]>
Commit | Line | Data |
---|---|---|
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 int vxlan_net_id; | |
82 | ||
83 | static const u8 all_zeros_mac[ETH_ALEN]; | |
84 | ||
85 | /* per UDP socket information */ | |
86 | struct vxlan_sock { | |
87 | struct hlist_node hlist; | |
88 | struct rcu_head rcu; | |
89 | struct work_struct del_work; | |
90 | atomic_t refcnt; | |
91 | struct socket *sock; | |
92 | struct hlist_head vni_list[VNI_HASH_SIZE]; | |
93 | }; | |
94 | ||
95 | /* per-network namespace private data for this module */ | |
96 | struct vxlan_net { | |
97 | struct list_head vxlan_list; | |
98 | struct hlist_head sock_list[PORT_HASH_SIZE]; | |
99 | spinlock_t sock_lock; | |
100 | }; | |
101 | ||
102 | struct vxlan_rdst { | |
103 | __be32 remote_ip; | |
104 | __be16 remote_port; | |
105 | u32 remote_vni; | |
106 | u32 remote_ifindex; | |
107 | struct list_head list; | |
108 | struct rcu_head rcu; | |
109 | }; | |
110 | ||
111 | /* Forwarding table entry */ | |
112 | struct vxlan_fdb { | |
113 | struct hlist_node hlist; /* linked list of entries */ | |
114 | struct rcu_head rcu; | |
115 | unsigned long updated; /* jiffies */ | |
116 | unsigned long used; | |
117 | struct list_head remotes; | |
118 | u16 state; /* see ndm_state */ | |
119 | u8 flags; /* see ndm_flags */ | |
120 | u8 eth_addr[ETH_ALEN]; | |
121 | }; | |
122 | ||
123 | /* Pseudo network device */ | |
124 | struct vxlan_dev { | |
125 | struct hlist_node hlist; /* vni hash table */ | |
126 | struct list_head next; /* vxlan's per namespace list */ | |
127 | struct vxlan_sock *vn_sock; /* listening socket */ | |
128 | struct net_device *dev; | |
129 | struct vxlan_rdst default_dst; /* default destination */ | |
130 | __be32 saddr; /* source address */ | |
131 | __be16 dst_port; | |
132 | __u16 port_min; /* source port range */ | |
133 | __u16 port_max; | |
134 | __u8 tos; /* TOS override */ | |
135 | __u8 ttl; | |
136 | u32 flags; /* VXLAN_F_* below */ | |
137 | ||
138 | struct work_struct sock_work; | |
139 | struct work_struct igmp_join; | |
140 | struct work_struct igmp_leave; | |
141 | ||
142 | unsigned long age_interval; | |
143 | struct timer_list age_timer; | |
144 | spinlock_t hash_lock; | |
145 | unsigned int addrcnt; | |
146 | unsigned int addrmax; | |
147 | ||
148 | struct hlist_head fdb_head[FDB_HASH_SIZE]; | |
149 | }; | |
150 | ||
151 | #define VXLAN_F_LEARN 0x01 | |
152 | #define VXLAN_F_PROXY 0x02 | |
153 | #define VXLAN_F_RSC 0x04 | |
154 | #define VXLAN_F_L2MISS 0x08 | |
155 | #define VXLAN_F_L3MISS 0x10 | |
156 | ||
157 | /* salt for hash table */ | |
158 | static u32 vxlan_salt __read_mostly; | |
159 | static struct workqueue_struct *vxlan_wq; | |
160 | ||
161 | static void vxlan_sock_work(struct work_struct *work); | |
162 | ||
163 | /* Virtual Network hash table head */ | |
164 | static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id) | |
165 | { | |
166 | return &vs->vni_list[hash_32(id, VNI_HASH_BITS)]; | |
167 | } | |
168 | ||
169 | /* Socket hash table head */ | |
170 | static inline struct hlist_head *vs_head(struct net *net, __be16 port) | |
171 | { | |
172 | struct vxlan_net *vn = net_generic(net, vxlan_net_id); | |
173 | ||
174 | return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)]; | |
175 | } | |
176 | ||
177 | /* First remote destination for a forwarding entry. | |
178 | * Guaranteed to be non-NULL because remotes are never deleted. | |
179 | */ | |
180 | static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb) | |
181 | { | |
182 | return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list); | |
183 | } | |
184 | ||
185 | static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb) | |
186 | { | |
187 | return list_first_entry(&fdb->remotes, struct vxlan_rdst, list); | |
188 | } | |
189 | ||
190 | /* Find VXLAN socket based on network namespace and UDP port */ | |
191 | static struct vxlan_sock *vxlan_find_port(struct net *net, __be16 port) | |
192 | { | |
193 | struct vxlan_sock *vs; | |
194 | ||
195 | hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) { | |
196 | if (inet_sk(vs->sock->sk)->inet_sport == port) | |
197 | return vs; | |
198 | } | |
199 | return NULL; | |
200 | } | |
201 | ||
202 | /* Look up VNI in a per net namespace table */ | |
203 | static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id, __be16 port) | |
204 | { | |
205 | struct vxlan_sock *vs; | |
206 | struct vxlan_dev *vxlan; | |
207 | ||
208 | vs = vxlan_find_port(net, port); | |
209 | if (!vs) | |
210 | return NULL; | |
211 | ||
212 | hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) { | |
213 | if (vxlan->default_dst.remote_vni == id) | |
214 | return vxlan; | |
215 | } | |
216 | ||
217 | return NULL; | |
218 | } | |
219 | ||
220 | /* Fill in neighbour message in skbuff. */ | |
221 | static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan, | |
222 | const struct vxlan_fdb *fdb, | |
223 | u32 portid, u32 seq, int type, unsigned int flags, | |
224 | const struct vxlan_rdst *rdst) | |
225 | { | |
226 | unsigned long now = jiffies; | |
227 | struct nda_cacheinfo ci; | |
228 | struct nlmsghdr *nlh; | |
229 | struct ndmsg *ndm; | |
230 | bool send_ip, send_eth; | |
231 | ||
232 | nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags); | |
233 | if (nlh == NULL) | |
234 | return -EMSGSIZE; | |
235 | ||
236 | ndm = nlmsg_data(nlh); | |
237 | memset(ndm, 0, sizeof(*ndm)); | |
238 | ||
239 | send_eth = send_ip = true; | |
240 | ||
241 | if (type == RTM_GETNEIGH) { | |
242 | ndm->ndm_family = AF_INET; | |
243 | send_ip = rdst->remote_ip != htonl(INADDR_ANY); | |
244 | send_eth = !is_zero_ether_addr(fdb->eth_addr); | |
245 | } else | |
246 | ndm->ndm_family = AF_BRIDGE; | |
247 | ndm->ndm_state = fdb->state; | |
248 | ndm->ndm_ifindex = vxlan->dev->ifindex; | |
249 | ndm->ndm_flags = fdb->flags; | |
250 | ndm->ndm_type = NDA_DST; | |
251 | ||
252 | if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr)) | |
253 | goto nla_put_failure; | |
254 | ||
255 | if (send_ip && nla_put_be32(skb, NDA_DST, rdst->remote_ip)) | |
256 | goto nla_put_failure; | |
257 | ||
258 | if (rdst->remote_port && rdst->remote_port != vxlan->dst_port && | |
259 | nla_put_be16(skb, NDA_PORT, rdst->remote_port)) | |
260 | goto nla_put_failure; | |
261 | if (rdst->remote_vni != vxlan->default_dst.remote_vni && | |
262 | nla_put_u32(skb, NDA_VNI, rdst->remote_vni)) | |
263 | goto nla_put_failure; | |
264 | if (rdst->remote_ifindex && | |
265 | nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex)) | |
266 | goto nla_put_failure; | |
267 | ||
268 | ci.ndm_used = jiffies_to_clock_t(now - fdb->used); | |
269 | ci.ndm_confirmed = 0; | |
270 | ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated); | |
271 | ci.ndm_refcnt = 0; | |
272 | ||
273 | if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci)) | |
274 | goto nla_put_failure; | |
275 | ||
276 | return nlmsg_end(skb, nlh); | |
277 | ||
278 | nla_put_failure: | |
279 | nlmsg_cancel(skb, nlh); | |
280 | return -EMSGSIZE; | |
281 | } | |
282 | ||
283 | static inline size_t vxlan_nlmsg_size(void) | |
284 | { | |
285 | return NLMSG_ALIGN(sizeof(struct ndmsg)) | |
286 | + nla_total_size(ETH_ALEN) /* NDA_LLADDR */ | |
287 | + nla_total_size(sizeof(__be32)) /* NDA_DST */ | |
288 | + nla_total_size(sizeof(__be16)) /* NDA_PORT */ | |
289 | + nla_total_size(sizeof(__be32)) /* NDA_VNI */ | |
290 | + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */ | |
291 | + nla_total_size(sizeof(struct nda_cacheinfo)); | |
292 | } | |
293 | ||
294 | static void vxlan_fdb_notify(struct vxlan_dev *vxlan, | |
295 | struct vxlan_fdb *fdb, int type) | |
296 | { | |
297 | struct net *net = dev_net(vxlan->dev); | |
298 | struct sk_buff *skb; | |
299 | int err = -ENOBUFS; | |
300 | ||
301 | skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC); | |
302 | if (skb == NULL) | |
303 | goto errout; | |
304 | ||
305 | err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, | |
306 | first_remote_rtnl(fdb)); | |
307 | if (err < 0) { | |
308 | /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */ | |
309 | WARN_ON(err == -EMSGSIZE); | |
310 | kfree_skb(skb); | |
311 | goto errout; | |
312 | } | |
313 | ||
314 | rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC); | |
315 | return; | |
316 | errout: | |
317 | if (err < 0) | |
318 | rtnl_set_sk_err(net, RTNLGRP_NEIGH, err); | |
319 | } | |
320 | ||
321 | static void vxlan_ip_miss(struct net_device *dev, __be32 ipa) | |
322 | { | |
323 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
324 | struct vxlan_fdb f = { | |
325 | .state = NUD_STALE, | |
326 | }; | |
327 | struct vxlan_rdst remote = { | |
328 | .remote_ip = ipa, /* goes to NDA_DST */ | |
329 | .remote_vni = VXLAN_N_VID, | |
330 | }; | |
331 | ||
332 | INIT_LIST_HEAD(&f.remotes); | |
333 | list_add_rcu(&remote.list, &f.remotes); | |
334 | ||
335 | vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH); | |
336 | } | |
337 | ||
338 | static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN]) | |
339 | { | |
340 | struct vxlan_fdb f = { | |
341 | .state = NUD_STALE, | |
342 | }; | |
343 | ||
344 | INIT_LIST_HEAD(&f.remotes); | |
345 | memcpy(f.eth_addr, eth_addr, ETH_ALEN); | |
346 | ||
347 | vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH); | |
348 | } | |
349 | ||
350 | /* Hash Ethernet address */ | |
351 | static u32 eth_hash(const unsigned char *addr) | |
352 | { | |
353 | u64 value = get_unaligned((u64 *)addr); | |
354 | ||
355 | /* only want 6 bytes */ | |
356 | #ifdef __BIG_ENDIAN | |
357 | value >>= 16; | |
358 | #else | |
359 | value <<= 16; | |
360 | #endif | |
361 | return hash_64(value, FDB_HASH_BITS); | |
362 | } | |
363 | ||
364 | /* Hash chain to use given mac address */ | |
365 | static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan, | |
366 | const u8 *mac) | |
367 | { | |
368 | return &vxlan->fdb_head[eth_hash(mac)]; | |
369 | } | |
370 | ||
371 | /* Look up Ethernet address in forwarding table */ | |
372 | static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan, | |
373 | const u8 *mac) | |
374 | ||
375 | { | |
376 | struct hlist_head *head = vxlan_fdb_head(vxlan, mac); | |
377 | struct vxlan_fdb *f; | |
378 | ||
379 | hlist_for_each_entry_rcu(f, head, hlist) { | |
380 | if (compare_ether_addr(mac, f->eth_addr) == 0) | |
381 | return f; | |
382 | } | |
383 | ||
384 | return NULL; | |
385 | } | |
386 | ||
387 | static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan, | |
388 | const u8 *mac) | |
389 | { | |
390 | struct vxlan_fdb *f; | |
391 | ||
392 | f = __vxlan_find_mac(vxlan, mac); | |
393 | if (f) | |
394 | f->used = jiffies; | |
395 | ||
396 | return f; | |
397 | } | |
398 | ||
399 | /* caller should hold vxlan->hash_lock */ | |
400 | static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f, | |
401 | __be32 ip, __be16 port, | |
402 | __u32 vni, __u32 ifindex) | |
403 | { | |
404 | struct vxlan_rdst *rd; | |
405 | ||
406 | list_for_each_entry(rd, &f->remotes, list) { | |
407 | if (rd->remote_ip == ip && | |
408 | rd->remote_port == port && | |
409 | rd->remote_vni == vni && | |
410 | rd->remote_ifindex == ifindex) | |
411 | return rd; | |
412 | } | |
413 | ||
414 | return NULL; | |
415 | } | |
416 | ||
417 | /* Replace destination of unicast mac */ | |
418 | static int vxlan_fdb_replace(struct vxlan_fdb *f, | |
419 | __be32 ip, __be16 port, __u32 vni, __u32 ifindex) | |
420 | { | |
421 | struct vxlan_rdst *rd; | |
422 | ||
423 | rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex); | |
424 | if (rd) | |
425 | return 0; | |
426 | ||
427 | rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list); | |
428 | if (!rd) | |
429 | return 0; | |
430 | rd->remote_ip = ip; | |
431 | rd->remote_port = port; | |
432 | rd->remote_vni = vni; | |
433 | rd->remote_ifindex = ifindex; | |
434 | return 1; | |
435 | } | |
436 | ||
437 | /* Add/update destinations for multicast */ | |
438 | static int vxlan_fdb_append(struct vxlan_fdb *f, | |
439 | __be32 ip, __be16 port, __u32 vni, __u32 ifindex) | |
440 | { | |
441 | struct vxlan_rdst *rd; | |
442 | ||
443 | rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex); | |
444 | if (rd) | |
445 | return 0; | |
446 | ||
447 | rd = kmalloc(sizeof(*rd), GFP_ATOMIC); | |
448 | if (rd == NULL) | |
449 | return -ENOBUFS; | |
450 | rd->remote_ip = ip; | |
451 | rd->remote_port = port; | |
452 | rd->remote_vni = vni; | |
453 | rd->remote_ifindex = ifindex; | |
454 | ||
455 | list_add_tail_rcu(&rd->list, &f->remotes); | |
456 | ||
457 | return 1; | |
458 | } | |
459 | ||
460 | /* Add new entry to forwarding table -- assumes lock held */ | |
461 | static int vxlan_fdb_create(struct vxlan_dev *vxlan, | |
462 | const u8 *mac, __be32 ip, | |
463 | __u16 state, __u16 flags, | |
464 | __be16 port, __u32 vni, __u32 ifindex, | |
465 | __u8 ndm_flags) | |
466 | { | |
467 | struct vxlan_fdb *f; | |
468 | int notify = 0; | |
469 | ||
470 | f = __vxlan_find_mac(vxlan, mac); | |
471 | if (f) { | |
472 | if (flags & NLM_F_EXCL) { | |
473 | netdev_dbg(vxlan->dev, | |
474 | "lost race to create %pM\n", mac); | |
475 | return -EEXIST; | |
476 | } | |
477 | if (f->state != state) { | |
478 | f->state = state; | |
479 | f->updated = jiffies; | |
480 | notify = 1; | |
481 | } | |
482 | if (f->flags != ndm_flags) { | |
483 | f->flags = ndm_flags; | |
484 | f->updated = jiffies; | |
485 | notify = 1; | |
486 | } | |
487 | if ((flags & NLM_F_REPLACE)) { | |
488 | /* Only change unicasts */ | |
489 | if (!(is_multicast_ether_addr(f->eth_addr) || | |
490 | is_zero_ether_addr(f->eth_addr))) { | |
491 | int rc = vxlan_fdb_replace(f, ip, port, vni, | |
492 | ifindex); | |
493 | ||
494 | if (rc < 0) | |
495 | return rc; | |
496 | notify |= rc; | |
497 | } else | |
498 | return -EOPNOTSUPP; | |
499 | } | |
500 | if ((flags & NLM_F_APPEND) && | |
501 | (is_multicast_ether_addr(f->eth_addr) || | |
502 | is_zero_ether_addr(f->eth_addr))) { | |
503 | int rc = vxlan_fdb_append(f, ip, port, vni, ifindex); | |
504 | ||
505 | if (rc < 0) | |
506 | return rc; | |
507 | notify |= rc; | |
508 | } | |
509 | } else { | |
510 | if (!(flags & NLM_F_CREATE)) | |
511 | return -ENOENT; | |
512 | ||
513 | if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax) | |
514 | return -ENOSPC; | |
515 | ||
516 | /* Disallow replace to add a multicast entry */ | |
517 | if ((flags & NLM_F_REPLACE) && | |
518 | (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac))) | |
519 | return -EOPNOTSUPP; | |
520 | ||
521 | netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip); | |
522 | f = kmalloc(sizeof(*f), GFP_ATOMIC); | |
523 | if (!f) | |
524 | return -ENOMEM; | |
525 | ||
526 | notify = 1; | |
527 | f->state = state; | |
528 | f->flags = ndm_flags; | |
529 | f->updated = f->used = jiffies; | |
530 | INIT_LIST_HEAD(&f->remotes); | |
531 | memcpy(f->eth_addr, mac, ETH_ALEN); | |
532 | ||
533 | vxlan_fdb_append(f, ip, port, vni, ifindex); | |
534 | ||
535 | ++vxlan->addrcnt; | |
536 | hlist_add_head_rcu(&f->hlist, | |
537 | vxlan_fdb_head(vxlan, mac)); | |
538 | } | |
539 | ||
540 | if (notify) | |
541 | vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH); | |
542 | ||
543 | return 0; | |
544 | } | |
545 | ||
546 | static void vxlan_fdb_free_rdst(struct rcu_head *head) | |
547 | { | |
548 | struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu); | |
549 | kfree(rd); | |
550 | } | |
551 | ||
552 | static void vxlan_fdb_free(struct rcu_head *head) | |
553 | { | |
554 | struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu); | |
555 | struct vxlan_rdst *rd, *nd; | |
556 | ||
557 | list_for_each_entry_safe(rd, nd, &f->remotes, list) | |
558 | kfree(rd); | |
559 | kfree(f); | |
560 | } | |
561 | ||
562 | static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f) | |
563 | { | |
564 | netdev_dbg(vxlan->dev, | |
565 | "delete %pM\n", f->eth_addr); | |
566 | ||
567 | --vxlan->addrcnt; | |
568 | vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH); | |
569 | ||
570 | hlist_del_rcu(&f->hlist); | |
571 | call_rcu(&f->rcu, vxlan_fdb_free); | |
572 | } | |
573 | ||
574 | static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan, | |
575 | __be32 *ip, __be16 *port, u32 *vni, u32 *ifindex) | |
576 | { | |
577 | struct net *net = dev_net(vxlan->dev); | |
578 | ||
579 | if (tb[NDA_DST]) { | |
580 | if (nla_len(tb[NDA_DST]) != sizeof(__be32)) | |
581 | return -EAFNOSUPPORT; | |
582 | ||
583 | *ip = nla_get_be32(tb[NDA_DST]); | |
584 | } else { | |
585 | *ip = htonl(INADDR_ANY); | |
586 | } | |
587 | ||
588 | if (tb[NDA_PORT]) { | |
589 | if (nla_len(tb[NDA_PORT]) != sizeof(__be16)) | |
590 | return -EINVAL; | |
591 | *port = nla_get_be16(tb[NDA_PORT]); | |
592 | } else { | |
593 | *port = vxlan->dst_port; | |
594 | } | |
595 | ||
596 | if (tb[NDA_VNI]) { | |
597 | if (nla_len(tb[NDA_VNI]) != sizeof(u32)) | |
598 | return -EINVAL; | |
599 | *vni = nla_get_u32(tb[NDA_VNI]); | |
600 | } else { | |
601 | *vni = vxlan->default_dst.remote_vni; | |
602 | } | |
603 | ||
604 | if (tb[NDA_IFINDEX]) { | |
605 | struct net_device *tdev; | |
606 | ||
607 | if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32)) | |
608 | return -EINVAL; | |
609 | *ifindex = nla_get_u32(tb[NDA_IFINDEX]); | |
610 | tdev = dev_get_by_index(net, *ifindex); | |
611 | if (!tdev) | |
612 | return -EADDRNOTAVAIL; | |
613 | dev_put(tdev); | |
614 | } else { | |
615 | *ifindex = 0; | |
616 | } | |
617 | ||
618 | return 0; | |
619 | } | |
620 | ||
621 | /* Add static entry (via netlink) */ | |
622 | static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[], | |
623 | struct net_device *dev, | |
624 | const unsigned char *addr, u16 flags) | |
625 | { | |
626 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
627 | /* struct net *net = dev_net(vxlan->dev); */ | |
628 | __be32 ip; | |
629 | __be16 port; | |
630 | u32 vni, ifindex; | |
631 | int err; | |
632 | ||
633 | if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) { | |
634 | pr_info("RTM_NEWNEIGH with invalid state %#x\n", | |
635 | ndm->ndm_state); | |
636 | return -EINVAL; | |
637 | } | |
638 | ||
639 | if (tb[NDA_DST] == NULL) | |
640 | return -EINVAL; | |
641 | ||
642 | err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex); | |
643 | if (err) | |
644 | return err; | |
645 | ||
646 | spin_lock_bh(&vxlan->hash_lock); | |
647 | err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags, | |
648 | port, vni, ifindex, ndm->ndm_flags); | |
649 | spin_unlock_bh(&vxlan->hash_lock); | |
650 | ||
651 | return err; | |
652 | } | |
653 | ||
654 | /* Delete entry (via netlink) */ | |
655 | static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[], | |
656 | struct net_device *dev, | |
657 | const unsigned char *addr) | |
658 | { | |
659 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
660 | struct vxlan_fdb *f; | |
661 | struct vxlan_rdst *rd = NULL; | |
662 | __be32 ip; | |
663 | __be16 port; | |
664 | u32 vni, ifindex; | |
665 | int err; | |
666 | ||
667 | err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex); | |
668 | if (err) | |
669 | return err; | |
670 | ||
671 | err = -ENOENT; | |
672 | ||
673 | spin_lock_bh(&vxlan->hash_lock); | |
674 | f = vxlan_find_mac(vxlan, addr); | |
675 | if (!f) | |
676 | goto out; | |
677 | ||
678 | if (ip != htonl(INADDR_ANY)) { | |
679 | rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex); | |
680 | if (!rd) | |
681 | goto out; | |
682 | } | |
683 | ||
684 | err = 0; | |
685 | ||
686 | /* remove a destination if it's not the only one on the list, | |
687 | * otherwise destroy the fdb entry | |
688 | */ | |
689 | if (rd && !list_is_singular(&f->remotes)) { | |
690 | list_del_rcu(&rd->list); | |
691 | call_rcu(&rd->rcu, vxlan_fdb_free_rdst); | |
692 | goto out; | |
693 | } | |
694 | ||
695 | vxlan_fdb_destroy(vxlan, f); | |
696 | ||
697 | out: | |
698 | spin_unlock_bh(&vxlan->hash_lock); | |
699 | ||
700 | return err; | |
701 | } | |
702 | ||
703 | /* Dump forwarding table */ | |
704 | static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb, | |
705 | struct net_device *dev, int idx) | |
706 | { | |
707 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
708 | unsigned int h; | |
709 | ||
710 | for (h = 0; h < FDB_HASH_SIZE; ++h) { | |
711 | struct vxlan_fdb *f; | |
712 | int err; | |
713 | ||
714 | hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) { | |
715 | struct vxlan_rdst *rd; | |
716 | ||
717 | if (idx < cb->args[0]) | |
718 | goto skip; | |
719 | ||
720 | list_for_each_entry_rcu(rd, &f->remotes, list) { | |
721 | err = vxlan_fdb_info(skb, vxlan, f, | |
722 | NETLINK_CB(cb->skb).portid, | |
723 | cb->nlh->nlmsg_seq, | |
724 | RTM_NEWNEIGH, | |
725 | NLM_F_MULTI, rd); | |
726 | if (err < 0) | |
727 | goto out; | |
728 | } | |
729 | skip: | |
730 | ++idx; | |
731 | } | |
732 | } | |
733 | out: | |
734 | return idx; | |
735 | } | |
736 | ||
737 | /* Watch incoming packets to learn mapping between Ethernet address | |
738 | * and Tunnel endpoint. | |
739 | * Return true if packet is bogus and should be droppped. | |
740 | */ | |
741 | static bool vxlan_snoop(struct net_device *dev, | |
742 | __be32 src_ip, const u8 *src_mac) | |
743 | { | |
744 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
745 | struct vxlan_fdb *f; | |
746 | ||
747 | f = vxlan_find_mac(vxlan, src_mac); | |
748 | if (likely(f)) { | |
749 | struct vxlan_rdst *rdst = first_remote_rcu(f); | |
750 | ||
751 | if (likely(rdst->remote_ip == src_ip)) | |
752 | return false; | |
753 | ||
754 | /* Don't migrate static entries, drop packets */ | |
755 | if (f->state & NUD_NOARP) | |
756 | return true; | |
757 | ||
758 | if (net_ratelimit()) | |
759 | netdev_info(dev, | |
760 | "%pM migrated from %pI4 to %pI4\n", | |
761 | src_mac, &rdst->remote_ip, &src_ip); | |
762 | ||
763 | rdst->remote_ip = src_ip; | |
764 | f->updated = jiffies; | |
765 | vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH); | |
766 | } else { | |
767 | /* learned new entry */ | |
768 | spin_lock(&vxlan->hash_lock); | |
769 | ||
770 | /* close off race between vxlan_flush and incoming packets */ | |
771 | if (netif_running(dev)) | |
772 | vxlan_fdb_create(vxlan, src_mac, src_ip, | |
773 | NUD_REACHABLE, | |
774 | NLM_F_EXCL|NLM_F_CREATE, | |
775 | vxlan->dst_port, | |
776 | vxlan->default_dst.remote_vni, | |
777 | 0, NTF_SELF); | |
778 | spin_unlock(&vxlan->hash_lock); | |
779 | } | |
780 | ||
781 | return false; | |
782 | } | |
783 | ||
784 | /* See if multicast group is already in use by other ID */ | |
785 | static bool vxlan_group_used(struct vxlan_net *vn, __be32 remote_ip) | |
786 | { | |
787 | struct vxlan_dev *vxlan; | |
788 | ||
789 | list_for_each_entry(vxlan, &vn->vxlan_list, next) { | |
790 | if (!netif_running(vxlan->dev)) | |
791 | continue; | |
792 | ||
793 | if (vxlan->default_dst.remote_ip == remote_ip) | |
794 | return true; | |
795 | } | |
796 | ||
797 | return false; | |
798 | } | |
799 | ||
800 | static void vxlan_sock_hold(struct vxlan_sock *vs) | |
801 | { | |
802 | atomic_inc(&vs->refcnt); | |
803 | } | |
804 | ||
805 | static void vxlan_sock_release(struct vxlan_net *vn, struct vxlan_sock *vs) | |
806 | { | |
807 | if (!atomic_dec_and_test(&vs->refcnt)) | |
808 | return; | |
809 | ||
810 | spin_lock(&vn->sock_lock); | |
811 | hlist_del_rcu(&vs->hlist); | |
812 | spin_unlock(&vn->sock_lock); | |
813 | ||
814 | queue_work(vxlan_wq, &vs->del_work); | |
815 | } | |
816 | ||
817 | /* Callback to update multicast group membership when first VNI on | |
818 | * multicast asddress is brought up | |
819 | * Done as workqueue because ip_mc_join_group acquires RTNL. | |
820 | */ | |
821 | static void vxlan_igmp_join(struct work_struct *work) | |
822 | { | |
823 | struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join); | |
824 | struct vxlan_net *vn = net_generic(dev_net(vxlan->dev), vxlan_net_id); | |
825 | struct vxlan_sock *vs = vxlan->vn_sock; | |
826 | struct sock *sk = vs->sock->sk; | |
827 | struct ip_mreqn mreq = { | |
828 | .imr_multiaddr.s_addr = vxlan->default_dst.remote_ip, | |
829 | .imr_ifindex = vxlan->default_dst.remote_ifindex, | |
830 | }; | |
831 | ||
832 | lock_sock(sk); | |
833 | ip_mc_join_group(sk, &mreq); | |
834 | release_sock(sk); | |
835 | ||
836 | vxlan_sock_release(vn, vs); | |
837 | dev_put(vxlan->dev); | |
838 | } | |
839 | ||
840 | /* Inverse of vxlan_igmp_join when last VNI is brought down */ | |
841 | static void vxlan_igmp_leave(struct work_struct *work) | |
842 | { | |
843 | struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave); | |
844 | struct vxlan_net *vn = net_generic(dev_net(vxlan->dev), vxlan_net_id); | |
845 | struct vxlan_sock *vs = vxlan->vn_sock; | |
846 | struct sock *sk = vs->sock->sk; | |
847 | struct ip_mreqn mreq = { | |
848 | .imr_multiaddr.s_addr = vxlan->default_dst.remote_ip, | |
849 | .imr_ifindex = vxlan->default_dst.remote_ifindex, | |
850 | }; | |
851 | ||
852 | lock_sock(sk); | |
853 | ip_mc_leave_group(sk, &mreq); | |
854 | release_sock(sk); | |
855 | ||
856 | vxlan_sock_release(vn, vs); | |
857 | dev_put(vxlan->dev); | |
858 | } | |
859 | ||
860 | /* Callback from net/ipv4/udp.c to receive packets */ | |
861 | static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb) | |
862 | { | |
863 | struct iphdr *oip; | |
864 | struct vxlanhdr *vxh; | |
865 | struct vxlan_dev *vxlan; | |
866 | struct pcpu_tstats *stats; | |
867 | __be16 port; | |
868 | __u32 vni; | |
869 | int err; | |
870 | ||
871 | /* pop off outer UDP header */ | |
872 | __skb_pull(skb, sizeof(struct udphdr)); | |
873 | ||
874 | /* Need Vxlan and inner Ethernet header to be present */ | |
875 | if (!pskb_may_pull(skb, sizeof(struct vxlanhdr))) | |
876 | goto error; | |
877 | ||
878 | /* Drop packets with reserved bits set */ | |
879 | vxh = (struct vxlanhdr *) skb->data; | |
880 | if (vxh->vx_flags != htonl(VXLAN_FLAGS) || | |
881 | (vxh->vx_vni & htonl(0xff))) { | |
882 | netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n", | |
883 | ntohl(vxh->vx_flags), ntohl(vxh->vx_vni)); | |
884 | goto error; | |
885 | } | |
886 | ||
887 | __skb_pull(skb, sizeof(struct vxlanhdr)); | |
888 | ||
889 | /* Is this VNI defined? */ | |
890 | vni = ntohl(vxh->vx_vni) >> 8; | |
891 | port = inet_sk(sk)->inet_sport; | |
892 | vxlan = vxlan_find_vni(sock_net(sk), vni, port); | |
893 | if (!vxlan) { | |
894 | netdev_dbg(skb->dev, "unknown vni %d port %u\n", | |
895 | vni, ntohs(port)); | |
896 | goto drop; | |
897 | } | |
898 | ||
899 | if (!pskb_may_pull(skb, ETH_HLEN)) { | |
900 | vxlan->dev->stats.rx_length_errors++; | |
901 | vxlan->dev->stats.rx_errors++; | |
902 | goto drop; | |
903 | } | |
904 | ||
905 | skb_reset_mac_header(skb); | |
906 | ||
907 | /* Re-examine inner Ethernet packet */ | |
908 | oip = ip_hdr(skb); | |
909 | skb->protocol = eth_type_trans(skb, vxlan->dev); | |
910 | ||
911 | /* Ignore packet loops (and multicast echo) */ | |
912 | if (compare_ether_addr(eth_hdr(skb)->h_source, | |
913 | vxlan->dev->dev_addr) == 0) | |
914 | goto drop; | |
915 | ||
916 | if ((vxlan->flags & VXLAN_F_LEARN) && | |
917 | vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source)) | |
918 | goto drop; | |
919 | ||
920 | __skb_tunnel_rx(skb, vxlan->dev); | |
921 | skb_reset_network_header(skb); | |
922 | ||
923 | /* If the NIC driver gave us an encapsulated packet with | |
924 | * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled, | |
925 | * leave the CHECKSUM_UNNECESSARY, the device checksummed it | |
926 | * for us. Otherwise force the upper layers to verify it. | |
927 | */ | |
928 | if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation || | |
929 | !(vxlan->dev->features & NETIF_F_RXCSUM)) | |
930 | skb->ip_summed = CHECKSUM_NONE; | |
931 | ||
932 | skb->encapsulation = 0; | |
933 | ||
934 | err = IP_ECN_decapsulate(oip, skb); | |
935 | if (unlikely(err)) { | |
936 | if (log_ecn_error) | |
937 | net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n", | |
938 | &oip->saddr, oip->tos); | |
939 | if (err > 1) { | |
940 | ++vxlan->dev->stats.rx_frame_errors; | |
941 | ++vxlan->dev->stats.rx_errors; | |
942 | goto drop; | |
943 | } | |
944 | } | |
945 | ||
946 | stats = this_cpu_ptr(vxlan->dev->tstats); | |
947 | u64_stats_update_begin(&stats->syncp); | |
948 | stats->rx_packets++; | |
949 | stats->rx_bytes += skb->len; | |
950 | u64_stats_update_end(&stats->syncp); | |
951 | ||
952 | netif_rx(skb); | |
953 | ||
954 | return 0; | |
955 | error: | |
956 | /* Put UDP header back */ | |
957 | __skb_push(skb, sizeof(struct udphdr)); | |
958 | ||
959 | return 1; | |
960 | drop: | |
961 | /* Consume bad packet */ | |
962 | kfree_skb(skb); | |
963 | return 0; | |
964 | } | |
965 | ||
966 | static int arp_reduce(struct net_device *dev, struct sk_buff *skb) | |
967 | { | |
968 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
969 | struct arphdr *parp; | |
970 | u8 *arpptr, *sha; | |
971 | __be32 sip, tip; | |
972 | struct neighbour *n; | |
973 | ||
974 | if (dev->flags & IFF_NOARP) | |
975 | goto out; | |
976 | ||
977 | if (!pskb_may_pull(skb, arp_hdr_len(dev))) { | |
978 | dev->stats.tx_dropped++; | |
979 | goto out; | |
980 | } | |
981 | parp = arp_hdr(skb); | |
982 | ||
983 | if ((parp->ar_hrd != htons(ARPHRD_ETHER) && | |
984 | parp->ar_hrd != htons(ARPHRD_IEEE802)) || | |
985 | parp->ar_pro != htons(ETH_P_IP) || | |
986 | parp->ar_op != htons(ARPOP_REQUEST) || | |
987 | parp->ar_hln != dev->addr_len || | |
988 | parp->ar_pln != 4) | |
989 | goto out; | |
990 | arpptr = (u8 *)parp + sizeof(struct arphdr); | |
991 | sha = arpptr; | |
992 | arpptr += dev->addr_len; /* sha */ | |
993 | memcpy(&sip, arpptr, sizeof(sip)); | |
994 | arpptr += sizeof(sip); | |
995 | arpptr += dev->addr_len; /* tha */ | |
996 | memcpy(&tip, arpptr, sizeof(tip)); | |
997 | ||
998 | if (ipv4_is_loopback(tip) || | |
999 | ipv4_is_multicast(tip)) | |
1000 | goto out; | |
1001 | ||
1002 | n = neigh_lookup(&arp_tbl, &tip, dev); | |
1003 | ||
1004 | if (n) { | |
1005 | struct vxlan_fdb *f; | |
1006 | struct sk_buff *reply; | |
1007 | ||
1008 | if (!(n->nud_state & NUD_CONNECTED)) { | |
1009 | neigh_release(n); | |
1010 | goto out; | |
1011 | } | |
1012 | ||
1013 | f = vxlan_find_mac(vxlan, n->ha); | |
1014 | if (f && first_remote_rcu(f)->remote_ip == htonl(INADDR_ANY)) { | |
1015 | /* bridge-local neighbor */ | |
1016 | neigh_release(n); | |
1017 | goto out; | |
1018 | } | |
1019 | ||
1020 | reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha, | |
1021 | n->ha, sha); | |
1022 | ||
1023 | neigh_release(n); | |
1024 | ||
1025 | skb_reset_mac_header(reply); | |
1026 | __skb_pull(reply, skb_network_offset(reply)); | |
1027 | reply->ip_summed = CHECKSUM_UNNECESSARY; | |
1028 | reply->pkt_type = PACKET_HOST; | |
1029 | ||
1030 | if (netif_rx_ni(reply) == NET_RX_DROP) | |
1031 | dev->stats.rx_dropped++; | |
1032 | } else if (vxlan->flags & VXLAN_F_L3MISS) | |
1033 | vxlan_ip_miss(dev, tip); | |
1034 | out: | |
1035 | consume_skb(skb); | |
1036 | return NETDEV_TX_OK; | |
1037 | } | |
1038 | ||
1039 | static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb) | |
1040 | { | |
1041 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1042 | struct neighbour *n; | |
1043 | struct iphdr *pip; | |
1044 | ||
1045 | if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) | |
1046 | return false; | |
1047 | ||
1048 | n = NULL; | |
1049 | switch (ntohs(eth_hdr(skb)->h_proto)) { | |
1050 | case ETH_P_IP: | |
1051 | if (!pskb_may_pull(skb, sizeof(struct iphdr))) | |
1052 | return false; | |
1053 | pip = ip_hdr(skb); | |
1054 | n = neigh_lookup(&arp_tbl, &pip->daddr, dev); | |
1055 | break; | |
1056 | default: | |
1057 | return false; | |
1058 | } | |
1059 | ||
1060 | if (n) { | |
1061 | bool diff; | |
1062 | ||
1063 | diff = compare_ether_addr(eth_hdr(skb)->h_dest, n->ha) != 0; | |
1064 | if (diff) { | |
1065 | memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest, | |
1066 | dev->addr_len); | |
1067 | memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len); | |
1068 | } | |
1069 | neigh_release(n); | |
1070 | return diff; | |
1071 | } else if (vxlan->flags & VXLAN_F_L3MISS) | |
1072 | vxlan_ip_miss(dev, pip->daddr); | |
1073 | return false; | |
1074 | } | |
1075 | ||
1076 | static void vxlan_sock_put(struct sk_buff *skb) | |
1077 | { | |
1078 | sock_put(skb->sk); | |
1079 | } | |
1080 | ||
1081 | /* On transmit, associate with the tunnel socket */ | |
1082 | static void vxlan_set_owner(struct net_device *dev, struct sk_buff *skb) | |
1083 | { | |
1084 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1085 | struct sock *sk = vxlan->vn_sock->sock->sk; | |
1086 | ||
1087 | skb_orphan(skb); | |
1088 | sock_hold(sk); | |
1089 | skb->sk = sk; | |
1090 | skb->destructor = vxlan_sock_put; | |
1091 | } | |
1092 | ||
1093 | /* Compute source port for outgoing packet | |
1094 | * first choice to use L4 flow hash since it will spread | |
1095 | * better and maybe available from hardware | |
1096 | * secondary choice is to use jhash on the Ethernet header | |
1097 | */ | |
1098 | static __be16 vxlan_src_port(const struct vxlan_dev *vxlan, struct sk_buff *skb) | |
1099 | { | |
1100 | unsigned int range = (vxlan->port_max - vxlan->port_min) + 1; | |
1101 | u32 hash; | |
1102 | ||
1103 | hash = skb_get_rxhash(skb); | |
1104 | if (!hash) | |
1105 | hash = jhash(skb->data, 2 * ETH_ALEN, | |
1106 | (__force u32) skb->protocol); | |
1107 | ||
1108 | return htons((((u64) hash * range) >> 32) + vxlan->port_min); | |
1109 | } | |
1110 | ||
1111 | static int handle_offloads(struct sk_buff *skb) | |
1112 | { | |
1113 | if (skb_is_gso(skb)) { | |
1114 | int err = skb_unclone(skb, GFP_ATOMIC); | |
1115 | if (unlikely(err)) | |
1116 | return err; | |
1117 | ||
1118 | skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL; | |
1119 | } else if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1120 | skb->ip_summed = CHECKSUM_NONE; | |
1121 | ||
1122 | return 0; | |
1123 | } | |
1124 | ||
1125 | /* Bypass encapsulation if the destination is local */ | |
1126 | static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan, | |
1127 | struct vxlan_dev *dst_vxlan) | |
1128 | { | |
1129 | struct pcpu_tstats *tx_stats = this_cpu_ptr(src_vxlan->dev->tstats); | |
1130 | struct pcpu_tstats *rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats); | |
1131 | ||
1132 | skb->pkt_type = PACKET_HOST; | |
1133 | skb->encapsulation = 0; | |
1134 | skb->dev = dst_vxlan->dev; | |
1135 | __skb_pull(skb, skb_network_offset(skb)); | |
1136 | ||
1137 | if (dst_vxlan->flags & VXLAN_F_LEARN) | |
1138 | vxlan_snoop(skb->dev, htonl(INADDR_LOOPBACK), | |
1139 | eth_hdr(skb)->h_source); | |
1140 | ||
1141 | u64_stats_update_begin(&tx_stats->syncp); | |
1142 | tx_stats->tx_packets++; | |
1143 | tx_stats->tx_bytes += skb->len; | |
1144 | u64_stats_update_end(&tx_stats->syncp); | |
1145 | ||
1146 | if (netif_rx(skb) == NET_RX_SUCCESS) { | |
1147 | u64_stats_update_begin(&rx_stats->syncp); | |
1148 | rx_stats->rx_packets++; | |
1149 | rx_stats->rx_bytes += skb->len; | |
1150 | u64_stats_update_end(&rx_stats->syncp); | |
1151 | } else { | |
1152 | skb->dev->stats.rx_dropped++; | |
1153 | } | |
1154 | } | |
1155 | ||
1156 | static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev, | |
1157 | struct vxlan_rdst *rdst, bool did_rsc) | |
1158 | { | |
1159 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1160 | struct rtable *rt; | |
1161 | const struct iphdr *old_iph; | |
1162 | struct vxlanhdr *vxh; | |
1163 | struct udphdr *uh; | |
1164 | struct flowi4 fl4; | |
1165 | __be32 dst; | |
1166 | __be16 src_port, dst_port; | |
1167 | u32 vni; | |
1168 | __be16 df = 0; | |
1169 | __u8 tos, ttl; | |
1170 | int err; | |
1171 | ||
1172 | dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port; | |
1173 | vni = rdst->remote_vni; | |
1174 | dst = rdst->remote_ip; | |
1175 | ||
1176 | if (!dst) { | |
1177 | if (did_rsc) { | |
1178 | /* short-circuited back to local bridge */ | |
1179 | vxlan_encap_bypass(skb, vxlan, vxlan); | |
1180 | return; | |
1181 | } | |
1182 | goto drop; | |
1183 | } | |
1184 | ||
1185 | if (!skb->encapsulation) { | |
1186 | skb_reset_inner_headers(skb); | |
1187 | skb->encapsulation = 1; | |
1188 | } | |
1189 | ||
1190 | /* Need space for new headers (invalidates iph ptr) */ | |
1191 | if (skb_cow_head(skb, VXLAN_HEADROOM)) | |
1192 | goto drop; | |
1193 | ||
1194 | old_iph = ip_hdr(skb); | |
1195 | ||
1196 | ttl = vxlan->ttl; | |
1197 | if (!ttl && IN_MULTICAST(ntohl(dst))) | |
1198 | ttl = 1; | |
1199 | ||
1200 | tos = vxlan->tos; | |
1201 | if (tos == 1) | |
1202 | tos = ip_tunnel_get_dsfield(old_iph, skb); | |
1203 | ||
1204 | src_port = vxlan_src_port(vxlan, skb); | |
1205 | ||
1206 | memset(&fl4, 0, sizeof(fl4)); | |
1207 | fl4.flowi4_oif = rdst->remote_ifindex; | |
1208 | fl4.flowi4_tos = RT_TOS(tos); | |
1209 | fl4.daddr = dst; | |
1210 | fl4.saddr = vxlan->saddr; | |
1211 | ||
1212 | rt = ip_route_output_key(dev_net(dev), &fl4); | |
1213 | if (IS_ERR(rt)) { | |
1214 | netdev_dbg(dev, "no route to %pI4\n", &dst); | |
1215 | dev->stats.tx_carrier_errors++; | |
1216 | goto tx_error; | |
1217 | } | |
1218 | ||
1219 | if (rt->dst.dev == dev) { | |
1220 | netdev_dbg(dev, "circular route to %pI4\n", &dst); | |
1221 | ip_rt_put(rt); | |
1222 | dev->stats.collisions++; | |
1223 | goto tx_error; | |
1224 | } | |
1225 | ||
1226 | /* Bypass encapsulation if the destination is local */ | |
1227 | if (rt->rt_flags & RTCF_LOCAL && | |
1228 | !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) { | |
1229 | struct vxlan_dev *dst_vxlan; | |
1230 | ||
1231 | ip_rt_put(rt); | |
1232 | dst_vxlan = vxlan_find_vni(dev_net(dev), vni, dst_port); | |
1233 | if (!dst_vxlan) | |
1234 | goto tx_error; | |
1235 | vxlan_encap_bypass(skb, vxlan, dst_vxlan); | |
1236 | return; | |
1237 | } | |
1238 | vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh)); | |
1239 | vxh->vx_flags = htonl(VXLAN_FLAGS); | |
1240 | vxh->vx_vni = htonl(vni << 8); | |
1241 | ||
1242 | __skb_push(skb, sizeof(*uh)); | |
1243 | skb_reset_transport_header(skb); | |
1244 | uh = udp_hdr(skb); | |
1245 | ||
1246 | uh->dest = dst_port; | |
1247 | uh->source = src_port; | |
1248 | ||
1249 | uh->len = htons(skb->len); | |
1250 | uh->check = 0; | |
1251 | ||
1252 | vxlan_set_owner(dev, skb); | |
1253 | ||
1254 | if (handle_offloads(skb)) | |
1255 | goto drop; | |
1256 | ||
1257 | tos = ip_tunnel_ecn_encap(tos, old_iph, skb); | |
1258 | ttl = ttl ? : ip4_dst_hoplimit(&rt->dst); | |
1259 | ||
1260 | err = iptunnel_xmit(dev_net(dev), rt, skb, fl4.saddr, dst, | |
1261 | IPPROTO_UDP, tos, ttl, df); | |
1262 | iptunnel_xmit_stats(err, &dev->stats, dev->tstats); | |
1263 | ||
1264 | return; | |
1265 | ||
1266 | drop: | |
1267 | dev->stats.tx_dropped++; | |
1268 | goto tx_free; | |
1269 | ||
1270 | tx_error: | |
1271 | dev->stats.tx_errors++; | |
1272 | tx_free: | |
1273 | dev_kfree_skb(skb); | |
1274 | } | |
1275 | ||
1276 | /* Transmit local packets over Vxlan | |
1277 | * | |
1278 | * Outer IP header inherits ECN and DF from inner header. | |
1279 | * Outer UDP destination is the VXLAN assigned port. | |
1280 | * source port is based on hash of flow | |
1281 | */ | |
1282 | static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev) | |
1283 | { | |
1284 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1285 | struct ethhdr *eth; | |
1286 | bool did_rsc = false; | |
1287 | struct vxlan_rdst *rdst; | |
1288 | struct vxlan_fdb *f; | |
1289 | ||
1290 | skb_reset_mac_header(skb); | |
1291 | eth = eth_hdr(skb); | |
1292 | ||
1293 | if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP) | |
1294 | return arp_reduce(dev, skb); | |
1295 | ||
1296 | f = vxlan_find_mac(vxlan, eth->h_dest); | |
1297 | did_rsc = false; | |
1298 | ||
1299 | if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) && | |
1300 | ntohs(eth->h_proto) == ETH_P_IP) { | |
1301 | did_rsc = route_shortcircuit(dev, skb); | |
1302 | if (did_rsc) | |
1303 | f = vxlan_find_mac(vxlan, eth->h_dest); | |
1304 | } | |
1305 | ||
1306 | if (f == NULL) { | |
1307 | f = vxlan_find_mac(vxlan, all_zeros_mac); | |
1308 | if (f == NULL) { | |
1309 | if ((vxlan->flags & VXLAN_F_L2MISS) && | |
1310 | !is_multicast_ether_addr(eth->h_dest)) | |
1311 | vxlan_fdb_miss(vxlan, eth->h_dest); | |
1312 | ||
1313 | dev->stats.tx_dropped++; | |
1314 | dev_kfree_skb(skb); | |
1315 | return NETDEV_TX_OK; | |
1316 | } | |
1317 | } | |
1318 | ||
1319 | list_for_each_entry_rcu(rdst, &f->remotes, list) { | |
1320 | struct sk_buff *skb1; | |
1321 | ||
1322 | skb1 = skb_clone(skb, GFP_ATOMIC); | |
1323 | if (skb1) | |
1324 | vxlan_xmit_one(skb1, dev, rdst, did_rsc); | |
1325 | } | |
1326 | ||
1327 | dev_kfree_skb(skb); | |
1328 | return NETDEV_TX_OK; | |
1329 | } | |
1330 | ||
1331 | /* Walk the forwarding table and purge stale entries */ | |
1332 | static void vxlan_cleanup(unsigned long arg) | |
1333 | { | |
1334 | struct vxlan_dev *vxlan = (struct vxlan_dev *) arg; | |
1335 | unsigned long next_timer = jiffies + FDB_AGE_INTERVAL; | |
1336 | unsigned int h; | |
1337 | ||
1338 | if (!netif_running(vxlan->dev)) | |
1339 | return; | |
1340 | ||
1341 | spin_lock_bh(&vxlan->hash_lock); | |
1342 | for (h = 0; h < FDB_HASH_SIZE; ++h) { | |
1343 | struct hlist_node *p, *n; | |
1344 | hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) { | |
1345 | struct vxlan_fdb *f | |
1346 | = container_of(p, struct vxlan_fdb, hlist); | |
1347 | unsigned long timeout; | |
1348 | ||
1349 | if (f->state & NUD_PERMANENT) | |
1350 | continue; | |
1351 | ||
1352 | timeout = f->used + vxlan->age_interval * HZ; | |
1353 | if (time_before_eq(timeout, jiffies)) { | |
1354 | netdev_dbg(vxlan->dev, | |
1355 | "garbage collect %pM\n", | |
1356 | f->eth_addr); | |
1357 | f->state = NUD_STALE; | |
1358 | vxlan_fdb_destroy(vxlan, f); | |
1359 | } else if (time_before(timeout, next_timer)) | |
1360 | next_timer = timeout; | |
1361 | } | |
1362 | } | |
1363 | spin_unlock_bh(&vxlan->hash_lock); | |
1364 | ||
1365 | mod_timer(&vxlan->age_timer, next_timer); | |
1366 | } | |
1367 | ||
1368 | /* Setup stats when device is created */ | |
1369 | static int vxlan_init(struct net_device *dev) | |
1370 | { | |
1371 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1372 | struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id); | |
1373 | struct vxlan_sock *vs; | |
1374 | __u32 vni = vxlan->default_dst.remote_vni; | |
1375 | ||
1376 | dev->tstats = alloc_percpu(struct pcpu_tstats); | |
1377 | if (!dev->tstats) | |
1378 | return -ENOMEM; | |
1379 | ||
1380 | spin_lock(&vn->sock_lock); | |
1381 | vs = vxlan_find_port(dev_net(dev), vxlan->dst_port); | |
1382 | if (vs) { | |
1383 | /* If we have a socket with same port already, reuse it */ | |
1384 | atomic_inc(&vs->refcnt); | |
1385 | vxlan->vn_sock = vs; | |
1386 | hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni)); | |
1387 | } else { | |
1388 | /* otherwise make new socket outside of RTNL */ | |
1389 | dev_hold(dev); | |
1390 | queue_work(vxlan_wq, &vxlan->sock_work); | |
1391 | } | |
1392 | spin_unlock(&vn->sock_lock); | |
1393 | ||
1394 | return 0; | |
1395 | } | |
1396 | ||
1397 | static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan) | |
1398 | { | |
1399 | struct vxlan_fdb *f; | |
1400 | ||
1401 | spin_lock_bh(&vxlan->hash_lock); | |
1402 | f = __vxlan_find_mac(vxlan, all_zeros_mac); | |
1403 | if (f) | |
1404 | vxlan_fdb_destroy(vxlan, f); | |
1405 | spin_unlock_bh(&vxlan->hash_lock); | |
1406 | } | |
1407 | ||
1408 | static void vxlan_uninit(struct net_device *dev) | |
1409 | { | |
1410 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1411 | struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id); | |
1412 | struct vxlan_sock *vs = vxlan->vn_sock; | |
1413 | ||
1414 | vxlan_fdb_delete_default(vxlan); | |
1415 | ||
1416 | if (vs) | |
1417 | vxlan_sock_release(vn, vs); | |
1418 | free_percpu(dev->tstats); | |
1419 | } | |
1420 | ||
1421 | /* Start ageing timer and join group when device is brought up */ | |
1422 | static int vxlan_open(struct net_device *dev) | |
1423 | { | |
1424 | struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id); | |
1425 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1426 | struct vxlan_sock *vs = vxlan->vn_sock; | |
1427 | ||
1428 | /* socket hasn't been created */ | |
1429 | if (!vs) | |
1430 | return -ENOTCONN; | |
1431 | ||
1432 | if (IN_MULTICAST(ntohl(vxlan->default_dst.remote_ip)) && | |
1433 | ! vxlan_group_used(vn, vxlan->default_dst.remote_ip)) { | |
1434 | vxlan_sock_hold(vs); | |
1435 | dev_hold(dev); | |
1436 | queue_work(vxlan_wq, &vxlan->igmp_join); | |
1437 | } | |
1438 | ||
1439 | if (vxlan->age_interval) | |
1440 | mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL); | |
1441 | ||
1442 | return 0; | |
1443 | } | |
1444 | ||
1445 | /* Purge the forwarding table */ | |
1446 | static void vxlan_flush(struct vxlan_dev *vxlan) | |
1447 | { | |
1448 | unsigned int h; | |
1449 | ||
1450 | spin_lock_bh(&vxlan->hash_lock); | |
1451 | for (h = 0; h < FDB_HASH_SIZE; ++h) { | |
1452 | struct hlist_node *p, *n; | |
1453 | hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) { | |
1454 | struct vxlan_fdb *f | |
1455 | = container_of(p, struct vxlan_fdb, hlist); | |
1456 | /* the all_zeros_mac entry is deleted at vxlan_uninit */ | |
1457 | if (!is_zero_ether_addr(f->eth_addr)) | |
1458 | vxlan_fdb_destroy(vxlan, f); | |
1459 | } | |
1460 | } | |
1461 | spin_unlock_bh(&vxlan->hash_lock); | |
1462 | } | |
1463 | ||
1464 | /* Cleanup timer and forwarding table on shutdown */ | |
1465 | static int vxlan_stop(struct net_device *dev) | |
1466 | { | |
1467 | struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id); | |
1468 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1469 | struct vxlan_sock *vs = vxlan->vn_sock; | |
1470 | ||
1471 | if (vs && IN_MULTICAST(ntohl(vxlan->default_dst.remote_ip)) && | |
1472 | ! vxlan_group_used(vn, vxlan->default_dst.remote_ip)) { | |
1473 | vxlan_sock_hold(vs); | |
1474 | dev_hold(dev); | |
1475 | queue_work(vxlan_wq, &vxlan->igmp_leave); | |
1476 | } | |
1477 | ||
1478 | del_timer_sync(&vxlan->age_timer); | |
1479 | ||
1480 | vxlan_flush(vxlan); | |
1481 | ||
1482 | return 0; | |
1483 | } | |
1484 | ||
1485 | /* Stub, nothing needs to be done. */ | |
1486 | static void vxlan_set_multicast_list(struct net_device *dev) | |
1487 | { | |
1488 | } | |
1489 | ||
1490 | static const struct net_device_ops vxlan_netdev_ops = { | |
1491 | .ndo_init = vxlan_init, | |
1492 | .ndo_uninit = vxlan_uninit, | |
1493 | .ndo_open = vxlan_open, | |
1494 | .ndo_stop = vxlan_stop, | |
1495 | .ndo_start_xmit = vxlan_xmit, | |
1496 | .ndo_get_stats64 = ip_tunnel_get_stats64, | |
1497 | .ndo_set_rx_mode = vxlan_set_multicast_list, | |
1498 | .ndo_change_mtu = eth_change_mtu, | |
1499 | .ndo_validate_addr = eth_validate_addr, | |
1500 | .ndo_set_mac_address = eth_mac_addr, | |
1501 | .ndo_fdb_add = vxlan_fdb_add, | |
1502 | .ndo_fdb_del = vxlan_fdb_delete, | |
1503 | .ndo_fdb_dump = vxlan_fdb_dump, | |
1504 | }; | |
1505 | ||
1506 | /* Info for udev, that this is a virtual tunnel endpoint */ | |
1507 | static struct device_type vxlan_type = { | |
1508 | .name = "vxlan", | |
1509 | }; | |
1510 | ||
1511 | /* Initialize the device structure. */ | |
1512 | static void vxlan_setup(struct net_device *dev) | |
1513 | { | |
1514 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1515 | unsigned int h; | |
1516 | int low, high; | |
1517 | ||
1518 | eth_hw_addr_random(dev); | |
1519 | ether_setup(dev); | |
1520 | dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM; | |
1521 | ||
1522 | dev->netdev_ops = &vxlan_netdev_ops; | |
1523 | dev->destructor = free_netdev; | |
1524 | SET_NETDEV_DEVTYPE(dev, &vxlan_type); | |
1525 | ||
1526 | dev->tx_queue_len = 0; | |
1527 | dev->features |= NETIF_F_LLTX; | |
1528 | dev->features |= NETIF_F_NETNS_LOCAL; | |
1529 | dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM; | |
1530 | dev->features |= NETIF_F_RXCSUM; | |
1531 | dev->features |= NETIF_F_GSO_SOFTWARE; | |
1532 | ||
1533 | dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM; | |
1534 | dev->hw_features |= NETIF_F_GSO_SOFTWARE; | |
1535 | dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; | |
1536 | dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; | |
1537 | ||
1538 | INIT_LIST_HEAD(&vxlan->next); | |
1539 | spin_lock_init(&vxlan->hash_lock); | |
1540 | INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join); | |
1541 | INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave); | |
1542 | INIT_WORK(&vxlan->sock_work, vxlan_sock_work); | |
1543 | ||
1544 | init_timer_deferrable(&vxlan->age_timer); | |
1545 | vxlan->age_timer.function = vxlan_cleanup; | |
1546 | vxlan->age_timer.data = (unsigned long) vxlan; | |
1547 | ||
1548 | inet_get_local_port_range(&low, &high); | |
1549 | vxlan->port_min = low; | |
1550 | vxlan->port_max = high; | |
1551 | vxlan->dst_port = htons(vxlan_port); | |
1552 | ||
1553 | vxlan->dev = dev; | |
1554 | ||
1555 | for (h = 0; h < FDB_HASH_SIZE; ++h) | |
1556 | INIT_HLIST_HEAD(&vxlan->fdb_head[h]); | |
1557 | } | |
1558 | ||
1559 | static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = { | |
1560 | [IFLA_VXLAN_ID] = { .type = NLA_U32 }, | |
1561 | [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) }, | |
1562 | [IFLA_VXLAN_LINK] = { .type = NLA_U32 }, | |
1563 | [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) }, | |
1564 | [IFLA_VXLAN_TOS] = { .type = NLA_U8 }, | |
1565 | [IFLA_VXLAN_TTL] = { .type = NLA_U8 }, | |
1566 | [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 }, | |
1567 | [IFLA_VXLAN_AGEING] = { .type = NLA_U32 }, | |
1568 | [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 }, | |
1569 | [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) }, | |
1570 | [IFLA_VXLAN_PROXY] = { .type = NLA_U8 }, | |
1571 | [IFLA_VXLAN_RSC] = { .type = NLA_U8 }, | |
1572 | [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 }, | |
1573 | [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 }, | |
1574 | [IFLA_VXLAN_PORT] = { .type = NLA_U16 }, | |
1575 | }; | |
1576 | ||
1577 | static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[]) | |
1578 | { | |
1579 | if (tb[IFLA_ADDRESS]) { | |
1580 | if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { | |
1581 | pr_debug("invalid link address (not ethernet)\n"); | |
1582 | return -EINVAL; | |
1583 | } | |
1584 | ||
1585 | if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { | |
1586 | pr_debug("invalid all zero ethernet address\n"); | |
1587 | return -EADDRNOTAVAIL; | |
1588 | } | |
1589 | } | |
1590 | ||
1591 | if (!data) | |
1592 | return -EINVAL; | |
1593 | ||
1594 | if (data[IFLA_VXLAN_ID]) { | |
1595 | __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]); | |
1596 | if (id >= VXLAN_VID_MASK) | |
1597 | return -ERANGE; | |
1598 | } | |
1599 | ||
1600 | if (data[IFLA_VXLAN_PORT_RANGE]) { | |
1601 | const struct ifla_vxlan_port_range *p | |
1602 | = nla_data(data[IFLA_VXLAN_PORT_RANGE]); | |
1603 | ||
1604 | if (ntohs(p->high) < ntohs(p->low)) { | |
1605 | pr_debug("port range %u .. %u not valid\n", | |
1606 | ntohs(p->low), ntohs(p->high)); | |
1607 | return -EINVAL; | |
1608 | } | |
1609 | } | |
1610 | ||
1611 | return 0; | |
1612 | } | |
1613 | ||
1614 | static void vxlan_get_drvinfo(struct net_device *netdev, | |
1615 | struct ethtool_drvinfo *drvinfo) | |
1616 | { | |
1617 | strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version)); | |
1618 | strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver)); | |
1619 | } | |
1620 | ||
1621 | static const struct ethtool_ops vxlan_ethtool_ops = { | |
1622 | .get_drvinfo = vxlan_get_drvinfo, | |
1623 | .get_link = ethtool_op_get_link, | |
1624 | }; | |
1625 | ||
1626 | static void vxlan_del_work(struct work_struct *work) | |
1627 | { | |
1628 | struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work); | |
1629 | ||
1630 | sk_release_kernel(vs->sock->sk); | |
1631 | kfree_rcu(vs, rcu); | |
1632 | } | |
1633 | ||
1634 | static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port) | |
1635 | { | |
1636 | struct vxlan_sock *vs; | |
1637 | struct sock *sk; | |
1638 | struct sockaddr_in vxlan_addr = { | |
1639 | .sin_family = AF_INET, | |
1640 | .sin_addr.s_addr = htonl(INADDR_ANY), | |
1641 | .sin_port = port, | |
1642 | }; | |
1643 | int rc; | |
1644 | unsigned int h; | |
1645 | ||
1646 | vs = kmalloc(sizeof(*vs), GFP_KERNEL); | |
1647 | if (!vs) | |
1648 | return ERR_PTR(-ENOMEM); | |
1649 | ||
1650 | for (h = 0; h < VNI_HASH_SIZE; ++h) | |
1651 | INIT_HLIST_HEAD(&vs->vni_list[h]); | |
1652 | ||
1653 | INIT_WORK(&vs->del_work, vxlan_del_work); | |
1654 | ||
1655 | /* Create UDP socket for encapsulation receive. */ | |
1656 | rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vs->sock); | |
1657 | if (rc < 0) { | |
1658 | pr_debug("UDP socket create failed\n"); | |
1659 | kfree(vs); | |
1660 | return ERR_PTR(rc); | |
1661 | } | |
1662 | ||
1663 | /* Put in proper namespace */ | |
1664 | sk = vs->sock->sk; | |
1665 | sk_change_net(sk, net); | |
1666 | ||
1667 | rc = kernel_bind(vs->sock, (struct sockaddr *) &vxlan_addr, | |
1668 | sizeof(vxlan_addr)); | |
1669 | if (rc < 0) { | |
1670 | pr_debug("bind for UDP socket %pI4:%u (%d)\n", | |
1671 | &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc); | |
1672 | sk_release_kernel(sk); | |
1673 | kfree(vs); | |
1674 | return ERR_PTR(rc); | |
1675 | } | |
1676 | ||
1677 | /* Disable multicast loopback */ | |
1678 | inet_sk(sk)->mc_loop = 0; | |
1679 | ||
1680 | /* Mark socket as an encapsulation socket. */ | |
1681 | udp_sk(sk)->encap_type = 1; | |
1682 | udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv; | |
1683 | udp_encap_enable(); | |
1684 | atomic_set(&vs->refcnt, 1); | |
1685 | ||
1686 | return vs; | |
1687 | } | |
1688 | ||
1689 | /* Scheduled at device creation to bind to a socket */ | |
1690 | static void vxlan_sock_work(struct work_struct *work) | |
1691 | { | |
1692 | struct vxlan_dev *vxlan | |
1693 | = container_of(work, struct vxlan_dev, sock_work); | |
1694 | struct net_device *dev = vxlan->dev; | |
1695 | struct net *net = dev_net(dev); | |
1696 | __u32 vni = vxlan->default_dst.remote_vni; | |
1697 | __be16 port = vxlan->dst_port; | |
1698 | struct vxlan_net *vn = net_generic(net, vxlan_net_id); | |
1699 | struct vxlan_sock *nvs, *ovs; | |
1700 | ||
1701 | nvs = vxlan_socket_create(net, port); | |
1702 | if (IS_ERR(nvs)) { | |
1703 | netdev_err(vxlan->dev, "Can not create UDP socket, %ld\n", | |
1704 | PTR_ERR(nvs)); | |
1705 | goto out; | |
1706 | } | |
1707 | ||
1708 | spin_lock(&vn->sock_lock); | |
1709 | /* Look again to see if can reuse socket */ | |
1710 | ovs = vxlan_find_port(net, port); | |
1711 | if (ovs) { | |
1712 | atomic_inc(&ovs->refcnt); | |
1713 | vxlan->vn_sock = ovs; | |
1714 | hlist_add_head_rcu(&vxlan->hlist, vni_head(ovs, vni)); | |
1715 | spin_unlock(&vn->sock_lock); | |
1716 | ||
1717 | sk_release_kernel(nvs->sock->sk); | |
1718 | kfree(nvs); | |
1719 | } else { | |
1720 | vxlan->vn_sock = nvs; | |
1721 | hlist_add_head_rcu(&nvs->hlist, vs_head(net, port)); | |
1722 | hlist_add_head_rcu(&vxlan->hlist, vni_head(nvs, vni)); | |
1723 | spin_unlock(&vn->sock_lock); | |
1724 | } | |
1725 | out: | |
1726 | dev_put(dev); | |
1727 | } | |
1728 | ||
1729 | static int vxlan_newlink(struct net *net, struct net_device *dev, | |
1730 | struct nlattr *tb[], struct nlattr *data[]) | |
1731 | { | |
1732 | struct vxlan_net *vn = net_generic(net, vxlan_net_id); | |
1733 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1734 | struct vxlan_rdst *dst = &vxlan->default_dst; | |
1735 | __u32 vni; | |
1736 | int err; | |
1737 | ||
1738 | if (!data[IFLA_VXLAN_ID]) | |
1739 | return -EINVAL; | |
1740 | ||
1741 | vni = nla_get_u32(data[IFLA_VXLAN_ID]); | |
1742 | dst->remote_vni = vni; | |
1743 | ||
1744 | if (data[IFLA_VXLAN_GROUP]) | |
1745 | dst->remote_ip = nla_get_be32(data[IFLA_VXLAN_GROUP]); | |
1746 | ||
1747 | if (data[IFLA_VXLAN_LOCAL]) | |
1748 | vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]); | |
1749 | ||
1750 | if (data[IFLA_VXLAN_LINK] && | |
1751 | (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) { | |
1752 | struct net_device *lowerdev | |
1753 | = __dev_get_by_index(net, dst->remote_ifindex); | |
1754 | ||
1755 | if (!lowerdev) { | |
1756 | pr_info("ifindex %d does not exist\n", dst->remote_ifindex); | |
1757 | return -ENODEV; | |
1758 | } | |
1759 | ||
1760 | if (!tb[IFLA_MTU]) | |
1761 | dev->mtu = lowerdev->mtu - VXLAN_HEADROOM; | |
1762 | ||
1763 | /* update header length based on lower device */ | |
1764 | dev->hard_header_len = lowerdev->hard_header_len + | |
1765 | VXLAN_HEADROOM; | |
1766 | } | |
1767 | ||
1768 | if (data[IFLA_VXLAN_TOS]) | |
1769 | vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]); | |
1770 | ||
1771 | if (data[IFLA_VXLAN_TTL]) | |
1772 | vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]); | |
1773 | ||
1774 | if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING])) | |
1775 | vxlan->flags |= VXLAN_F_LEARN; | |
1776 | ||
1777 | if (data[IFLA_VXLAN_AGEING]) | |
1778 | vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]); | |
1779 | else | |
1780 | vxlan->age_interval = FDB_AGE_DEFAULT; | |
1781 | ||
1782 | if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY])) | |
1783 | vxlan->flags |= VXLAN_F_PROXY; | |
1784 | ||
1785 | if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC])) | |
1786 | vxlan->flags |= VXLAN_F_RSC; | |
1787 | ||
1788 | if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS])) | |
1789 | vxlan->flags |= VXLAN_F_L2MISS; | |
1790 | ||
1791 | if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS])) | |
1792 | vxlan->flags |= VXLAN_F_L3MISS; | |
1793 | ||
1794 | if (data[IFLA_VXLAN_LIMIT]) | |
1795 | vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]); | |
1796 | ||
1797 | if (data[IFLA_VXLAN_PORT_RANGE]) { | |
1798 | const struct ifla_vxlan_port_range *p | |
1799 | = nla_data(data[IFLA_VXLAN_PORT_RANGE]); | |
1800 | vxlan->port_min = ntohs(p->low); | |
1801 | vxlan->port_max = ntohs(p->high); | |
1802 | } | |
1803 | ||
1804 | if (data[IFLA_VXLAN_PORT]) | |
1805 | vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]); | |
1806 | ||
1807 | if (vxlan_find_vni(net, vni, vxlan->dst_port)) { | |
1808 | pr_info("duplicate VNI %u\n", vni); | |
1809 | return -EEXIST; | |
1810 | } | |
1811 | ||
1812 | SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops); | |
1813 | ||
1814 | /* create an fdb entry for default destination */ | |
1815 | err = vxlan_fdb_create(vxlan, all_zeros_mac, | |
1816 | vxlan->default_dst.remote_ip, | |
1817 | NUD_REACHABLE|NUD_PERMANENT, | |
1818 | NLM_F_EXCL|NLM_F_CREATE, | |
1819 | vxlan->dst_port, vxlan->default_dst.remote_vni, | |
1820 | vxlan->default_dst.remote_ifindex, NTF_SELF); | |
1821 | if (err) | |
1822 | return err; | |
1823 | ||
1824 | err = register_netdevice(dev); | |
1825 | if (err) { | |
1826 | vxlan_fdb_delete_default(vxlan); | |
1827 | return err; | |
1828 | } | |
1829 | ||
1830 | list_add(&vxlan->next, &vn->vxlan_list); | |
1831 | ||
1832 | return 0; | |
1833 | } | |
1834 | ||
1835 | static void vxlan_dellink(struct net_device *dev, struct list_head *head) | |
1836 | { | |
1837 | struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id); | |
1838 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1839 | ||
1840 | flush_workqueue(vxlan_wq); | |
1841 | ||
1842 | spin_lock(&vn->sock_lock); | |
1843 | hlist_del_rcu(&vxlan->hlist); | |
1844 | spin_unlock(&vn->sock_lock); | |
1845 | ||
1846 | list_del(&vxlan->next); | |
1847 | unregister_netdevice_queue(dev, head); | |
1848 | } | |
1849 | ||
1850 | static size_t vxlan_get_size(const struct net_device *dev) | |
1851 | { | |
1852 | ||
1853 | return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */ | |
1854 | nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */ | |
1855 | nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */ | |
1856 | nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */ | |
1857 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */ | |
1858 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */ | |
1859 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */ | |
1860 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */ | |
1861 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */ | |
1862 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */ | |
1863 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */ | |
1864 | nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */ | |
1865 | nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */ | |
1866 | nla_total_size(sizeof(struct ifla_vxlan_port_range)) + | |
1867 | nla_total_size(sizeof(__be16))+ /* IFLA_VXLAN_PORT */ | |
1868 | 0; | |
1869 | } | |
1870 | ||
1871 | static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev) | |
1872 | { | |
1873 | const struct vxlan_dev *vxlan = netdev_priv(dev); | |
1874 | const struct vxlan_rdst *dst = &vxlan->default_dst; | |
1875 | struct ifla_vxlan_port_range ports = { | |
1876 | .low = htons(vxlan->port_min), | |
1877 | .high = htons(vxlan->port_max), | |
1878 | }; | |
1879 | ||
1880 | if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni)) | |
1881 | goto nla_put_failure; | |
1882 | ||
1883 | if (dst->remote_ip && nla_put_be32(skb, IFLA_VXLAN_GROUP, dst->remote_ip)) | |
1884 | goto nla_put_failure; | |
1885 | ||
1886 | if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex)) | |
1887 | goto nla_put_failure; | |
1888 | ||
1889 | if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr)) | |
1890 | goto nla_put_failure; | |
1891 | ||
1892 | if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) || | |
1893 | nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) || | |
1894 | nla_put_u8(skb, IFLA_VXLAN_LEARNING, | |
1895 | !!(vxlan->flags & VXLAN_F_LEARN)) || | |
1896 | nla_put_u8(skb, IFLA_VXLAN_PROXY, | |
1897 | !!(vxlan->flags & VXLAN_F_PROXY)) || | |
1898 | nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) || | |
1899 | nla_put_u8(skb, IFLA_VXLAN_L2MISS, | |
1900 | !!(vxlan->flags & VXLAN_F_L2MISS)) || | |
1901 | nla_put_u8(skb, IFLA_VXLAN_L3MISS, | |
1902 | !!(vxlan->flags & VXLAN_F_L3MISS)) || | |
1903 | nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) || | |
1904 | nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) || | |
1905 | nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port)) | |
1906 | goto nla_put_failure; | |
1907 | ||
1908 | if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports)) | |
1909 | goto nla_put_failure; | |
1910 | ||
1911 | return 0; | |
1912 | ||
1913 | nla_put_failure: | |
1914 | return -EMSGSIZE; | |
1915 | } | |
1916 | ||
1917 | static struct rtnl_link_ops vxlan_link_ops __read_mostly = { | |
1918 | .kind = "vxlan", | |
1919 | .maxtype = IFLA_VXLAN_MAX, | |
1920 | .policy = vxlan_policy, | |
1921 | .priv_size = sizeof(struct vxlan_dev), | |
1922 | .setup = vxlan_setup, | |
1923 | .validate = vxlan_validate, | |
1924 | .newlink = vxlan_newlink, | |
1925 | .dellink = vxlan_dellink, | |
1926 | .get_size = vxlan_get_size, | |
1927 | .fill_info = vxlan_fill_info, | |
1928 | }; | |
1929 | ||
1930 | static __net_init int vxlan_init_net(struct net *net) | |
1931 | { | |
1932 | struct vxlan_net *vn = net_generic(net, vxlan_net_id); | |
1933 | unsigned int h; | |
1934 | ||
1935 | INIT_LIST_HEAD(&vn->vxlan_list); | |
1936 | spin_lock_init(&vn->sock_lock); | |
1937 | ||
1938 | for (h = 0; h < PORT_HASH_SIZE; ++h) | |
1939 | INIT_HLIST_HEAD(&vn->sock_list[h]); | |
1940 | ||
1941 | return 0; | |
1942 | } | |
1943 | ||
1944 | static __net_exit void vxlan_exit_net(struct net *net) | |
1945 | { | |
1946 | struct vxlan_net *vn = net_generic(net, vxlan_net_id); | |
1947 | struct vxlan_dev *vxlan; | |
1948 | LIST_HEAD(list); | |
1949 | ||
1950 | rtnl_lock(); | |
1951 | list_for_each_entry(vxlan, &vn->vxlan_list, next) | |
1952 | unregister_netdevice_queue(vxlan->dev, &list); | |
1953 | unregister_netdevice_many(&list); | |
1954 | rtnl_unlock(); | |
1955 | } | |
1956 | ||
1957 | static struct pernet_operations vxlan_net_ops = { | |
1958 | .init = vxlan_init_net, | |
1959 | .exit = vxlan_exit_net, | |
1960 | .id = &vxlan_net_id, | |
1961 | .size = sizeof(struct vxlan_net), | |
1962 | }; | |
1963 | ||
1964 | static int __init vxlan_init_module(void) | |
1965 | { | |
1966 | int rc; | |
1967 | ||
1968 | vxlan_wq = alloc_workqueue("vxlan", 0, 0); | |
1969 | if (!vxlan_wq) | |
1970 | return -ENOMEM; | |
1971 | ||
1972 | get_random_bytes(&vxlan_salt, sizeof(vxlan_salt)); | |
1973 | ||
1974 | rc = register_pernet_device(&vxlan_net_ops); | |
1975 | if (rc) | |
1976 | goto out1; | |
1977 | ||
1978 | rc = rtnl_link_register(&vxlan_link_ops); | |
1979 | if (rc) | |
1980 | goto out2; | |
1981 | ||
1982 | return 0; | |
1983 | ||
1984 | out2: | |
1985 | unregister_pernet_device(&vxlan_net_ops); | |
1986 | out1: | |
1987 | destroy_workqueue(vxlan_wq); | |
1988 | return rc; | |
1989 | } | |
1990 | late_initcall(vxlan_init_module); | |
1991 | ||
1992 | static void __exit vxlan_cleanup_module(void) | |
1993 | { | |
1994 | rtnl_link_unregister(&vxlan_link_ops); | |
1995 | destroy_workqueue(vxlan_wq); | |
1996 | unregister_pernet_device(&vxlan_net_ops); | |
1997 | rcu_barrier(); | |
1998 | } | |
1999 | module_exit(vxlan_cleanup_module); | |
2000 | ||
2001 | MODULE_LICENSE("GPL"); | |
2002 | MODULE_VERSION(VXLAN_VERSION); | |
2003 | MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>"); | |
2004 | MODULE_ALIAS_RTNL_LINK("vxlan"); |