projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'timers-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[deliverable/linux.git] / drivers / net / vxlan.c
1 /*
2 * VXLAN: Virtual eXtensiable Local Area Network
3 *
4 * Copyright (c) 2012 Vyatta Inc.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 *
10 * TODO
11 * - use IANA UDP port number (when defined)
12 * - IPv6 (not in RFC)
13 */
14
15 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16
17 #include <linux/kernel.h>
18 #include <linux/types.h>
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/skbuff.h>
23 #include <linux/rculist.h>
24 #include <linux/netdevice.h>
25 #include <linux/in.h>
26 #include <linux/ip.h>
27 #include <linux/udp.h>
28 #include <linux/igmp.h>
29 #include <linux/etherdevice.h>
30 #include <linux/if_ether.h>
31 #include <linux/hash.h>
32 #include <net/ip.h>
33 #include <net/icmp.h>
34 #include <net/udp.h>
35 #include <net/rtnetlink.h>
36 #include <net/route.h>
37 #include <net/dsfield.h>
38 #include <net/inet_ecn.h>
39 #include <net/net_namespace.h>
40 #include <net/netns/generic.h>
41
42 #define VXLAN_VERSION "0.1"
43
44 #define VNI_HASH_BITS 10
45 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
46 #define FDB_HASH_BITS 8
47 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
48 #define FDB_AGE_DEFAULT 300 /* 5 min */
49 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
50
51 #define VXLAN_N_VID (1u << 24)
52 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
53 /* VLAN + IP header + UDP + VXLAN */
54 #define VXLAN_HEADROOM (4 + 20 + 8 + 8)
55
56 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
57
58 /* VXLAN protocol header */
59 struct vxlanhdr {
60 __be32 vx_flags;
61 __be32 vx_vni;
62 };
63
64 /* UDP port for VXLAN traffic. */
65 static unsigned int vxlan_port __read_mostly = 8472;
66 module_param_named(udp_port, vxlan_port, uint, 0444);
67 MODULE_PARM_DESC(udp_port, "Destination UDP port");
68
69 static bool log_ecn_error = true;
70 module_param(log_ecn_error, bool, 0644);
71 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
72
73 /* per-net private data for this module */
74 static unsigned int vxlan_net_id;
75 struct vxlan_net {
76 struct socket *sock; /* UDP encap socket */
77 struct hlist_head vni_list[VNI_HASH_SIZE];
78 };
79
80 /* Forwarding table entry */
81 struct vxlan_fdb {
82 struct hlist_node hlist; /* linked list of entries */
83 struct rcu_head rcu;
84 unsigned long updated; /* jiffies */
85 unsigned long used;
86 __be32 remote_ip;
87 u16 state; /* see ndm_state */
88 u8 eth_addr[ETH_ALEN];
89 };
90
91 /* Per-cpu network traffic stats */
92 struct vxlan_stats {
93 u64 rx_packets;
94 u64 rx_bytes;
95 u64 tx_packets;
96 u64 tx_bytes;
97 struct u64_stats_sync syncp;
98 };
99
100 /* Pseudo network device */
101 struct vxlan_dev {
102 struct hlist_node hlist;
103 struct net_device *dev;
104 struct vxlan_stats __percpu *stats;
105 __u32 vni; /* virtual network id */
106 __be32 gaddr; /* multicast group */
107 __be32 saddr; /* source address */
108 unsigned int link; /* link to multicast over */
109 __u8 tos; /* TOS override */
110 __u8 ttl;
111 bool learn;
112
113 unsigned long age_interval;
114 struct timer_list age_timer;
115 spinlock_t hash_lock;
116 unsigned int addrcnt;
117 unsigned int addrmax;
118 unsigned int addrexceeded;
119
120 struct hlist_head fdb_head[FDB_HASH_SIZE];
121 };
122
123 /* salt for hash table */
124 static u32 vxlan_salt __read_mostly;
125
126 static inline struct hlist_head *vni_head(struct net *net, u32 id)
127 {
128 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
129
130 return &vn->vni_list[hash_32(id, VNI_HASH_BITS)];
131 }
132
133 /* Look up VNI in a per net namespace table */
134 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
135 {
136 struct vxlan_dev *vxlan;
137 struct hlist_node *node;
138
139 hlist_for_each_entry_rcu(vxlan, node, vni_head(net, id), hlist) {
140 if (vxlan->vni == id)
141 return vxlan;
142 }
143
144 return NULL;
145 }
146
147 /* Fill in neighbour message in skbuff. */
148 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
149 const struct vxlan_fdb *fdb,
150 u32 portid, u32 seq, int type, unsigned int flags)
151 {
152 unsigned long now = jiffies;
153 struct nda_cacheinfo ci;
154 struct nlmsghdr *nlh;
155 struct ndmsg *ndm;
156
157 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
158 if (nlh == NULL)
159 return -EMSGSIZE;
160
161 ndm = nlmsg_data(nlh);
162 memset(ndm, 0, sizeof(*ndm));
163 ndm->ndm_family = AF_BRIDGE;
164 ndm->ndm_state = fdb->state;
165 ndm->ndm_ifindex = vxlan->dev->ifindex;
166 ndm->ndm_flags = NTF_SELF;
167 ndm->ndm_type = NDA_DST;
168
169 if (nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
170 goto nla_put_failure;
171
172 if (nla_put_be32(skb, NDA_DST, fdb->remote_ip))
173 goto nla_put_failure;
174
175 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
176 ci.ndm_confirmed = 0;
177 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
178 ci.ndm_refcnt = 0;
179
180 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
181 goto nla_put_failure;
182
183 return nlmsg_end(skb, nlh);
184
185 nla_put_failure:
186 nlmsg_cancel(skb, nlh);
187 return -EMSGSIZE;
188 }
189
190 static inline size_t vxlan_nlmsg_size(void)
191 {
192 return NLMSG_ALIGN(sizeof(struct ndmsg))
193 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
194 + nla_total_size(sizeof(__be32)) /* NDA_DST */
195 + nla_total_size(sizeof(struct nda_cacheinfo));
196 }
197
198 static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
199 const struct vxlan_fdb *fdb, int type)
200 {
201 struct net *net = dev_net(vxlan->dev);
202 struct sk_buff *skb;
203 int err = -ENOBUFS;
204
205 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
206 if (skb == NULL)
207 goto errout;
208
209 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0);
210 if (err < 0) {
211 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
212 WARN_ON(err == -EMSGSIZE);
213 kfree_skb(skb);
214 goto errout;
215 }
216
217 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
218 return;
219 errout:
220 if (err < 0)
221 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
222 }
223
224 /* Hash Ethernet address */
225 static u32 eth_hash(const unsigned char *addr)
226 {
227 u64 value = get_unaligned((u64 *)addr);
228
229 /* only want 6 bytes */
230 #ifdef __BIG_ENDIAN
231 value <<= 16;
232 #else
233 value >>= 16;
234 #endif
235 return hash_64(value, FDB_HASH_BITS);
236 }
237
238 /* Hash chain to use given mac address */
239 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
240 const u8 *mac)
241 {
242 return &vxlan->fdb_head[eth_hash(mac)];
243 }
244
245 /* Look up Ethernet address in forwarding table */
246 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
247 const u8 *mac)
248
249 {
250 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
251 struct vxlan_fdb *f;
252 struct hlist_node *node;
253
254 hlist_for_each_entry_rcu(f, node, head, hlist) {
255 if (compare_ether_addr(mac, f->eth_addr) == 0)
256 return f;
257 }
258
259 return NULL;
260 }
261
262 /* Add new entry to forwarding table -- assumes lock held */
263 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
264 const u8 *mac, __be32 ip,
265 __u16 state, __u16 flags)
266 {
267 struct vxlan_fdb *f;
268 int notify = 0;
269
270 f = vxlan_find_mac(vxlan, mac);
271 if (f) {
272 if (flags & NLM_F_EXCL) {
273 netdev_dbg(vxlan->dev,
274 "lost race to create %pM\n", mac);
275 return -EEXIST;
276 }
277 if (f->state != state) {
278 f->state = state;
279 f->updated = jiffies;
280 notify = 1;
281 }
282 } else {
283 if (!(flags & NLM_F_CREATE))
284 return -ENOENT;
285
286 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
287 return -ENOSPC;
288
289 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
290 f = kmalloc(sizeof(*f), GFP_ATOMIC);
291 if (!f)
292 return -ENOMEM;
293
294 notify = 1;
295 f->remote_ip = ip;
296 f->state = state;
297 f->updated = f->used = jiffies;
298 memcpy(f->eth_addr, mac, ETH_ALEN);
299
300 ++vxlan->addrcnt;
301 hlist_add_head_rcu(&f->hlist,
302 vxlan_fdb_head(vxlan, mac));
303 }
304
305 if (notify)
306 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
307
308 return 0;
309 }
310
311 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
312 {
313 netdev_dbg(vxlan->dev,
314 "delete %pM\n", f->eth_addr);
315
316 --vxlan->addrcnt;
317 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
318
319 hlist_del_rcu(&f->hlist);
320 kfree_rcu(f, rcu);
321 }
322
323 /* Add static entry (via netlink) */
324 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
325 struct net_device *dev,
326 const unsigned char *addr, u16 flags)
327 {
328 struct vxlan_dev *vxlan = netdev_priv(dev);
329 __be32 ip;
330 int err;
331
332 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
333 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
334 ndm->ndm_state);
335 return -EINVAL;
336 }
337
338 if (tb[NDA_DST] == NULL)
339 return -EINVAL;
340
341 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
342 return -EAFNOSUPPORT;
343
344 ip = nla_get_be32(tb[NDA_DST]);
345
346 spin_lock_bh(&vxlan->hash_lock);
347 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags);
348 spin_unlock_bh(&vxlan->hash_lock);
349
350 return err;
351 }
352
353 /* Delete entry (via netlink) */
354 static int vxlan_fdb_delete(struct ndmsg *ndm, struct net_device *dev,
355 const unsigned char *addr)
356 {
357 struct vxlan_dev *vxlan = netdev_priv(dev);
358 struct vxlan_fdb *f;
359 int err = -ENOENT;
360
361 spin_lock_bh(&vxlan->hash_lock);
362 f = vxlan_find_mac(vxlan, addr);
363 if (f) {
364 vxlan_fdb_destroy(vxlan, f);
365 err = 0;
366 }
367 spin_unlock_bh(&vxlan->hash_lock);
368
369 return err;
370 }
371
372 /* Dump forwarding table */
373 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
374 struct net_device *dev, int idx)
375 {
376 struct vxlan_dev *vxlan = netdev_priv(dev);
377 unsigned int h;
378
379 for (h = 0; h < FDB_HASH_SIZE; ++h) {
380 struct vxlan_fdb *f;
381 struct hlist_node *n;
382 int err;
383
384 hlist_for_each_entry_rcu(f, n, &vxlan->fdb_head[h], hlist) {
385 if (idx < cb->args[0])
386 goto skip;
387
388 err = vxlan_fdb_info(skb, vxlan, f,
389 NETLINK_CB(cb->skb).portid,
390 cb->nlh->nlmsg_seq,
391 RTM_NEWNEIGH,
392 NLM_F_MULTI);
393 if (err < 0)
394 break;
395 skip:
396 ++idx;
397 }
398 }
399
400 return idx;
401 }
402
403 /* Watch incoming packets to learn mapping between Ethernet address
404 * and Tunnel endpoint.
405 */
406 static void vxlan_snoop(struct net_device *dev,
407 __be32 src_ip, const u8 *src_mac)
408 {
409 struct vxlan_dev *vxlan = netdev_priv(dev);
410 struct vxlan_fdb *f;
411 int err;
412
413 f = vxlan_find_mac(vxlan, src_mac);
414 if (likely(f)) {
415 f->used = jiffies;
416 if (likely(f->remote_ip == src_ip))
417 return;
418
419 if (net_ratelimit())
420 netdev_info(dev,
421 "%pM migrated from %pI4 to %pI4\n",
422 src_mac, &f->remote_ip, &src_ip);
423
424 f->remote_ip = src_ip;
425 f->updated = jiffies;
426 } else {
427 /* learned new entry */
428 spin_lock(&vxlan->hash_lock);
429 err = vxlan_fdb_create(vxlan, src_mac, src_ip,
430 NUD_REACHABLE,
431 NLM_F_EXCL|NLM_F_CREATE);
432 spin_unlock(&vxlan->hash_lock);
433 }
434 }
435
436
437 /* See if multicast group is already in use by other ID */
438 static bool vxlan_group_used(struct vxlan_net *vn,
439 const struct vxlan_dev *this)
440 {
441 const struct vxlan_dev *vxlan;
442 struct hlist_node *node;
443 unsigned h;
444
445 for (h = 0; h < VNI_HASH_SIZE; ++h)
446 hlist_for_each_entry(vxlan, node, &vn->vni_list[h], hlist) {
447 if (vxlan == this)
448 continue;
449
450 if (!netif_running(vxlan->dev))
451 continue;
452
453 if (vxlan->gaddr == this->gaddr)
454 return true;
455 }
456
457 return false;
458 }
459
460 /* kernel equivalent to IP_ADD_MEMBERSHIP */
461 static int vxlan_join_group(struct net_device *dev)
462 {
463 struct vxlan_dev *vxlan = netdev_priv(dev);
464 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
465 struct sock *sk = vn->sock->sk;
466 struct ip_mreqn mreq = {
467 .imr_multiaddr.s_addr = vxlan->gaddr,
468 };
469 int err;
470
471 /* Already a member of group */
472 if (vxlan_group_used(vn, vxlan))
473 return 0;
474
475 /* Need to drop RTNL to call multicast join */
476 rtnl_unlock();
477 lock_sock(sk);
478 err = ip_mc_join_group(sk, &mreq);
479 release_sock(sk);
480 rtnl_lock();
481
482 return err;
483 }
484
485
486 /* kernel equivalent to IP_DROP_MEMBERSHIP */
487 static int vxlan_leave_group(struct net_device *dev)
488 {
489 struct vxlan_dev *vxlan = netdev_priv(dev);
490 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
491 int err = 0;
492 struct sock *sk = vn->sock->sk;
493 struct ip_mreqn mreq = {
494 .imr_multiaddr.s_addr = vxlan->gaddr,
495 };
496
497 /* Only leave group when last vxlan is done. */
498 if (vxlan_group_used(vn, vxlan))
499 return 0;
500
501 /* Need to drop RTNL to call multicast leave */
502 rtnl_unlock();
503 lock_sock(sk);
504 err = ip_mc_leave_group(sk, &mreq);
505 release_sock(sk);
506 rtnl_lock();
507
508 return err;
509 }
510
511 /* Callback from net/ipv4/udp.c to receive packets */
512 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
513 {
514 struct iphdr *oip;
515 struct vxlanhdr *vxh;
516 struct vxlan_dev *vxlan;
517 struct vxlan_stats *stats;
518 __u32 vni;
519 int err;
520
521 /* pop off outer UDP header */
522 __skb_pull(skb, sizeof(struct udphdr));
523
524 /* Need Vxlan and inner Ethernet header to be present */
525 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
526 goto error;
527
528 /* Drop packets with reserved bits set */
529 vxh = (struct vxlanhdr *) skb->data;
530 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
531 (vxh->vx_vni & htonl(0xff))) {
532 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
533 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
534 goto error;
535 }
536
537 __skb_pull(skb, sizeof(struct vxlanhdr));
538 skb_postpull_rcsum(skb, eth_hdr(skb), sizeof(struct vxlanhdr));
539
540 /* Is this VNI defined? */
541 vni = ntohl(vxh->vx_vni) >> 8;
542 vxlan = vxlan_find_vni(sock_net(sk), vni);
543 if (!vxlan) {
544 netdev_dbg(skb->dev, "unknown vni %d\n", vni);
545 goto drop;
546 }
547
548 if (!pskb_may_pull(skb, ETH_HLEN)) {
549 vxlan->dev->stats.rx_length_errors++;
550 vxlan->dev->stats.rx_errors++;
551 goto drop;
552 }
553
554 /* Re-examine inner Ethernet packet */
555 oip = ip_hdr(skb);
556 skb->protocol = eth_type_trans(skb, vxlan->dev);
557 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
558
559 /* Ignore packet loops (and multicast echo) */
560 if (compare_ether_addr(eth_hdr(skb)->h_source,
561 vxlan->dev->dev_addr) == 0)
562 goto drop;
563
564 if (vxlan->learn)
565 vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source);
566
567 __skb_tunnel_rx(skb, vxlan->dev);
568 skb_reset_network_header(skb);
569
570 err = IP_ECN_decapsulate(oip, skb);
571 if (unlikely(err)) {
572 if (log_ecn_error)
573 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
574 &oip->saddr, oip->tos);
575 if (err > 1) {
576 ++vxlan->dev->stats.rx_frame_errors;
577 ++vxlan->dev->stats.rx_errors;
578 goto drop;
579 }
580 }
581
582 stats = this_cpu_ptr(vxlan->stats);
583 u64_stats_update_begin(&stats->syncp);
584 stats->rx_packets++;
585 stats->rx_bytes += skb->len;
586 u64_stats_update_end(&stats->syncp);
587
588 netif_rx(skb);
589
590 return 0;
591 error:
592 /* Put UDP header back */
593 __skb_push(skb, sizeof(struct udphdr));
594
595 return 1;
596 drop:
597 /* Consume bad packet */
598 kfree_skb(skb);
599 return 0;
600 }
601
602 /* Extract dsfield from inner protocol */
603 static inline u8 vxlan_get_dsfield(const struct iphdr *iph,
604 const struct sk_buff *skb)
605 {
606 if (skb->protocol == htons(ETH_P_IP))
607 return iph->tos;
608 else if (skb->protocol == htons(ETH_P_IPV6))
609 return ipv6_get_dsfield((const struct ipv6hdr *)iph);
610 else
611 return 0;
612 }
613
614 /* Propogate ECN bits out */
615 static inline u8 vxlan_ecn_encap(u8 tos,
616 const struct iphdr *iph,
617 const struct sk_buff *skb)
618 {
619 u8 inner = vxlan_get_dsfield(iph, skb);
620
621 return INET_ECN_encapsulate(tos, inner);
622 }
623
624 /* Transmit local packets over Vxlan
625 *
626 * Outer IP header inherits ECN and DF from inner header.
627 * Outer UDP destination is the VXLAN assigned port.
628 * source port is based on hash of flow if available
629 * otherwise use a random value
630 */
631 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
632 {
633 struct vxlan_dev *vxlan = netdev_priv(dev);
634 struct rtable *rt;
635 const struct ethhdr *eth;
636 const struct iphdr *old_iph;
637 struct iphdr *iph;
638 struct vxlanhdr *vxh;
639 struct udphdr *uh;
640 struct flowi4 fl4;
641 struct vxlan_fdb *f;
642 unsigned int pkt_len = skb->len;
643 u32 hash;
644 __be32 dst;
645 __be16 df = 0;
646 __u8 tos, ttl;
647 int err;
648
649 /* Need space for new headers (invalidates iph ptr) */
650 if (skb_cow_head(skb, VXLAN_HEADROOM))
651 goto drop;
652
653 eth = (void *)skb->data;
654 old_iph = ip_hdr(skb);
655
656 if (!is_multicast_ether_addr(eth->h_dest) &&
657 (f = vxlan_find_mac(vxlan, eth->h_dest)))
658 dst = f->remote_ip;
659 else if (vxlan->gaddr) {
660 dst = vxlan->gaddr;
661 } else
662 goto drop;
663
664 ttl = vxlan->ttl;
665 if (!ttl && IN_MULTICAST(ntohl(dst)))
666 ttl = 1;
667
668 tos = vxlan->tos;
669 if (tos == 1)
670 tos = vxlan_get_dsfield(old_iph, skb);
671
672 hash = skb_get_rxhash(skb);
673
674 rt = ip_route_output_gre(dev_net(dev), &fl4, dst,
675 vxlan->saddr, vxlan->vni,
676 RT_TOS(tos), vxlan->link);
677 if (IS_ERR(rt)) {
678 netdev_dbg(dev, "no route to %pI4\n", &dst);
679 dev->stats.tx_carrier_errors++;
680 goto tx_error;
681 }
682
683 if (rt->dst.dev == dev) {
684 netdev_dbg(dev, "circular route to %pI4\n", &dst);
685 ip_rt_put(rt);
686 dev->stats.collisions++;
687 goto tx_error;
688 }
689
690 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
691 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
692 IPSKB_REROUTED);
693 skb_dst_drop(skb);
694 skb_dst_set(skb, &rt->dst);
695
696 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
697 vxh->vx_flags = htonl(VXLAN_FLAGS);
698 vxh->vx_vni = htonl(vxlan->vni << 8);
699
700 __skb_push(skb, sizeof(*uh));
701 skb_reset_transport_header(skb);
702 uh = udp_hdr(skb);
703
704 uh->dest = htons(vxlan_port);
705 uh->source = hash ? :random32();
706
707 uh->len = htons(skb->len);
708 uh->check = 0;
709
710 __skb_push(skb, sizeof(*iph));
711 skb_reset_network_header(skb);
712 iph = ip_hdr(skb);
713 iph->version = 4;
714 iph->ihl = sizeof(struct iphdr) >> 2;
715 iph->frag_off = df;
716 iph->protocol = IPPROTO_UDP;
717 iph->tos = vxlan_ecn_encap(tos, old_iph, skb);
718 iph->daddr = fl4.daddr;
719 iph->saddr = fl4.saddr;
720 iph->ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
721
722 /* See __IPTUNNEL_XMIT */
723 skb->ip_summed = CHECKSUM_NONE;
724 ip_select_ident(iph, &rt->dst, NULL);
725
726 err = ip_local_out(skb);
727 if (likely(net_xmit_eval(err) == 0)) {
728 struct vxlan_stats *stats = this_cpu_ptr(vxlan->stats);
729
730 u64_stats_update_begin(&stats->syncp);
731 stats->tx_packets++;
732 stats->tx_bytes += pkt_len;
733 u64_stats_update_end(&stats->syncp);
734 } else {
735 dev->stats.tx_errors++;
736 dev->stats.tx_aborted_errors++;
737 }
738 return NETDEV_TX_OK;
739
740 drop:
741 dev->stats.tx_dropped++;
742 goto tx_free;
743
744 tx_error:
745 dev->stats.tx_errors++;
746 tx_free:
747 dev_kfree_skb(skb);
748 return NETDEV_TX_OK;
749 }
750
751 /* Walk the forwarding table and purge stale entries */
752 static void vxlan_cleanup(unsigned long arg)
753 {
754 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
755 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
756 unsigned int h;
757
758 if (!netif_running(vxlan->dev))
759 return;
760
761 spin_lock_bh(&vxlan->hash_lock);
762 for (h = 0; h < FDB_HASH_SIZE; ++h) {
763 struct hlist_node *p, *n;
764 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
765 struct vxlan_fdb *f
766 = container_of(p, struct vxlan_fdb, hlist);
767 unsigned long timeout;
768
769 if (f->state == NUD_PERMANENT)
770 continue;
771
772 timeout = f->used + vxlan->age_interval * HZ;
773 if (time_before_eq(timeout, jiffies)) {
774 netdev_dbg(vxlan->dev,
775 "garbage collect %pM\n",
776 f->eth_addr);
777 f->state = NUD_STALE;
778 vxlan_fdb_destroy(vxlan, f);
779 } else if (time_before(timeout, next_timer))
780 next_timer = timeout;
781 }
782 }
783 spin_unlock_bh(&vxlan->hash_lock);
784
785 mod_timer(&vxlan->age_timer, next_timer);
786 }
787
788 /* Setup stats when device is created */
789 static int vxlan_init(struct net_device *dev)
790 {
791 struct vxlan_dev *vxlan = netdev_priv(dev);
792
793 vxlan->stats = alloc_percpu(struct vxlan_stats);
794 if (!vxlan->stats)
795 return -ENOMEM;
796
797 return 0;
798 }
799
800 /* Start ageing timer and join group when device is brought up */
801 static int vxlan_open(struct net_device *dev)
802 {
803 struct vxlan_dev *vxlan = netdev_priv(dev);
804 int err;
805
806 if (vxlan->gaddr) {
807 err = vxlan_join_group(dev);
808 if (err)
809 return err;
810 }
811
812 if (vxlan->age_interval)
813 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
814
815 return 0;
816 }
817
818 /* Purge the forwarding table */
819 static void vxlan_flush(struct vxlan_dev *vxlan)
820 {
821 unsigned h;
822
823 spin_lock_bh(&vxlan->hash_lock);
824 for (h = 0; h < FDB_HASH_SIZE; ++h) {
825 struct hlist_node *p, *n;
826 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
827 struct vxlan_fdb *f
828 = container_of(p, struct vxlan_fdb, hlist);
829 vxlan_fdb_destroy(vxlan, f);
830 }
831 }
832 spin_unlock_bh(&vxlan->hash_lock);
833 }
834
835 /* Cleanup timer and forwarding table on shutdown */
836 static int vxlan_stop(struct net_device *dev)
837 {
838 struct vxlan_dev *vxlan = netdev_priv(dev);
839
840 if (vxlan->gaddr)
841 vxlan_leave_group(dev);
842
843 del_timer_sync(&vxlan->age_timer);
844
845 vxlan_flush(vxlan);
846
847 return 0;
848 }
849
850 /* Merge per-cpu statistics */
851 static struct rtnl_link_stats64 *vxlan_stats64(struct net_device *dev,
852 struct rtnl_link_stats64 *stats)
853 {
854 struct vxlan_dev *vxlan = netdev_priv(dev);
855 struct vxlan_stats tmp, sum = { 0 };
856 unsigned int cpu;
857
858 for_each_possible_cpu(cpu) {
859 unsigned int start;
860 const struct vxlan_stats *stats
861 = per_cpu_ptr(vxlan->stats, cpu);
862
863 do {
864 start = u64_stats_fetch_begin_bh(&stats->syncp);
865 memcpy(&tmp, stats, sizeof(tmp));
866 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
867
868 sum.tx_bytes += tmp.tx_bytes;
869 sum.tx_packets += tmp.tx_packets;
870 sum.rx_bytes += tmp.rx_bytes;
871 sum.rx_packets += tmp.rx_packets;
872 }
873
874 stats->tx_bytes = sum.tx_bytes;
875 stats->tx_packets = sum.tx_packets;
876 stats->rx_bytes = sum.rx_bytes;
877 stats->rx_packets = sum.rx_packets;
878
879 stats->multicast = dev->stats.multicast;
880 stats->rx_length_errors = dev->stats.rx_length_errors;
881 stats->rx_frame_errors = dev->stats.rx_frame_errors;
882 stats->rx_errors = dev->stats.rx_errors;
883
884 stats->tx_dropped = dev->stats.tx_dropped;
885 stats->tx_carrier_errors = dev->stats.tx_carrier_errors;
886 stats->tx_aborted_errors = dev->stats.tx_aborted_errors;
887 stats->collisions = dev->stats.collisions;
888 stats->tx_errors = dev->stats.tx_errors;
889
890 return stats;
891 }
892
893 /* Stub, nothing needs to be done. */
894 static void vxlan_set_multicast_list(struct net_device *dev)
895 {
896 }
897
898 static const struct net_device_ops vxlan_netdev_ops = {
899 .ndo_init = vxlan_init,
900 .ndo_open = vxlan_open,
901 .ndo_stop = vxlan_stop,
902 .ndo_start_xmit = vxlan_xmit,
903 .ndo_get_stats64 = vxlan_stats64,
904 .ndo_set_rx_mode = vxlan_set_multicast_list,
905 .ndo_change_mtu = eth_change_mtu,
906 .ndo_validate_addr = eth_validate_addr,
907 .ndo_set_mac_address = eth_mac_addr,
908 .ndo_fdb_add = vxlan_fdb_add,
909 .ndo_fdb_del = vxlan_fdb_delete,
910 .ndo_fdb_dump = vxlan_fdb_dump,
911 };
912
913 /* Info for udev, that this is a virtual tunnel endpoint */
914 static struct device_type vxlan_type = {
915 .name = "vxlan",
916 };
917
918 static void vxlan_free(struct net_device *dev)
919 {
920 struct vxlan_dev *vxlan = netdev_priv(dev);
921
922 free_percpu(vxlan->stats);
923 free_netdev(dev);
924 }
925
926 /* Initialize the device structure. */
927 static void vxlan_setup(struct net_device *dev)
928 {
929 struct vxlan_dev *vxlan = netdev_priv(dev);
930 unsigned h;
931
932 eth_hw_addr_random(dev);
933 ether_setup(dev);
934
935 dev->netdev_ops = &vxlan_netdev_ops;
936 dev->destructor = vxlan_free;
937 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
938
939 dev->tx_queue_len = 0;
940 dev->features |= NETIF_F_LLTX;
941 dev->features |= NETIF_F_NETNS_LOCAL;
942 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
943
944 spin_lock_init(&vxlan->hash_lock);
945
946 init_timer_deferrable(&vxlan->age_timer);
947 vxlan->age_timer.function = vxlan_cleanup;
948 vxlan->age_timer.data = (unsigned long) vxlan;
949
950 vxlan->dev = dev;
951
952 for (h = 0; h < FDB_HASH_SIZE; ++h)
953 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
954 }
955
956 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
957 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
958 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
959 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
960 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
961 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
962 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
963 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
964 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
965 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
966 };
967
968 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
969 {
970 if (tb[IFLA_ADDRESS]) {
971 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
972 pr_debug("invalid link address (not ethernet)\n");
973 return -EINVAL;
974 }
975
976 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
977 pr_debug("invalid all zero ethernet address\n");
978 return -EADDRNOTAVAIL;
979 }
980 }
981
982 if (!data)
983 return -EINVAL;
984
985 if (data[IFLA_VXLAN_ID]) {
986 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
987 if (id >= VXLAN_VID_MASK)
988 return -ERANGE;
989 }
990
991 if (data[IFLA_VXLAN_GROUP]) {
992 __be32 gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
993 if (!IN_MULTICAST(ntohl(gaddr))) {
994 pr_debug("group address is not IPv4 multicast\n");
995 return -EADDRNOTAVAIL;
996 }
997 }
998 return 0;
999 }
1000
1001 static int vxlan_newlink(struct net *net, struct net_device *dev,
1002 struct nlattr *tb[], struct nlattr *data[])
1003 {
1004 struct vxlan_dev *vxlan = netdev_priv(dev);
1005 __u32 vni;
1006 int err;
1007
1008 if (!data[IFLA_VXLAN_ID])
1009 return -EINVAL;
1010
1011 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1012 if (vxlan_find_vni(net, vni)) {
1013 pr_info("duplicate VNI %u\n", vni);
1014 return -EEXIST;
1015 }
1016 vxlan->vni = vni;
1017
1018 if (data[IFLA_VXLAN_GROUP])
1019 vxlan->gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1020
1021 if (data[IFLA_VXLAN_LOCAL])
1022 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1023
1024 if (data[IFLA_VXLAN_LINK]) {
1025 vxlan->link = nla_get_u32(data[IFLA_VXLAN_LINK]);
1026
1027 if (!tb[IFLA_MTU]) {
1028 struct net_device *lowerdev;
1029 lowerdev = __dev_get_by_index(net, vxlan->link);
1030 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
1031 }
1032 }
1033
1034 if (data[IFLA_VXLAN_TOS])
1035 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1036
1037 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
1038 vxlan->learn = true;
1039
1040 if (data[IFLA_VXLAN_AGEING])
1041 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1042 else
1043 vxlan->age_interval = FDB_AGE_DEFAULT;
1044
1045 if (data[IFLA_VXLAN_LIMIT])
1046 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1047
1048 err = register_netdevice(dev);
1049 if (!err)
1050 hlist_add_head_rcu(&vxlan->hlist, vni_head(net, vxlan->vni));
1051
1052 return err;
1053 }
1054
1055 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1056 {
1057 struct vxlan_dev *vxlan = netdev_priv(dev);
1058
1059 hlist_del_rcu(&vxlan->hlist);
1060
1061 unregister_netdevice_queue(dev, head);
1062 }
1063
1064 static size_t vxlan_get_size(const struct net_device *dev)
1065 {
1066
1067 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1068 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1069 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1070 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1071 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1072 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1073 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
1074 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1075 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
1076 0;
1077 }
1078
1079 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1080 {
1081 const struct vxlan_dev *vxlan = netdev_priv(dev);
1082
1083 if (nla_put_u32(skb, IFLA_VXLAN_ID, vxlan->vni))
1084 goto nla_put_failure;
1085
1086 if (vxlan->gaddr && nla_put_be32(skb, IFLA_VXLAN_GROUP, vxlan->gaddr))
1087 goto nla_put_failure;
1088
1089 if (vxlan->link && nla_put_u32(skb, IFLA_VXLAN_LINK, vxlan->link))
1090 goto nla_put_failure;
1091
1092 if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
1093 goto nla_put_failure;
1094
1095 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1096 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
1097 nla_put_u8(skb, IFLA_VXLAN_LEARNING, vxlan->learn) ||
1098 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1099 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax))
1100 goto nla_put_failure;
1101
1102 return 0;
1103
1104 nla_put_failure:
1105 return -EMSGSIZE;
1106 }
1107
1108 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1109 .kind = "vxlan",
1110 .maxtype = IFLA_VXLAN_MAX,
1111 .policy = vxlan_policy,
1112 .priv_size = sizeof(struct vxlan_dev),
1113 .setup = vxlan_setup,
1114 .validate = vxlan_validate,
1115 .newlink = vxlan_newlink,
1116 .dellink = vxlan_dellink,
1117 .get_size = vxlan_get_size,
1118 .fill_info = vxlan_fill_info,
1119 };
1120
1121 static __net_init int vxlan_init_net(struct net *net)
1122 {
1123 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1124 struct sock *sk;
1125 struct sockaddr_in vxlan_addr = {
1126 .sin_family = AF_INET,
1127 .sin_addr.s_addr = htonl(INADDR_ANY),
1128 };
1129 int rc;
1130 unsigned h;
1131
1132 /* Create UDP socket for encapsulation receive. */
1133 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vn->sock);
1134 if (rc < 0) {
1135 pr_debug("UDP socket create failed\n");
1136 return rc;
1137 }
1138 /* Put in proper namespace */
1139 sk = vn->sock->sk;
1140 sk_change_net(sk, net);
1141
1142 vxlan_addr.sin_port = htons(vxlan_port);
1143
1144 rc = kernel_bind(vn->sock, (struct sockaddr *) &vxlan_addr,
1145 sizeof(vxlan_addr));
1146 if (rc < 0) {
1147 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1148 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
1149 sk_release_kernel(sk);
1150 vn->sock = NULL;
1151 return rc;
1152 }
1153
1154 /* Disable multicast loopback */
1155 inet_sk(sk)->mc_loop = 0;
1156
1157 /* Mark socket as an encapsulation socket. */
1158 udp_sk(sk)->encap_type = 1;
1159 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1160 udp_encap_enable();
1161
1162 for (h = 0; h < VNI_HASH_SIZE; ++h)
1163 INIT_HLIST_HEAD(&vn->vni_list[h]);
1164
1165 return 0;
1166 }
1167
1168 static __net_exit void vxlan_exit_net(struct net *net)
1169 {
1170 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1171
1172 if (vn->sock) {
1173 sk_release_kernel(vn->sock->sk);
1174 vn->sock = NULL;
1175 }
1176 }
1177
1178 static struct pernet_operations vxlan_net_ops = {
1179 .init = vxlan_init_net,
1180 .exit = vxlan_exit_net,
1181 .id = &vxlan_net_id,
1182 .size = sizeof(struct vxlan_net),
1183 };
1184
1185 static int __init vxlan_init_module(void)
1186 {
1187 int rc;
1188
1189 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1190
1191 rc = register_pernet_device(&vxlan_net_ops);
1192 if (rc)
1193 goto out1;
1194
1195 rc = rtnl_link_register(&vxlan_link_ops);
1196 if (rc)
1197 goto out2;
1198
1199 return 0;
1200
1201 out2:
1202 unregister_pernet_device(&vxlan_net_ops);
1203 out1:
1204 return rc;
1205 }
1206 module_init(vxlan_init_module);
1207
1208 static void __exit vxlan_cleanup_module(void)
1209 {
1210 rtnl_link_unregister(&vxlan_link_ops);
1211 unregister_pernet_device(&vxlan_net_ops);
1212 }
1213 module_exit(vxlan_cleanup_module);
1214
1215 MODULE_LICENSE("GPL");
1216 MODULE_VERSION(VXLAN_VERSION);
1217 MODULE_AUTHOR("Stephen Hemminger <shemminger@vyatta.com>");
1218 MODULE_ALIAS_RTNL_LINK("vxlan");
Linux kernel - Deliverables
This page took 0.05834 seconds and 6 git commands to generate.