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