?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
rbtree: optimize case selection logic in __rb_erase_color()
[linux-2.6/btrfs-unstable.git] / drivers / net / vxlan.c
blob51de9edb55f55d7eca5fe9a9956632ca5c403452
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/version.h>
32 #include <linux/hash.h>
33 #include <net/ip.h>
34 #include <net/icmp.h>
35 #include <net/udp.h>
36 #include <net/rtnetlink.h>
37 #include <net/route.h>
38 #include <net/dsfield.h>
39 #include <net/inet_ecn.h>
40 #include <net/net_namespace.h>
41 #include <net/netns/generic.h>
42
43 #define VXLAN_VERSION "0.1"
44
45 #define VNI_HASH_BITS 10
46 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
47 #define FDB_HASH_BITS 8
48 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
49 #define FDB_AGE_DEFAULT 300 /* 5 min */
50 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
51
52 #define VXLAN_N_VID (1u << 24)
53 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
54 /* VLAN + IP header + UDP + VXLAN */
55 #define VXLAN_HEADROOM (4 + 20 + 8 + 8)
56
57 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
58
59 /* VXLAN protocol header */
60 struct vxlanhdr {
61 __be32 vx_flags;
62 __be32 vx_vni;
63 };
64
65 /* UDP port for VXLAN traffic. */
66 static unsigned int vxlan_port __read_mostly = 8472;
67 module_param_named(udp_port, vxlan_port, uint, 0444);
68 MODULE_PARM_DESC(udp_port, "Destination UDP port");
69
70 static bool log_ecn_error = true;
71 module_param(log_ecn_error, bool, 0644);
72 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
73
74 /* per-net private data for this module */
75 static unsigned int vxlan_net_id;
76 struct vxlan_net {
77 struct socket *sock; /* UDP encap socket */
78 struct hlist_head vni_list[VNI_HASH_SIZE];
79 };
80
81 /* Forwarding table entry */
82 struct vxlan_fdb {
83 struct hlist_node hlist; /* linked list of entries */
84 struct rcu_head rcu;
85 unsigned long updated; /* jiffies */
86 unsigned long used;
87 __be32 remote_ip;
88 u16 state; /* see ndm_state */
89 u8 eth_addr[ETH_ALEN];
90 };
91
92 /* Per-cpu network traffic stats */
93 struct vxlan_stats {
94 u64 rx_packets;
95 u64 rx_bytes;
96 u64 tx_packets;
97 u64 tx_bytes;
98 struct u64_stats_sync syncp;
99 };
100
101 /* Pseudo network device */
102 struct vxlan_dev {
103 struct hlist_node hlist;
104 struct net_device *dev;
105 struct vxlan_stats __percpu *stats;
106 __u32 vni; /* virtual network id */
107 __be32 gaddr; /* multicast group */
108 __be32 saddr; /* source address */
109 unsigned int link; /* link to multicast over */
110 __u8 tos; /* TOS override */
111 __u8 ttl;
112 bool learn;
113
114 unsigned long age_interval;
115 struct timer_list age_timer;
116 spinlock_t hash_lock;
117 unsigned int addrcnt;
118 unsigned int addrmax;
119 unsigned int addrexceeded;
120
121 struct hlist_head fdb_head[FDB_HASH_SIZE];
122 };
123
124 /* salt for hash table */
125 static u32 vxlan_salt __read_mostly;
126
127 static inline struct hlist_head *vni_head(struct net *net, u32 id)
128 {
129 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
130
131 return &vn->vni_list[hash_32(id, VNI_HASH_BITS)];
132 }
133
134 /* Look up VNI in a per net namespace table */
135 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
136 {
137 struct vxlan_dev *vxlan;
138 struct hlist_node *node;
139
140 hlist_for_each_entry_rcu(vxlan, node, vni_head(net, id), hlist) {
141 if (vxlan->vni == id)
142 return vxlan;
143 }
144
145 return NULL;
146 }
147
148 /* Fill in neighbour message in skbuff. */
149 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
150 const struct vxlan_fdb *fdb,
151 u32 portid, u32 seq, int type, unsigned int flags)
152 {
153 unsigned long now = jiffies;
154 struct nda_cacheinfo ci;
155 struct nlmsghdr *nlh;
156 struct ndmsg *ndm;
157
158 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
159 if (nlh == NULL)
160 return -EMSGSIZE;
161
162 ndm = nlmsg_data(nlh);
163 memset(ndm, 0, sizeof(*ndm));
164 ndm->ndm_family = AF_BRIDGE;
165 ndm->ndm_state = fdb->state;
166 ndm->ndm_ifindex = vxlan->dev->ifindex;
167 ndm->ndm_flags = NTF_SELF;
168 ndm->ndm_type = NDA_DST;
169
170 if (nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
171 goto nla_put_failure;
172
173 if (nla_put_be32(skb, NDA_DST, fdb->remote_ip))
174 goto nla_put_failure;
175
176 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
177 ci.ndm_confirmed = 0;
178 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
179 ci.ndm_refcnt = 0;
180
181 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
182 goto nla_put_failure;
183
184 return nlmsg_end(skb, nlh);
185
186 nla_put_failure:
187 nlmsg_cancel(skb, nlh);
188 return -EMSGSIZE;
189 }
190
191 static inline size_t vxlan_nlmsg_size(void)
192 {
193 return NLMSG_ALIGN(sizeof(struct ndmsg))
194 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
195 + nla_total_size(sizeof(__be32)) /* NDA_DST */
196 + nla_total_size(sizeof(struct nda_cacheinfo));
197 }
198
199 static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
200 const struct vxlan_fdb *fdb, int type)
201 {
202 struct net *net = dev_net(vxlan->dev);
203 struct sk_buff *skb;
204 int err = -ENOBUFS;
205
206 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
207 if (skb == NULL)
208 goto errout;
209
210 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0);
211 if (err < 0) {
212 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
213 WARN_ON(err == -EMSGSIZE);
214 kfree_skb(skb);
215 goto errout;
216 }
217
218 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
219 return;
220 errout:
221 if (err < 0)
222 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
223 }
224
225 /* Hash Ethernet address */
226 static u32 eth_hash(const unsigned char *addr)
227 {
228 u64 value = get_unaligned((u64 *)addr);
229
230 /* only want 6 bytes */
231 #ifdef __BIG_ENDIAN
232 value <<= 16;
233 #else
234 value >>= 16;
235 #endif
236 return hash_64(value, FDB_HASH_BITS);
237 }
238
239 /* Hash chain to use given mac address */
240 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
241 const u8 *mac)
242 {
243 return &vxlan->fdb_head[eth_hash(mac)];
244 }
245
246 /* Look up Ethernet address in forwarding table */
247 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
248 const u8 *mac)
249
250 {
251 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
252 struct vxlan_fdb *f;
253 struct hlist_node *node;
254
255 hlist_for_each_entry_rcu(f, node, head, hlist) {
256 if (compare_ether_addr(mac, f->eth_addr) == 0)
257 return f;
258 }
259
260 return NULL;
261 }
262
263 /* Add new entry to forwarding table -- assumes lock held */
264 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
265 const u8 *mac, __be32 ip,
266 __u16 state, __u16 flags)
267 {
268 struct vxlan_fdb *f;
269 int notify = 0;
270
271 f = vxlan_find_mac(vxlan, mac);
272 if (f) {
273 if (flags & NLM_F_EXCL) {
274 netdev_dbg(vxlan->dev,
275 "lost race to create %pM\n", mac);
276 return -EEXIST;
277 }
278 if (f->state != state) {
279 f->state = state;
280 f->updated = jiffies;
281 notify = 1;
282 }
283 } else {
284 if (!(flags & NLM_F_CREATE))
285 return -ENOENT;
286
287 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
288 return -ENOSPC;
289
290 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
291 f = kmalloc(sizeof(*f), GFP_ATOMIC);
292 if (!f)
293 return -ENOMEM;
294
295 notify = 1;
296 f->remote_ip = ip;
297 f->state = state;
298 f->updated = f->used = jiffies;
299 memcpy(f->eth_addr, mac, ETH_ALEN);
300
301 ++vxlan->addrcnt;
302 hlist_add_head_rcu(&f->hlist,
303 vxlan_fdb_head(vxlan, mac));
304 }
305
306 if (notify)
307 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
308
309 return 0;
310 }
311
312 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
313 {
314 netdev_dbg(vxlan->dev,
315 "delete %pM\n", f->eth_addr);
316
317 --vxlan->addrcnt;
318 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
319
320 hlist_del_rcu(&f->hlist);
321 kfree_rcu(f, rcu);
322 }
323
324 /* Add static entry (via netlink) */
325 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
326 struct net_device *dev,
327 const unsigned char *addr, u16 flags)
328 {
329 struct vxlan_dev *vxlan = netdev_priv(dev);
330 __be32 ip;
331 int err;
332
333 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
334 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
335 ndm->ndm_state);
336 return -EINVAL;
337 }
338
339 if (tb[NDA_DST] == NULL)
340 return -EINVAL;
341
342 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
343 return -EAFNOSUPPORT;
344
345 ip = nla_get_be32(tb[NDA_DST]);
346
347 spin_lock_bh(&vxlan->hash_lock);
348 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags);
349 spin_unlock_bh(&vxlan->hash_lock);
350
351 return err;
352 }
353
354 /* Delete entry (via netlink) */
355 static int vxlan_fdb_delete(struct ndmsg *ndm, struct net_device *dev,
356 const unsigned char *addr)
357 {
358 struct vxlan_dev *vxlan = netdev_priv(dev);
359 struct vxlan_fdb *f;
360 int err = -ENOENT;
361
362 spin_lock_bh(&vxlan->hash_lock);
363 f = vxlan_find_mac(vxlan, addr);
364 if (f) {
365 vxlan_fdb_destroy(vxlan, f);
366 err = 0;
367 }
368 spin_unlock_bh(&vxlan->hash_lock);
369
370 return err;
371 }
372
373 /* Dump forwarding table */
374 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
375 struct net_device *dev, int idx)
376 {
377 struct vxlan_dev *vxlan = netdev_priv(dev);
378 unsigned int h;
379
380 for (h = 0; h < FDB_HASH_SIZE; ++h) {
381 struct vxlan_fdb *f;
382 struct hlist_node *n;
383 int err;
384
385 hlist_for_each_entry_rcu(f, n, &vxlan->fdb_head[h], hlist) {
386 if (idx < cb->args[0])
387 goto skip;
388
389 err = vxlan_fdb_info(skb, vxlan, f,
390 NETLINK_CB(cb->skb).portid,
391 cb->nlh->nlmsg_seq,
392 RTM_NEWNEIGH,
393 NLM_F_MULTI);
394 if (err < 0)
395 break;
396 skip:
397 ++idx;
398 }
399 }
400
401 return idx;
402 }
403
404 /* Watch incoming packets to learn mapping between Ethernet address
405 * and Tunnel endpoint.
406 */
407 static void vxlan_snoop(struct net_device *dev,
408 __be32 src_ip, const u8 *src_mac)
409 {
410 struct vxlan_dev *vxlan = netdev_priv(dev);
411 struct vxlan_fdb *f;
412 int err;
413
414 f = vxlan_find_mac(vxlan, src_mac);
415 if (likely(f)) {
416 f->used = jiffies;
417 if (likely(f->remote_ip == src_ip))
418 return;
419
420 if (net_ratelimit())
421 netdev_info(dev,
422 "%pM migrated from %pI4 to %pI4\n",
423 src_mac, &f->remote_ip, &src_ip);
424
425 f->remote_ip = src_ip;
426 f->updated = jiffies;
427 } else {
428 /* learned new entry */
429 spin_lock(&vxlan->hash_lock);
430 err = vxlan_fdb_create(vxlan, src_mac, src_ip,
431 NUD_REACHABLE,
432 NLM_F_EXCL|NLM_F_CREATE);
433 spin_unlock(&vxlan->hash_lock);
434 }
435 }
436
437
438 /* See if multicast group is already in use by other ID */
439 static bool vxlan_group_used(struct vxlan_net *vn,
440 const struct vxlan_dev *this)
441 {
442 const struct vxlan_dev *vxlan;
443 struct hlist_node *node;
444 unsigned h;
445
446 for (h = 0; h < VNI_HASH_SIZE; ++h)
447 hlist_for_each_entry(vxlan, node, &vn->vni_list[h], hlist) {
448 if (vxlan == this)
449 continue;
450
451 if (!netif_running(vxlan->dev))
452 continue;
453
454 if (vxlan->gaddr == this->gaddr)
455 return true;
456 }
457
458 return false;
459 }
460
461 /* kernel equivalent to IP_ADD_MEMBERSHIP */
462 static int vxlan_join_group(struct net_device *dev)
463 {
464 struct vxlan_dev *vxlan = netdev_priv(dev);
465 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
466 struct sock *sk = vn->sock->sk;
467 struct ip_mreqn mreq = {
468 .imr_multiaddr.s_addr = vxlan->gaddr,
469 };
470 int err;
471
472 /* Already a member of group */
473 if (vxlan_group_used(vn, vxlan))
474 return 0;
475
476 /* Need to drop RTNL to call multicast join */
477 rtnl_unlock();
478 lock_sock(sk);
479 err = ip_mc_join_group(sk, &mreq);
480 release_sock(sk);
481 rtnl_lock();
482
483 return err;
484 }
485
486
487 /* kernel equivalent to IP_DROP_MEMBERSHIP */
488 static int vxlan_leave_group(struct net_device *dev)
489 {
490 struct vxlan_dev *vxlan = netdev_priv(dev);
491 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
492 int err = 0;
493 struct sock *sk = vn->sock->sk;
494 struct ip_mreqn mreq = {
495 .imr_multiaddr.s_addr = vxlan->gaddr,
496 };
497
498 /* Only leave group when last vxlan is done. */
499 if (vxlan_group_used(vn, vxlan))
500 return 0;
501
502 /* Need to drop RTNL to call multicast leave */
503 rtnl_unlock();
504 lock_sock(sk);
505 err = ip_mc_leave_group(sk, &mreq);
506 release_sock(sk);
507 rtnl_lock();
508
509 return err;
510 }
511
512 /* Callback from net/ipv4/udp.c to receive packets */
513 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
514 {
515 struct iphdr *oip;
516 struct vxlanhdr *vxh;
517 struct vxlan_dev *vxlan;
518 struct vxlan_stats *stats;
519 __u32 vni;
520 int err;
521
522 /* pop off outer UDP header */
523 __skb_pull(skb, sizeof(struct udphdr));
524
525 /* Need Vxlan and inner Ethernet header to be present */
526 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
527 goto error;
528
529 /* Drop packets with reserved bits set */
530 vxh = (struct vxlanhdr *) skb->data;
531 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
532 (vxh->vx_vni & htonl(0xff))) {
533 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
534 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
535 goto error;
536 }
537
538 __skb_pull(skb, sizeof(struct vxlanhdr));
539 skb_postpull_rcsum(skb, eth_hdr(skb), sizeof(struct vxlanhdr));
540
541 /* Is this VNI defined? */
542 vni = ntohl(vxh->vx_vni) >> 8;
543 vxlan = vxlan_find_vni(sock_net(sk), vni);
544 if (!vxlan) {
545 netdev_dbg(skb->dev, "unknown vni %d\n", vni);
546 goto drop;
547 }
548
549 if (!pskb_may_pull(skb, ETH_HLEN)) {
550 vxlan->dev->stats.rx_length_errors++;
551 vxlan->dev->stats.rx_errors++;
552 goto drop;
553 }
554
555 /* Re-examine inner Ethernet packet */
556 oip = ip_hdr(skb);
557 skb->protocol = eth_type_trans(skb, vxlan->dev);
558 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
559
560 /* Ignore packet loops (and multicast echo) */
561 if (compare_ether_addr(eth_hdr(skb)->h_source,
562 vxlan->dev->dev_addr) == 0)
563 goto drop;
564
565 if (vxlan->learn)
566 vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source);
567
568 __skb_tunnel_rx(skb, vxlan->dev);
569 skb_reset_network_header(skb);
570
571 err = IP_ECN_decapsulate(oip, skb);
572 if (unlikely(err)) {
573 if (log_ecn_error)
574 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
575 &oip->saddr, oip->tos);
576 if (err > 1) {
577 ++vxlan->dev->stats.rx_frame_errors;
578 ++vxlan->dev->stats.rx_errors;
579 goto drop;
580 }
581 }
582
583 stats = this_cpu_ptr(vxlan->stats);
584 u64_stats_update_begin(&stats->syncp);
585 stats->rx_packets++;
586 stats->rx_bytes += skb->len;
587 u64_stats_update_end(&stats->syncp);
588
589 netif_rx(skb);
590
591 return 0;
592 error:
593 /* Put UDP header back */
594 __skb_push(skb, sizeof(struct udphdr));
595
596 return 1;
597 drop:
598 /* Consume bad packet */
599 kfree_skb(skb);
600 return 0;
601 }
602
603 /* Extract dsfield from inner protocol */
604 static inline u8 vxlan_get_dsfield(const struct iphdr *iph,
605 const struct sk_buff *skb)
606 {
607 if (skb->protocol == htons(ETH_P_IP))
608 return iph->tos;
609 else if (skb->protocol == htons(ETH_P_IPV6))
610 return ipv6_get_dsfield((const struct ipv6hdr *)iph);
611 else
612 return 0;
613 }
614
615 /* Propogate ECN bits out */
616 static inline u8 vxlan_ecn_encap(u8 tos,
617 const struct iphdr *iph,
618 const struct sk_buff *skb)
619 {
620 u8 inner = vxlan_get_dsfield(iph, skb);
621
622 return INET_ECN_encapsulate(tos, inner);
623 }
624
625 /* Transmit local packets over Vxlan
626 *
627 * Outer IP header inherits ECN and DF from inner header.
628 * Outer UDP destination is the VXLAN assigned port.
629 * source port is based on hash of flow if available
630 * otherwise use a random value
631 */
632 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
633 {
634 struct vxlan_dev *vxlan = netdev_priv(dev);
635 struct rtable *rt;
636 const struct ethhdr *eth;
637 const struct iphdr *old_iph;
638 struct iphdr *iph;
639 struct vxlanhdr *vxh;
640 struct udphdr *uh;
641 struct flowi4 fl4;
642 struct vxlan_fdb *f;
643 unsigned int pkt_len = skb->len;
644 u32 hash;
645 __be32 dst;
646 __be16 df = 0;
647 __u8 tos, ttl;
648 int err;
649
650 /* Need space for new headers (invalidates iph ptr) */
651 if (skb_cow_head(skb, VXLAN_HEADROOM))
652 goto drop;
653
654 eth = (void *)skb->data;
655 old_iph = ip_hdr(skb);
656
657 if (!is_multicast_ether_addr(eth->h_dest) &&
658 (f = vxlan_find_mac(vxlan, eth->h_dest)))
659 dst = f->remote_ip;
660 else if (vxlan->gaddr) {
661 dst = vxlan->gaddr;
662 } else
663 goto drop;
664
665 ttl = vxlan->ttl;
666 if (!ttl && IN_MULTICAST(ntohl(dst)))
667 ttl = 1;
668
669 tos = vxlan->tos;
670 if (tos == 1)
671 tos = vxlan_get_dsfield(old_iph, skb);
672
673 hash = skb_get_rxhash(skb);
674
675 rt = ip_route_output_gre(dev_net(dev), &fl4, dst,
676 vxlan->saddr, vxlan->vni,
677 RT_TOS(tos), vxlan->link);
678 if (IS_ERR(rt)) {
679 netdev_dbg(dev, "no route to %pI4\n", &dst);
680 dev->stats.tx_carrier_errors++;
681 goto tx_error;
682 }
683
684 if (rt->dst.dev == dev) {
685 netdev_dbg(dev, "circular route to %pI4\n", &dst);
686 ip_rt_put(rt);
687 dev->stats.collisions++;
688 goto tx_error;
689 }
690
691 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
692 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
693 IPSKB_REROUTED);
694 skb_dst_drop(skb);
695 skb_dst_set(skb, &rt->dst);
696
697 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
698 vxh->vx_flags = htonl(VXLAN_FLAGS);
699 vxh->vx_vni = htonl(vxlan->vni << 8);
700
701 __skb_push(skb, sizeof(*uh));
702 skb_reset_transport_header(skb);
703 uh = udp_hdr(skb);
704
705 uh->dest = htons(vxlan_port);
706 uh->source = hash ? :random32();
707
708 uh->len = htons(skb->len);
709 uh->check = 0;
710
711 __skb_push(skb, sizeof(*iph));
712 skb_reset_network_header(skb);
713 iph = ip_hdr(skb);
714 iph->version = 4;
715 iph->ihl = sizeof(struct iphdr) >> 2;
716 iph->frag_off = df;
717 iph->protocol = IPPROTO_UDP;
718 iph->tos = vxlan_ecn_encap(tos, old_iph, skb);
719 iph->daddr = fl4.daddr;
720 iph->saddr = fl4.saddr;
721 iph->ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
722
723 /* See __IPTUNNEL_XMIT */
724 skb->ip_summed = CHECKSUM_NONE;
725 ip_select_ident(iph, &rt->dst, NULL);
726
727 err = ip_local_out(skb);
728 if (likely(net_xmit_eval(err) == 0)) {
729 struct vxlan_stats *stats = this_cpu_ptr(vxlan->stats);
730
731 u64_stats_update_begin(&stats->syncp);
732 stats->tx_packets++;
733 stats->tx_bytes += pkt_len;
734 u64_stats_update_end(&stats->syncp);
735 } else {
736 dev->stats.tx_errors++;
737 dev->stats.tx_aborted_errors++;
738 }
739 return NETDEV_TX_OK;
740
741 drop:
742 dev->stats.tx_dropped++;
743 goto tx_free;
744
745 tx_error:
746 dev->stats.tx_errors++;
747 tx_free:
748 dev_kfree_skb(skb);
749 return NETDEV_TX_OK;
750 }
751
752 /* Walk the forwarding table and purge stale entries */
753 static void vxlan_cleanup(unsigned long arg)
754 {
755 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
756 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
757 unsigned int h;
758
759 if (!netif_running(vxlan->dev))
760 return;
761
762 spin_lock_bh(&vxlan->hash_lock);
763 for (h = 0; h < FDB_HASH_SIZE; ++h) {
764 struct hlist_node *p, *n;
765 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
766 struct vxlan_fdb *f
767 = container_of(p, struct vxlan_fdb, hlist);
768 unsigned long timeout;
769
770 if (f->state == NUD_PERMANENT)
771 continue;
772
773 timeout = f->used + vxlan->age_interval * HZ;
774 if (time_before_eq(timeout, jiffies)) {
775 netdev_dbg(vxlan->dev,
776 "garbage collect %pM\n",
777 f->eth_addr);
778 f->state = NUD_STALE;
779 vxlan_fdb_destroy(vxlan, f);
780 } else if (time_before(timeout, next_timer))
781 next_timer = timeout;
782 }
783 }
784 spin_unlock_bh(&vxlan->hash_lock);
785
786 mod_timer(&vxlan->age_timer, next_timer);
787 }
788
789 /* Setup stats when device is created */
790 static int vxlan_init(struct net_device *dev)
791 {
792 struct vxlan_dev *vxlan = netdev_priv(dev);
793
794 vxlan->stats = alloc_percpu(struct vxlan_stats);
795 if (!vxlan->stats)
796 return -ENOMEM;
797
798 return 0;
799 }
800
801 /* Start ageing timer and join group when device is brought up */
802 static int vxlan_open(struct net_device *dev)
803 {
804 struct vxlan_dev *vxlan = netdev_priv(dev);
805 int err;
806
807 if (vxlan->gaddr) {
808 err = vxlan_join_group(dev);
809 if (err)
810 return err;
811 }
812
813 if (vxlan->age_interval)
814 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
815
816 return 0;
817 }
818
819 /* Purge the forwarding table */
820 static void vxlan_flush(struct vxlan_dev *vxlan)
821 {
822 unsigned h;
823
824 spin_lock_bh(&vxlan->hash_lock);
825 for (h = 0; h < FDB_HASH_SIZE; ++h) {
826 struct hlist_node *p, *n;
827 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
828 struct vxlan_fdb *f
829 = container_of(p, struct vxlan_fdb, hlist);
830 vxlan_fdb_destroy(vxlan, f);
831 }
832 }
833 spin_unlock_bh(&vxlan->hash_lock);
834 }
835
836 /* Cleanup timer and forwarding table on shutdown */
837 static int vxlan_stop(struct net_device *dev)
838 {
839 struct vxlan_dev *vxlan = netdev_priv(dev);
840
841 if (vxlan->gaddr)
842 vxlan_leave_group(dev);
843
844 del_timer_sync(&vxlan->age_timer);
845
846 vxlan_flush(vxlan);
847
848 return 0;
849 }
850
851 /* Merge per-cpu statistics */
852 static struct rtnl_link_stats64 *vxlan_stats64(struct net_device *dev,
853 struct rtnl_link_stats64 *stats)
854 {
855 struct vxlan_dev *vxlan = netdev_priv(dev);
856 struct vxlan_stats tmp, sum = { 0 };
857 unsigned int cpu;
858
859 for_each_possible_cpu(cpu) {
860 unsigned int start;
861 const struct vxlan_stats *stats
862 = per_cpu_ptr(vxlan->stats, cpu);
863
864 do {
865 start = u64_stats_fetch_begin_bh(&stats->syncp);
866 memcpy(&tmp, stats, sizeof(tmp));
867 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
868
869 sum.tx_bytes += tmp.tx_bytes;
870 sum.tx_packets += tmp.tx_packets;
871 sum.rx_bytes += tmp.rx_bytes;
872 sum.rx_packets += tmp.rx_packets;
873 }
874
875 stats->tx_bytes = sum.tx_bytes;
876 stats->tx_packets = sum.tx_packets;
877 stats->rx_bytes = sum.rx_bytes;
878 stats->rx_packets = sum.rx_packets;
879
880 stats->multicast = dev->stats.multicast;
881 stats->rx_length_errors = dev->stats.rx_length_errors;
882 stats->rx_frame_errors = dev->stats.rx_frame_errors;
883 stats->rx_errors = dev->stats.rx_errors;
884
885 stats->tx_dropped = dev->stats.tx_dropped;
886 stats->tx_carrier_errors = dev->stats.tx_carrier_errors;
887 stats->tx_aborted_errors = dev->stats.tx_aborted_errors;
888 stats->collisions = dev->stats.collisions;
889 stats->tx_errors = dev->stats.tx_errors;
890
891 return stats;
892 }
893
894 /* Stub, nothing needs to be done. */
895 static void vxlan_set_multicast_list(struct net_device *dev)
896 {
897 }
898
899 static const struct net_device_ops vxlan_netdev_ops = {
900 .ndo_init = vxlan_init,
901 .ndo_open = vxlan_open,
902 .ndo_stop = vxlan_stop,
903 .ndo_start_xmit = vxlan_xmit,
904 .ndo_get_stats64 = vxlan_stats64,
905 .ndo_set_rx_mode = vxlan_set_multicast_list,
906 .ndo_change_mtu = eth_change_mtu,
907 .ndo_validate_addr = eth_validate_addr,
908 .ndo_set_mac_address = eth_mac_addr,
909 .ndo_fdb_add = vxlan_fdb_add,
910 .ndo_fdb_del = vxlan_fdb_delete,
911 .ndo_fdb_dump = vxlan_fdb_dump,
912 };
913
914 /* Info for udev, that this is a virtual tunnel endpoint */
915 static struct device_type vxlan_type = {
916 .name = "vxlan",
917 };
918
919 static void vxlan_free(struct net_device *dev)
920 {
921 struct vxlan_dev *vxlan = netdev_priv(dev);
922
923 free_percpu(vxlan->stats);
924 free_netdev(dev);
925 }
926
927 /* Initialize the device structure. */
928 static void vxlan_setup(struct net_device *dev)
929 {
930 struct vxlan_dev *vxlan = netdev_priv(dev);
931 unsigned h;
932
933 eth_hw_addr_random(dev);
934 ether_setup(dev);
935
936 dev->netdev_ops = &vxlan_netdev_ops;
937 dev->destructor = vxlan_free;
938 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
939
940 dev->tx_queue_len = 0;
941 dev->features |= NETIF_F_LLTX;
942 dev->features |= NETIF_F_NETNS_LOCAL;
943 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
944
945 spin_lock_init(&vxlan->hash_lock);
946
947 init_timer_deferrable(&vxlan->age_timer);
948 vxlan->age_timer.function = vxlan_cleanup;
949 vxlan->age_timer.data = (unsigned long) vxlan;
950
951 vxlan->dev = dev;
952
953 for (h = 0; h < FDB_HASH_SIZE; ++h)
954 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
955 }
956
957 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
958 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
959 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
960 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
961 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
962 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
963 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
964 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
965 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
966 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
967 };
968
969 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
970 {
971 if (tb[IFLA_ADDRESS]) {
972 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
973 pr_debug("invalid link address (not ethernet)\n");
974 return -EINVAL;
975 }
976
977 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
978 pr_debug("invalid all zero ethernet address\n");
979 return -EADDRNOTAVAIL;
980 }
981 }
982
983 if (!data)
984 return -EINVAL;
985
986 if (data[IFLA_VXLAN_ID]) {
987 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
988 if (id >= VXLAN_VID_MASK)
989 return -ERANGE;
990 }
991
992 if (data[IFLA_VXLAN_GROUP]) {
993 __be32 gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
994 if (!IN_MULTICAST(ntohl(gaddr))) {
995 pr_debug("group address is not IPv4 multicast\n");
996 return -EADDRNOTAVAIL;
997 }
998 }
999 return 0;
1000 }
1001
1002 static int vxlan_newlink(struct net *net, struct net_device *dev,
1003 struct nlattr *tb[], struct nlattr *data[])
1004 {
1005 struct vxlan_dev *vxlan = netdev_priv(dev);
1006 __u32 vni;
1007 int err;
1008
1009 if (!data[IFLA_VXLAN_ID])
1010 return -EINVAL;
1011
1012 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1013 if (vxlan_find_vni(net, vni)) {
1014 pr_info("duplicate VNI %u\n", vni);
1015 return -EEXIST;
1016 }
1017 vxlan->vni = vni;
1018
1019 if (data[IFLA_VXLAN_GROUP])
1020 vxlan->gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1021
1022 if (data[IFLA_VXLAN_LOCAL])
1023 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1024
1025 if (data[IFLA_VXLAN_LINK]) {
1026 vxlan->link = nla_get_u32(data[IFLA_VXLAN_LINK]);
1027
1028 if (!tb[IFLA_MTU]) {
1029 struct net_device *lowerdev;
1030 lowerdev = __dev_get_by_index(net, vxlan->link);
1031 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
1032 }
1033 }
1034
1035 if (data[IFLA_VXLAN_TOS])
1036 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1037
1038 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
1039 vxlan->learn = true;
1040
1041 if (data[IFLA_VXLAN_AGEING])
1042 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1043 else
1044 vxlan->age_interval = FDB_AGE_DEFAULT;
1045
1046 if (data[IFLA_VXLAN_LIMIT])
1047 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1048
1049 err = register_netdevice(dev);
1050 if (!err)
1051 hlist_add_head_rcu(&vxlan->hlist, vni_head(net, vxlan->vni));
1052
1053 return err;
1054 }
1055
1056 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1057 {
1058 struct vxlan_dev *vxlan = netdev_priv(dev);
1059
1060 hlist_del_rcu(&vxlan->hlist);
1061
1062 unregister_netdevice_queue(dev, head);
1063 }
1064
1065 static size_t vxlan_get_size(const struct net_device *dev)
1066 {
1067
1068 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1069 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1070 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1071 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1072 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1073 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1074 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
1075 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1076 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
1077 0;
1078 }
1079
1080 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1081 {
1082 const struct vxlan_dev *vxlan = netdev_priv(dev);
1083
1084 if (nla_put_u32(skb, IFLA_VXLAN_ID, vxlan->vni))
1085 goto nla_put_failure;
1086
1087 if (vxlan->gaddr && nla_put_u32(skb, IFLA_VXLAN_GROUP, vxlan->gaddr))
1088 goto nla_put_failure;
1089
1090 if (vxlan->link && nla_put_u32(skb, IFLA_VXLAN_LINK, vxlan->link))
1091 goto nla_put_failure;
1092
1093 if (vxlan->saddr && nla_put_u32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
1094 goto nla_put_failure;
1095
1096 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1097 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
1098 nla_put_u8(skb, IFLA_VXLAN_LEARNING, vxlan->learn) ||
1099 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1100 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax))
1101 goto nla_put_failure;
1102
1103 return 0;
1104
1105 nla_put_failure:
1106 return -EMSGSIZE;
1107 }
1108
1109 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1110 .kind = "vxlan",
1111 .maxtype = IFLA_VXLAN_MAX,
1112 .policy = vxlan_policy,
1113 .priv_size = sizeof(struct vxlan_dev),
1114 .setup = vxlan_setup,
1115 .validate = vxlan_validate,
1116 .newlink = vxlan_newlink,
1117 .dellink = vxlan_dellink,
1118 .get_size = vxlan_get_size,
1119 .fill_info = vxlan_fill_info,
1120 };
1121
1122 static __net_init int vxlan_init_net(struct net *net)
1123 {
1124 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1125 struct sock *sk;
1126 struct sockaddr_in vxlan_addr = {
1127 .sin_family = AF_INET,
1128 .sin_addr.s_addr = htonl(INADDR_ANY),
1129 };
1130 int rc;
1131 unsigned h;
1132
1133 /* Create UDP socket for encapsulation receive. */
1134 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vn->sock);
1135 if (rc < 0) {
1136 pr_debug("UDP socket create failed\n");
1137 return rc;
1138 }
1139 /* Put in proper namespace */
1140 sk = vn->sock->sk;
1141 sk_change_net(sk, net);
1142
1143 vxlan_addr.sin_port = htons(vxlan_port);
1144
1145 rc = kernel_bind(vn->sock, (struct sockaddr *) &vxlan_addr,
1146 sizeof(vxlan_addr));
1147 if (rc < 0) {
1148 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1149 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
1150 sk_release_kernel(sk);
1151 vn->sock = NULL;
1152 return rc;
1153 }
1154
1155 /* Disable multicast loopback */
1156 inet_sk(sk)->mc_loop = 0;
1157
1158 /* Mark socket as an encapsulation socket. */
1159 udp_sk(sk)->encap_type = 1;
1160 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1161 udp_encap_enable();
1162
1163 for (h = 0; h < VNI_HASH_SIZE; ++h)
1164 INIT_HLIST_HEAD(&vn->vni_list[h]);
1165
1166 return 0;
1167 }
1168
1169 static __net_exit void vxlan_exit_net(struct net *net)
1170 {
1171 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1172
1173 if (vn->sock) {
1174 sk_release_kernel(vn->sock->sk);
1175 vn->sock = NULL;
1176 }
1177 }
1178
1179 static struct pernet_operations vxlan_net_ops = {
1180 .init = vxlan_init_net,
1181 .exit = vxlan_exit_net,
1182 .id = &vxlan_net_id,
1183 .size = sizeof(struct vxlan_net),
1184 };
1185
1186 static int __init vxlan_init_module(void)
1187 {
1188 int rc;
1189
1190 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1191
1192 rc = register_pernet_device(&vxlan_net_ops);
1193 if (rc)
1194 goto out1;
1195
1196 rc = rtnl_link_register(&vxlan_link_ops);
1197 if (rc)
1198 goto out2;
1199
1200 return 0;
1201
1202 out2:
1203 unregister_pernet_device(&vxlan_net_ops);
1204 out1:
1205 return rc;
1206 }
1207 module_init(vxlan_init_module);
1208
1209 static void __exit vxlan_cleanup_module(void)
1210 {
1211 rtnl_link_unregister(&vxlan_link_ops);
1212 unregister_pernet_device(&vxlan_net_ops);
1213 }
1214 module_exit(vxlan_cleanup_module);
1215
1216 MODULE_LICENSE("GPL");
1217 MODULE_VERSION(VXLAN_VERSION);
1218 MODULE_AUTHOR("Stephen Hemminger <shemminger@vyatta.com>");
1219 MODULE_ALIAS_RTNL_LINK("vxlan");
(deprecated) Btrfs devel tree