continue;
1217
1218 timeout = f->used + vxlan->age_interval * HZ;
1219 if (time_before_eq(timeout, jiffies)) {
1220 netdev_dbg(vxlan->dev,
1221 "garbage collect %pM\n",
1222 f->eth_addr);
1223 f->state = NUD_STALE;
1224 vxlan_fdb_destroy(vxlan, f);
1225 } else if (time_before(timeout, next_timer))
1226 next_timer = timeout;
1227 }
1228 }
1229 spin_unlock_bh(&vxlan->hash_lock);
1230
1231 mod_timer(&vxlan->age_timer, next_timer);
1232 }
1233
1234 /* Setup stats when device is created */
1235 static int vxlan_init(struct net_device *dev)
1236 {
1237 struct vxlan_dev *vxlan = netdev_priv(dev);
1238 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
1239 struct vxlan_sock *vs;
1240 __u32 vni = vxlan->default_dst.remote_vni;
1241
1242 dev->tstats = alloc_percpu(struct pcpu_tstats);
1243 if (!dev->tstats)
1244 return -ENOMEM;
1245
1246 spin_lock(&vn->sock_lock);
1247 vs = vxlan_find_port(dev_net(dev), vxlan->dst_port);
1248 if (vs) {
1249 /* If we have a socket with same port already, reuse it */
1250 atomic_inc(&vs->refcnt);
1251 vxlan->vn_sock = vs;
1252 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
1253 } else {
1254 /* otherwise make new socket outside of RTNL */
1255 dev_hold(dev);
1256 queue_work(vxlan_wq, &vxlan->sock_work);
1257 }
1258 spin_unlock(&vn->sock_lock);
1259
1260 return 0;
1261 }
1262
1263 static void vxlan_uninit(struct net_device *dev)
1264 {
1265 struct vxlan_dev *vxlan = netdev_priv(dev);
1266 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
1267 struct vxlan_sock *vs = vxlan->vn_sock;
1268
1269 if (vs)
1270 vxlan_sock_release(vn, vs);
1271 free_percpu(dev->tstats);
1272 }
1273
1274 /* Start ageing timer and join group when device is brought up */
1275 static int vxlan_open(struct net_device *dev)
1276 {
1277 struct vxlan_dev *vxlan = netdev_priv(dev);
1278 struct vxlan_sock *vs = vxlan->vn_sock;
1279
1280 /* socket hasn't been created */
1281 if (!vs)
1282 return -ENOTCONN;
1283
1284 if (IN_MULTICAST(ntohl(vxlan->default_dst.remote_ip))) {
1285 vxlan_sock_hold(vs);
1286 dev_hold(dev);
1287 queue_work(vxlan_wq, &vxlan->igmp_work);
1288 }
1289
1290 if (vxlan->age_interval)
1291 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
1292
1293 return 0;
1294 }
1295
1296 /* Purge the forwarding table */
1297 static void vxlan_flush(struct vxlan_dev *vxlan)
1298 {
1299 unsigned int h;
1300
1301 spin_lock_bh(&vxlan->hash_lock);
1302 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1303 struct hlist_node *p, *n;
1304 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1305 struct vxlan_fdb *f
1306 = container_of(p, struct vxlan_fdb, hlist);
1307 vxlan_fdb_destroy(vxlan, f);
1308 }
1309 }
1310 spin_unlock_bh(&vxlan->hash_lock);
1311 }
1312
1313 /* Cleanup timer and forwarding table on shutdown */
1314 static int vxlan_stop(struct net_device *dev)
1315 {
1316 struct vxlan_dev *vxlan = netdev_priv(dev);
1317 struct vxlan_sock *vs = vxlan->vn_sock;
1318
1319 if (vs && IN_MULTICAST(ntohl(vxlan->default_dst.remote_ip))) {
1320 vxlan_sock_hold(vs);
1321 dev_hold(dev);
1322 queue_work(vxlan_wq, &vxlan->igmp_work);
1323 }
1324
1325 del_timer_sync(&vxlan->age_timer);
1326
1327 vxlan_flush(vxlan);
1328
1329 return 0;
1330 }
1331
1332 /* Stub, nothing needs to be done. */
1333 static void vxlan_set_multicast_list(struct net_device *dev)
1334 {
1335 }
1336
1337 static const struct net_device_ops vxlan_netdev_ops = {
1338 .ndo_init = vxlan_init,
1339 .ndo_uninit = vxlan_uninit,
1340 .ndo_open = vxlan_open,
1341 .ndo_stop = vxlan_stop,
1342 .ndo_start_xmit = vxlan_xmit,
1343 .ndo_get_stats64 = ip_tunnel_get_stats64,
1344 .ndo_set_rx_mode = vxlan_set_multicast_list,
1345 .ndo_change_mtu = eth_change_mtu,
1346 .ndo_validate_addr = eth_validate_addr,
1347 .ndo_set_mac_address = eth_mac_addr,
1348 .ndo_fdb_add = vxlan_fdb_add,
1349 .ndo_fdb_del = vxlan_fdb_delete,
1350 .ndo_fdb_dump = vxlan_fdb_dump,
1351 };
1352
1353 /* Info for udev, that this is a virtual tunnel endpoint */
1354 static struct device_type vxlan_type = {
1355 .name = "vxlan",
1356 };
1357
1358 /* Initialize the device structure. */
1359 static void vxlan_setup(struct net_device *dev)
1360 {
1361 struct vxlan_dev *vxlan = netdev_priv(dev);
1362 unsigned int h;
1363 int low, high;
1364
1365 eth_hw_addr_random(dev);
1366 ether_setup(dev);
1367 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
1368
1369 dev->netdev_ops = &vxlan_netdev_ops;
1370 dev->destructor = free_netdev;
1371 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
1372
1373 dev->tx_queue_len = 0;
1374 dev->features |= NETIF_F_LLTX;
1375 dev->features |= NETIF_F_NETNS_LOCAL;
1376 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1377 dev->features |= NETIF_F_RXCSUM;
1378 dev->features |= NETIF_F_GSO_SOFTWARE;
1379
1380 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1381 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1382 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1383 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1384
1385 INIT_LIST_HEAD(&vxlan->next);
1386 spin_lock_init(&vxlan->hash_lock);
1387 INIT_WORK(&vxlan->igmp_work, vxlan_igmp_work);
1388 INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
1389
1390 init_timer_deferrable(&vxlan->age_timer);
1391 vxlan->age_timer.function = vxlan_cleanup;
1392 vxlan->age_timer.data = (unsigned long) vxlan;
1393
1394 inet_get_local_port_range(&low, &high);
1395 vxlan->port_min = low;
1396 vxlan->port_max = high;
1397 vxlan->dst_port = htons(vxlan_port);
1398
1399 vxlan->dev = dev;
1400
1401 for (h = 0; h < FDB_HASH_SIZE; ++h)
1402 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
1403 }
1404
1405 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
1406 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
1407 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1408 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
1409 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1410 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
1411 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
1412 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
1413 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
1414 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
1415 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
1416 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
1417 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
1418 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
1419 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
1420 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
1421 };
1422
1423 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
1424 {
1425 if (tb[IFLA_ADDRESS]) {
1426 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1427 pr_debug("invalid link address (not ethernet)\n");
1428 return -EINVAL;
1429 }
1430
1431 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1432 pr_debug("invalid all zero ethernet address\n");
1433 return -EADDRNOTAVAIL;
1434 }
1435 }
1436
1437 if (!data)
1438 return -EINVAL;
1439
1440 if (data[IFLA_VXLAN_ID]) {
1441 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
1442 if (id >= VXLAN_VID_MASK)
1443 return -ERANGE;
1444 }
1445
1446 if (data[IFLA_VXLAN_PORT_RANGE]) {
1447 const struct ifla_vxlan_port_range *p
1448 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1449
1450 if (ntohs(p->high) < ntohs(p->low)) {
1451 pr_debug("port range %u .. %u not valid\n",
1452 ntohs(p->low), ntohs(p->high));
1453 return -EINVAL;
1454 }
1455 }
1456
1457 return 0;
1458 }
1459
1460 static void vxlan_get_drvinfo(struct net_device *netdev,
1461 struct ethtool_drvinfo *drvinfo)
1462 {
1463 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
1464 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
1465 }
1466
1467 static const struct ethtool_ops vxlan_ethtool_ops = {
1468 .get_drvinfo = vxlan_get_drvinfo,
1469 .get_link = ethtool_op_get_link,
1470 };
1471
1472 static void vxlan_del_work(struct work_struct *work)
1473 {
1474 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
1475
1476 sk_release_kernel(vs->sock->sk);
1477 kfree_rcu(vs, rcu);
1478 }
1479
1480 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port)
1481 {
1482 struct vxlan_sock *vs;
1483 struct sock *sk;
1484 struct sockaddr_in vxlan_addr = {
1485 .sin_family = AF_INET,
1486 .sin_addr.s_addr = htonl(INADDR_ANY),
1487 .sin_port = port,
1488 };
1489 int rc;
1490 unsigned int h;
1491
1492 vs = kmalloc(sizeof(*vs), GFP_KERNEL);
1493 if (!vs)
1494 return ERR_PTR(-ENOMEM);
1495
1496 for (h = 0; h < VNI_HASH_SIZE; ++h)
1497 INIT_HLIST_HEAD(&vs->vni_list[h]);
1498
1499 INIT_WORK(&vs->del_work, vxlan_del_work);
1500
1501 /* Create UDP socket for encapsulation receive. */
1502 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vs->sock);
1503 if (rc < 0) {
1504 pr_debug("UDP socket create failed\n");
1505 kfree(vs);
1506 return ERR_PTR(rc);
1507 }
1508
1509 /* Put in proper namespace */
1510 sk = vs->sock->sk;
1511 sk_change_net(sk, net);
1512
1513 rc = kernel_bind(vs->sock, (struct sockaddr *) &vxlan_addr,
1514 sizeof(vxlan_addr));
1515 if (rc < 0) {
1516 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1517 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
1518 sk_release_kernel(sk);
1519 kfree(vs);
1520 return ERR_PTR(rc);
1521 }
1522
1523 /* Disable multicast loopback */
1524 inet_sk(sk)->mc_loop = 0;
1525
1526 /* Mark socket as an encapsulation socket. */
1527 udp_sk(sk)->encap_type = 1;
1528 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1529 udp_encap_enable();
1530 atomic_set(&vs->refcnt, 1);
1531
1532 return vs;
1533 }
1534
1535 /* Scheduled at device creation to bind to a socket */
1536 static void vxlan_sock_work(struct work_struct *work)
1537 {
1538 struct vxlan_dev *vxlan
1539 = container_of(work, struct vxlan_dev, sock_work);
1540 struct net_device *dev = vxlan->dev;
1541 struct net *net = dev_net(dev);
1542 __u32 vni = vxlan->default_dst.remote_vni;
1543 __be16 port = vxlan->dst_port;
1544 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1545 struct vxlan_sock *nvs, *ovs;
1546
1547 nvs = vxlan_socket_create(net, port);
1548 if (IS_ERR(nvs)) {
1549 netdev_err(vxlan->dev, "Can not create UDP socket, %ld\n",
1550 PTR_ERR(nvs));
1551 goto out;
1552 }
1553
1554 spin_lock(&vn->sock_lock);
1555 /* Look again to see if can reuse socket */
1556 ovs = vxlan_find_port(net, port);
1557 if (ovs) {
1558 atomic_inc(&ovs->refcnt);
1559 vxlan->vn_sock = ovs;
1560 hlist_add_head_rcu(&vxlan->hlist, vni_head(ovs, vni));
1561 spin_unlock(&vn->sock_lock);
1562
1563 sk_release_kernel(nvs->sock->sk);
1564 kfree(nvs);
1565 } else {
1566 vxlan->vn_sock = nvs;
1567 hlist_add_head_rcu(&nvs->hlist, vs_head(net, port));
1568 hlist_add_head_rcu(&vxlan->hlist, vni_head(nvs, vni));
1569 spin_unlock(&vn->sock_lock);
1570 }
1571 out:
1572 dev_put(dev);
1573 }
1574
1575 static int vxlan_newlink(struct net *net, struct net_device *dev,
1576 struct nlattr *tb[], struct nlattr *data[])
1577 {
1578 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1579 struct vxlan_dev *vxlan = netdev_priv(dev);
1580 struct vxlan_rdst *dst = &vxlan->default_dst;
1581 __u32 vni;
1582 int err;
1583
1584 if (!data[IFLA_VXLAN_ID])
1585 return -EINVAL;
1586
1587 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1588 dst->remote_vni = vni;
1589
1590 if (data[IFLA_VXLAN_GROUP])
1591 dst->remote_ip = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1592
1593 if (data[IFLA_VXLAN_LOCAL])
1594 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1595
1596 if (data[IFLA_VXLAN_LINK] &&
1597 (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
1598 struct net_device *lowerdev
1599 = __dev_get_by_index(net, dst->remote_ifindex);
1600
1601 if (!lowerdev) {
1602 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
1603 return -ENODEV;
1604 }
1605
1606 if (!tb[IFLA_MTU])
1607 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
1608
1609 /* update header length based on lower device */
1610 dev->hard_header_len = lowerdev->hard_header_len +
1611 VXLAN_HEADROOM;
1612 }
1613
1614 if (data[IFLA_VXLAN_TOS])
1615 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1616
1617 if (data[IFLA_VXLAN_TTL])
1618 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
1619
1620 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
1621 vxlan->flags |= VXLAN_F_LEARN;
1622
1623 if (data[IFLA_VXLAN_AGEING])
1624 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1625 else
1626 vxlan->age_interval = FDB_AGE_DEFAULT;
1627
1628 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
1629 vxlan->flags |= VXLAN_F_PROXY;
1630
1631 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
1632 vxlan->flags |= VXLAN_F_RSC;
1633
1634 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
1635 vxlan->flags |= VXLAN_F_L2MISS;
1636
1637 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
1638 vxlan->flags |= VXLAN_F_L3MISS;
1639
1640 if (data[IFLA_VXLAN_LIMIT])
1641 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1642
1643 if (data[IFLA_VXLAN_PORT_RANGE]) {
1644 const struct ifla_vxlan_port_range *p
1645 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1646 vxlan->port_min = ntohs(p->low);
1647 vxlan->port_max = ntohs(p->high);
1648 }
1649
1650 if (data[IFLA_VXLAN_PORT])
1651 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
1652
1653 if (vxlan_find_vni(net, vni, vxlan->dst_port)) {
1654 pr_info("duplicate VNI %u\n", vni);
1655 return -EEXIST;
1656 }
1657
1658 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
1659
1660 err = register_netdevice(dev);
1661 if (err)
1662 return err;
1663
1664 list_add(&vxlan->next, &vn->vxlan_list);
1665
1666 return 0;
1667 }
1668
1669 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1670 {
1671 struct vxlan_dev *vxlan = netdev_priv(dev);
1672
1673 hlist_del_rcu(&vxlan->hlist);
1674 list_del(&vxlan->next);
1675 unregister_netdevice_queue(dev, head);
1676 }
1677
1678 static size_t vxlan_get_size(const struct net_device *dev)
1679 {
1680
1681 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1682 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1683 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1684 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1685 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1686 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1687 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
1688 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
1689 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
1690 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
1691 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
1692 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1693 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
1694 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
1695 nla_total_size(sizeof(__be16))+ /* IFLA_VXLAN_PORT */
1696 0;
1697 }
1698
1699 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1700 {
1701 const struct vxlan_dev *vxlan = netdev_priv(dev);
1702 const struct vxlan_rdst *dst = &vxlan->default_dst;
1703 struct ifla_vxlan_port_range ports = {
1704 .low = htons(vxlan->port_min),
1705 .high = htons(vxlan->port_max),
1706 };
1707
1708 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
1709 goto nla_put_failure;
1710
1711 if (dst->remote_ip && nla_put_be32(skb, IFLA_VXLAN_GROUP, dst->remote_ip))
1712 goto nla_put_failure;
1713
1714 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
1715 goto nla_put_failure;
1716
1717 if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
1718 goto nla_put_failure;
1719
1720 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1721 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
1722 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
1723 !!(vxlan->flags & VXLAN_F_LEARN)) ||
1724 nla_put_u8(skb, IFLA_VXLAN_PROXY,
1725 !!(vxlan->flags & VXLAN_F_PROXY)) ||
1726 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
1727 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
1728 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
1729 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
1730 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
1731 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1732 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
1733 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port))
1734 goto nla_put_failure;
1735
1736 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
1737 goto nla_put_failure;
1738
1739 return 0;
1740
1741 nla_put_failure:
1742 return -EMSGSIZE;
1743 }
1744
1745 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1746 .kind = "vxlan",
1747 .maxtype = IFLA_VXLAN_MAX,
1748 .policy = vxlan_policy,
1749 .priv_size = sizeof(struct vxlan_dev),
1750 .setup = vxlan_setup,
1751 .validate = vxlan_validate,
1752 .newlink = vxlan_newlink,
1753 .dellink = vxlan_dellink,
1754 .get_size = vxlan_get_size,
1755 .fill_info = vxlan_fill_info,
1756 };
1757
1758 static __net_init int vxlan_init_net(struct net *net)
1759 {
1760 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1761 unsigned int h;
1762
1763 INIT_LIST_HEAD(&vn->vxlan_list);
1764 spin_lock_init(&vn->sock_lock);
1765
1766 for (h = 0; h < PORT_HASH_SIZE; ++h)
1767 INIT_HLIST_HEAD(&vn->sock_list[h]);
1768
1769 return 0;
1770 }
1771
1772 static __net_exit void vxlan_exit_net(struct net *net)
1773 {
1774 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1775 struct vxlan_dev *vxlan;
1776
1777 rtnl_lock();
1778 list_for_each_entry(vxlan, &vn->vxlan_list, next)
1779 dev_close(vxlan->dev);
1780 rtnl_unlock();
1781 }
1782
1783 static struct pernet_operations vxlan_net_ops = {
1784 .init = vxlan_init_net,
1785 .exit = vxlan_exit_net,
1786 .id = &vxlan_net_id,
1787 .size = sizeof(struct vxlan_net),
1788 };
1789
1790 static int __init vxlan_init_module(void)
1791 {
1792 int rc;
1793
1794 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
1795 if (!vxlan_wq)
1796 return -ENOMEM;
1797
1798 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1799
1800 rc = register_pernet_device(&vxlan_net_ops);
1801 if (rc)
1802 goto out1;
1803
1804 rc = rtnl_link_register(&vxlan_link_ops);
1805 if (rc)
1806 goto out2;
1807
1808 return 0;
1809
1810 out2:
1811 unregister_pernet_device(&vxlan_net_ops);
1812 out1:
1813 destroy_workqueue(vxlan_wq);
1814 return rc;
1815 }
1816 late_initcall(vxlan_init_module);
1817
1818 static void __exit vxlan_cleanup_module(void)
1819 {
1820 unregister_pernet_device(&vxlan_net_ops);
1821 rtnl_link_unregister(&vxlan_link_ops);
1822 destroy_workqueue(vxlan_wq);
1823 rcu_barrier();
1824 }
1825 module_exit(vxlan_cleanup_module);
1826
1827 MODULE_LICENSE("GPL");
1828 MODULE_VERSION(VXLAN_VERSION);
1829 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
1830 MODULE_ALIAS_RTNL_LINK("vxlan");
Linux kernel - Deliverables
This page took 0.068895 seconds and 5 git commands to generate.