projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ipv6: setsockopt(IPIPPROTO_IPV6, IPV6_MINHOPCOUNT) forgot to set return value
[linux-2.6-block.git] / drivers / net / vxlan.c
CommitLineData
d342894c 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>
d342894c 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)
52b702ff
AD		 53/* IP header + UDP + VXLAN + Ethernet header */
54#define VXLAN_HEADROOM (20 + 8 + 8 + 14)
d342894c 55
56#define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
57
58/* VXLAN protocol header */
59struct vxlanhdr {
60 __be32 vx_flags;
61 __be32 vx_vni;
62};
63
64/* UDP port for VXLAN traffic. */
65static unsigned int vxlan_port __read_mostly = 8472;
66module_param_named(udp_port, vxlan_port, uint, 0444);
67MODULE_PARM_DESC(udp_port, "Destination UDP port");
68
69static bool log_ecn_error = true;
70module_param(log_ecn_error, bool, 0644);
71MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
72
73/* per-net private data for this module */
74static unsigned int vxlan_net_id;
75struct vxlan_net {
76 struct socket *sock; /* UDP encap socket */
77 struct hlist_head vni_list[VNI_HASH_SIZE];
78};
79
80/* Forwarding table entry */
81struct 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 */
92struct 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 */
101struct 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 */
05f47d69 109 __u16 port_min; /* source port range */
110 __u16 port_max;
d342894c 111 __u8 tos; /* TOS override */
112 __u8 ttl;
113 bool learn;
114
115 unsigned long age_interval;
116 struct timer_list age_timer;
117 spinlock_t hash_lock;
118 unsigned int addrcnt;
119 unsigned int addrmax;
120 unsigned int addrexceeded;
121
122 struct hlist_head fdb_head[FDB_HASH_SIZE];
123};
124
125/* salt for hash table */
126static u32 vxlan_salt __read_mostly;
127
128static inline struct hlist_head *vni_head(struct net *net, u32 id)
129{
130 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
131
132 return &vn->vni_list[hash_32(id, VNI_HASH_BITS)];
133}
134
135/* Look up VNI in a per net namespace table */
136static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
137{
138 struct vxlan_dev *vxlan;
139 struct hlist_node *node;
140
141 hlist_for_each_entry_rcu(vxlan, node, vni_head(net, id), hlist) {
142 if (vxlan->vni == id)
143 return vxlan;
144 }
145
146 return NULL;
147}
148
149/* Fill in neighbour message in skbuff. */
150static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
151 const struct vxlan_fdb *fdb,
152 u32 portid, u32 seq, int type, unsigned int flags)
153{
154 unsigned long now = jiffies;
155 struct nda_cacheinfo ci;
156 struct nlmsghdr *nlh;
157 struct ndmsg *ndm;
158
159 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
160 if (nlh == NULL)
161 return -EMSGSIZE;
162
163 ndm = nlmsg_data(nlh);
164 memset(ndm, 0, sizeof(*ndm));
165 ndm->ndm_family = AF_BRIDGE;
166 ndm->ndm_state = fdb->state;
167 ndm->ndm_ifindex = vxlan->dev->ifindex;
168 ndm->ndm_flags = NTF_SELF;
169 ndm->ndm_type = NDA_DST;
170
171 if (nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
172 goto nla_put_failure;
173
174 if (nla_put_be32(skb, NDA_DST, fdb->remote_ip))
175 goto nla_put_failure;
176
177 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
178 ci.ndm_confirmed = 0;
179 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
180 ci.ndm_refcnt = 0;
181
182 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
183 goto nla_put_failure;
184
185 return nlmsg_end(skb, nlh);
186
187nla_put_failure:
188 nlmsg_cancel(skb, nlh);
189 return -EMSGSIZE;
190}
191
192static inline size_t vxlan_nlmsg_size(void)
193{
194 return NLMSG_ALIGN(sizeof(struct ndmsg))
195 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
196 + nla_total_size(sizeof(__be32)) /* NDA_DST */
197 + nla_total_size(sizeof(struct nda_cacheinfo));
198}
199
200static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
201 const struct vxlan_fdb *fdb, int type)
202{
203 struct net *net = dev_net(vxlan->dev);
204 struct sk_buff *skb;
205 int err = -ENOBUFS;
206
207 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
208 if (skb == NULL)
209 goto errout;
210
211 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0);
212 if (err < 0) {
213 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
214 WARN_ON(err == -EMSGSIZE);
215 kfree_skb(skb);
216 goto errout;
217 }
218
219 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
220 return;
221errout:
222 if (err < 0)
223 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
224}
225
226/* Hash Ethernet address */
227static u32 eth_hash(const unsigned char *addr)
228{
229 u64 value = get_unaligned((u64 *)addr);
230
231 /* only want 6 bytes */
232#ifdef __BIG_ENDIAN
d342894c 233 value >>= 16;
321fb991 234#else
235 value <<= 16;
d342894c 236#endif
237 return hash_64(value, FDB_HASH_BITS);
238}
239
240/* Hash chain to use given mac address */
241static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
242 const u8 *mac)
243{
244 return &vxlan->fdb_head[eth_hash(mac)];
245}
246
247/* Look up Ethernet address in forwarding table */
248static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
249 const u8 *mac)
250
251{
252 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
253 struct vxlan_fdb *f;
254 struct hlist_node *node;
255
256 hlist_for_each_entry_rcu(f, node, head, hlist) {
257 if (compare_ether_addr(mac, f->eth_addr) == 0)
258 return f;
259 }
260
261 return NULL;
262}
263
264/* Add new entry to forwarding table -- assumes lock held */
265static int vxlan_fdb_create(struct vxlan_dev *vxlan,
266 const u8 *mac, __be32 ip,
267 __u16 state, __u16 flags)
268{
269 struct vxlan_fdb *f;
270 int notify = 0;
271
272 f = vxlan_find_mac(vxlan, mac);
273 if (f) {
274 if (flags & NLM_F_EXCL) {
275 netdev_dbg(vxlan->dev,
276 "lost race to create %pM\n", mac);
277 return -EEXIST;
278 }
279 if (f->state != state) {
280 f->state = state;
281 f->updated = jiffies;
282 notify = 1;
283 }
284 } else {
285 if (!(flags & NLM_F_CREATE))
286 return -ENOENT;
287
288 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
289 return -ENOSPC;
290
291 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
292 f = kmalloc(sizeof(*f), GFP_ATOMIC);
293 if (!f)
294 return -ENOMEM;
295
296 notify = 1;
297 f->remote_ip = ip;
298 f->state = state;
299 f->updated = f->used = jiffies;
300 memcpy(f->eth_addr, mac, ETH_ALEN);
301
302 ++vxlan->addrcnt;
303 hlist_add_head_rcu(&f->hlist,
304 vxlan_fdb_head(vxlan, mac));
305 }
306
307 if (notify)
308 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
309
310 return 0;
311}
312
313static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
314{
315 netdev_dbg(vxlan->dev,
316 "delete %pM\n", f->eth_addr);
317
318 --vxlan->addrcnt;
319 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
320
321 hlist_del_rcu(&f->hlist);
322 kfree_rcu(f, rcu);
323}
324
325/* Add static entry (via netlink) */
326static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
327 struct net_device *dev,
328 const unsigned char *addr, u16 flags)
329{
330 struct vxlan_dev *vxlan = netdev_priv(dev);
331 __be32 ip;
332 int err;
333
334 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
335 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
336 ndm->ndm_state);
337 return -EINVAL;
338 }
339
340 if (tb[NDA_DST] == NULL)
341 return -EINVAL;
342
343 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
344 return -EAFNOSUPPORT;
345
346 ip = nla_get_be32(tb[NDA_DST]);
347
348 spin_lock_bh(&vxlan->hash_lock);
349 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags);
350 spin_unlock_bh(&vxlan->hash_lock);
351
352 return err;
353}
354
355/* Delete entry (via netlink) */
356static int vxlan_fdb_delete(struct ndmsg *ndm, struct net_device *dev,
357 const unsigned char *addr)
358{
359 struct vxlan_dev *vxlan = netdev_priv(dev);
360 struct vxlan_fdb *f;
361 int err = -ENOENT;
362
363 spin_lock_bh(&vxlan->hash_lock);
364 f = vxlan_find_mac(vxlan, addr);
365 if (f) {
366 vxlan_fdb_destroy(vxlan, f);
367 err = 0;
368 }
369 spin_unlock_bh(&vxlan->hash_lock);
370
371 return err;
372}
373
374/* Dump forwarding table */
375static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
376 struct net_device *dev, int idx)
377{
378 struct vxlan_dev *vxlan = netdev_priv(dev);
379 unsigned int h;
380
381 for (h = 0; h < FDB_HASH_SIZE; ++h) {
382 struct vxlan_fdb *f;
383 struct hlist_node *n;
384 int err;
385
386 hlist_for_each_entry_rcu(f, n, &vxlan->fdb_head[h], hlist) {
387 if (idx < cb->args[0])
388 goto skip;
389
390 err = vxlan_fdb_info(skb, vxlan, f,
391 NETLINK_CB(cb->skb).portid,
392 cb->nlh->nlmsg_seq,
393 RTM_NEWNEIGH,
394 NLM_F_MULTI);
395 if (err < 0)
396 break;
397skip:
398 ++idx;
399 }
400 }
401
402 return idx;
403}
404
405/* Watch incoming packets to learn mapping between Ethernet address
406 * and Tunnel endpoint.
407 */
408static void vxlan_snoop(struct net_device *dev,
409 __be32 src_ip, const u8 *src_mac)
410{
411 struct vxlan_dev *vxlan = netdev_priv(dev);
412 struct vxlan_fdb *f;
413 int err;
414
415 f = vxlan_find_mac(vxlan, src_mac);
416 if (likely(f)) {
417 f->used = jiffies;
418 if (likely(f->remote_ip == src_ip))
419 return;
420
421 if (net_ratelimit())
422 netdev_info(dev,
423 "%pM migrated from %pI4 to %pI4\n",
424 src_mac, &f->remote_ip, &src_ip);
425
426 f->remote_ip = src_ip;
427 f->updated = jiffies;
428 } else {
429 /* learned new entry */
430 spin_lock(&vxlan->hash_lock);
431 err = vxlan_fdb_create(vxlan, src_mac, src_ip,
432 NUD_REACHABLE,
433 NLM_F_EXCL|NLM_F_CREATE);
434 spin_unlock(&vxlan->hash_lock);
435 }
436}
437
438
439/* See if multicast group is already in use by other ID */
440static bool vxlan_group_used(struct vxlan_net *vn,
441 const struct vxlan_dev *this)
442{
443 const struct vxlan_dev *vxlan;
444 struct hlist_node *node;
445 unsigned h;
446
447 for (h = 0; h < VNI_HASH_SIZE; ++h)
448 hlist_for_each_entry(vxlan, node, &vn->vni_list[h], hlist) {
449 if (vxlan == this)
450 continue;
451
452 if (!netif_running(vxlan->dev))
453 continue;
454
455 if (vxlan->gaddr == this->gaddr)
456 return true;
457 }
458
459 return false;
460}
461
462/* kernel equivalent to IP_ADD_MEMBERSHIP */
463static int vxlan_join_group(struct net_device *dev)
464{
465 struct vxlan_dev *vxlan = netdev_priv(dev);
466 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
467 struct sock *sk = vn->sock->sk;
468 struct ip_mreqn mreq = {
469 .imr_multiaddr.s_addr = vxlan->gaddr,
470 };
471 int err;
472
473 /* Already a member of group */
474 if (vxlan_group_used(vn, vxlan))
475 return 0;
476
477 /* Need to drop RTNL to call multicast join */
478 rtnl_unlock();
479 lock_sock(sk);
480 err = ip_mc_join_group(sk, &mreq);
481 release_sock(sk);
482 rtnl_lock();
483
484 return err;
485}
486
487
488/* kernel equivalent to IP_DROP_MEMBERSHIP */
489static int vxlan_leave_group(struct net_device *dev)
490{
491 struct vxlan_dev *vxlan = netdev_priv(dev);
492 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
493 int err = 0;
494 struct sock *sk = vn->sock->sk;
495 struct ip_mreqn mreq = {
496 .imr_multiaddr.s_addr = vxlan->gaddr,
497 };
498
499 /* Only leave group when last vxlan is done. */
500 if (vxlan_group_used(vn, vxlan))
501 return 0;
502
503 /* Need to drop RTNL to call multicast leave */
504 rtnl_unlock();
505 lock_sock(sk);
506 err = ip_mc_leave_group(sk, &mreq);
507 release_sock(sk);
508 rtnl_lock();
509
510 return err;
511}
512
513/* Callback from net/ipv4/udp.c to receive packets */
514static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
515{
516 struct iphdr *oip;
517 struct vxlanhdr *vxh;
518 struct vxlan_dev *vxlan;
519 struct vxlan_stats *stats;
520 __u32 vni;
521 int err;
522
523 /* pop off outer UDP header */
524 __skb_pull(skb, sizeof(struct udphdr));
525
526 /* Need Vxlan and inner Ethernet header to be present */
527 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
528 goto error;
529
530 /* Drop packets with reserved bits set */
531 vxh = (struct vxlanhdr *) skb->data;
532 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
533 (vxh->vx_vni & htonl(0xff))) {
534 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
535 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
536 goto error;
537 }
538
539 __skb_pull(skb, sizeof(struct vxlanhdr));
d342894c 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);
d342894c 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);
d97c00a3 569 skb->ip_summed = CHECKSUM_NONE;
d342894c 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;
592error:
593 /* Put UDP header back */
594 __skb_push(skb, sizeof(struct udphdr));
595
596 return 1;
597drop:
598 /* Consume bad packet */
599 kfree_skb(skb);
600 return 0;
601}
602
603/* Extract dsfield from inner protocol */
604static 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 */
616static 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
ef59febe 625static __be32 vxlan_find_dst(struct vxlan_dev *vxlan, struct sk_buff *skb)
626{
627 const struct ethhdr *eth = (struct ethhdr *) skb->data;
628 const struct vxlan_fdb *f;
629
630 if (is_multicast_ether_addr(eth->h_dest))
631 return vxlan->gaddr;
632
633 f = vxlan_find_mac(vxlan, eth->h_dest);
634 if (f)
635 return f->remote_ip;
636 else
637 return vxlan->gaddr;
638
639}
640
1cad8715 641static void vxlan_sock_free(struct sk_buff *skb)
642{
643 sock_put(skb->sk);
644}
645
646/* On transmit, associate with the tunnel socket */
647static void vxlan_set_owner(struct net_device *dev, struct sk_buff *skb)
648{
649 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
650 struct sock *sk = vn->sock->sk;
651
652 skb_orphan(skb);
653 sock_hold(sk);
654 skb->sk = sk;
655 skb->destructor = vxlan_sock_free;
656}
657
05f47d69 658/* Compute source port for outgoing packet
659 * first choice to use L4 flow hash since it will spread
660 * better and maybe available from hardware
661 * secondary choice is to use jhash on the Ethernet header
662 */
663static u16 vxlan_src_port(const struct vxlan_dev *vxlan, struct sk_buff *skb)
664{
665 unsigned int range = (vxlan->port_max - vxlan->port_min) + 1;
666 u32 hash;
667
668 hash = skb_get_rxhash(skb);
669 if (!hash)
670 hash = jhash(skb->data, 2 * ETH_ALEN,
671 (__force u32) skb->protocol);
672
673 return (((u64) hash * range) >> 32) + vxlan->port_min;
674}
675
d342894c 676/* Transmit local packets over Vxlan
677 *
678 * Outer IP header inherits ECN and DF from inner header.
679 * Outer UDP destination is the VXLAN assigned port.
05f47d69 680 * source port is based on hash of flow
d342894c 681 */
682static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
683{
684 struct vxlan_dev *vxlan = netdev_priv(dev);
685 struct rtable *rt;
d342894c 686 const struct iphdr *old_iph;
687 struct iphdr *iph;
688 struct vxlanhdr *vxh;
689 struct udphdr *uh;
690 struct flowi4 fl4;
d342894c 691 unsigned int pkt_len = skb->len;
d342894c 692 __be32 dst;
05f47d69 693 __u16 src_port;
d342894c 694 __be16 df = 0;
695 __u8 tos, ttl;
696 int err;
697
ef59febe 698 dst = vxlan_find_dst(vxlan, skb);
699 if (!dst)
700 goto drop;
701
d342894c 702 /* Need space for new headers (invalidates iph ptr) */
703 if (skb_cow_head(skb, VXLAN_HEADROOM))
704 goto drop;
705
d342894c 706 old_iph = ip_hdr(skb);
707
d342894c 708 ttl = vxlan->ttl;
709 if (!ttl && IN_MULTICAST(ntohl(dst)))
710 ttl = 1;
711
712 tos = vxlan->tos;
713 if (tos == 1)
714 tos = vxlan_get_dsfield(old_iph, skb);
715
05f47d69 716 src_port = vxlan_src_port(vxlan, skb);
d342894c 717
ca78f181 718 memset(&fl4, 0, sizeof(fl4));
719 fl4.flowi4_oif = vxlan->link;
720 fl4.flowi4_tos = RT_TOS(tos);
721 fl4.daddr = dst;
722 fl4.saddr = vxlan->saddr;
723
724 rt = ip_route_output_key(dev_net(dev), &fl4);
d342894c 725 if (IS_ERR(rt)) {
726 netdev_dbg(dev, "no route to %pI4\n", &dst);
727 dev->stats.tx_carrier_errors++;
728 goto tx_error;
729 }
730
731 if (rt->dst.dev == dev) {
732 netdev_dbg(dev, "circular route to %pI4\n", &dst);
733 ip_rt_put(rt);
734 dev->stats.collisions++;
735 goto tx_error;
736 }
737
738 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
739 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
740 IPSKB_REROUTED);
741 skb_dst_drop(skb);
742 skb_dst_set(skb, &rt->dst);
743
744 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
745 vxh->vx_flags = htonl(VXLAN_FLAGS);
746 vxh->vx_vni = htonl(vxlan->vni << 8);
747
748 __skb_push(skb, sizeof(*uh));
749 skb_reset_transport_header(skb);
750 uh = udp_hdr(skb);
751
752 uh->dest = htons(vxlan_port);
05f47d69 753 uh->source = htons(src_port);
d342894c 754
755 uh->len = htons(skb->len);
756 uh->check = 0;
757
758 __skb_push(skb, sizeof(*iph));
759 skb_reset_network_header(skb);
760 iph = ip_hdr(skb);
761 iph->version = 4;
762 iph->ihl = sizeof(struct iphdr) >> 2;
763 iph->frag_off = df;
764 iph->protocol = IPPROTO_UDP;
765 iph->tos = vxlan_ecn_encap(tos, old_iph, skb);
ca78f181 766 iph->daddr = dst;
d342894c 767 iph->saddr = fl4.saddr;
768 iph->ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
769
1cad8715 770 vxlan_set_owner(dev, skb);
771
d342894c 772 /* See __IPTUNNEL_XMIT */
773 skb->ip_summed = CHECKSUM_NONE;
774 ip_select_ident(iph, &rt->dst, NULL);
775
776 err = ip_local_out(skb);
777 if (likely(net_xmit_eval(err) == 0)) {
778 struct vxlan_stats *stats = this_cpu_ptr(vxlan->stats);
779
780 u64_stats_update_begin(&stats->syncp);
781 stats->tx_packets++;
782 stats->tx_bytes += pkt_len;
783 u64_stats_update_end(&stats->syncp);
784 } else {
785 dev->stats.tx_errors++;
786 dev->stats.tx_aborted_errors++;
787 }
788 return NETDEV_TX_OK;
789
790drop:
791 dev->stats.tx_dropped++;
792 goto tx_free;
793
794tx_error:
795 dev->stats.tx_errors++;
796tx_free:
797 dev_kfree_skb(skb);
798 return NETDEV_TX_OK;
799}
800
801/* Walk the forwarding table and purge stale entries */
802static void vxlan_cleanup(unsigned long arg)
803{
804 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
805 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
806 unsigned int h;
807
808 if (!netif_running(vxlan->dev))
809 return;
810
811 spin_lock_bh(&vxlan->hash_lock);
812 for (h = 0; h < FDB_HASH_SIZE; ++h) {
813 struct hlist_node *p, *n;
814 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
815 struct vxlan_fdb *f
816 = container_of(p, struct vxlan_fdb, hlist);
817 unsigned long timeout;
818
3c172868 819 if (f->state & NUD_PERMANENT)
d342894c 820 continue;
821
822 timeout = f->used + vxlan->age_interval * HZ;
823 if (time_before_eq(timeout, jiffies)) {
824 netdev_dbg(vxlan->dev,
825 "garbage collect %pM\n",
826 f->eth_addr);
827 f->state = NUD_STALE;
828 vxlan_fdb_destroy(vxlan, f);
829 } else if (time_before(timeout, next_timer))
830 next_timer = timeout;
831 }
832 }
833 spin_unlock_bh(&vxlan->hash_lock);
834
835 mod_timer(&vxlan->age_timer, next_timer);
836}
837
838/* Setup stats when device is created */
839static int vxlan_init(struct net_device *dev)
840{
841 struct vxlan_dev *vxlan = netdev_priv(dev);
842
843 vxlan->stats = alloc_percpu(struct vxlan_stats);
844 if (!vxlan->stats)
845 return -ENOMEM;
846
847 return 0;
848}
849
850/* Start ageing timer and join group when device is brought up */
851static int vxlan_open(struct net_device *dev)
852{
853 struct vxlan_dev *vxlan = netdev_priv(dev);
854 int err;
855
856 if (vxlan->gaddr) {
857 err = vxlan_join_group(dev);
858 if (err)
859 return err;
860 }
861
862 if (vxlan->age_interval)
863 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
864
865 return 0;
866}
867
868/* Purge the forwarding table */
869static void vxlan_flush(struct vxlan_dev *vxlan)
870{
871 unsigned h;
872
873 spin_lock_bh(&vxlan->hash_lock);
874 for (h = 0; h < FDB_HASH_SIZE; ++h) {
875 struct hlist_node *p, *n;
876 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
877 struct vxlan_fdb *f
878 = container_of(p, struct vxlan_fdb, hlist);
879 vxlan_fdb_destroy(vxlan, f);
880 }
881 }
882 spin_unlock_bh(&vxlan->hash_lock);
883}
884
885/* Cleanup timer and forwarding table on shutdown */
886static int vxlan_stop(struct net_device *dev)
887{
888 struct vxlan_dev *vxlan = netdev_priv(dev);
889
890 if (vxlan->gaddr)
891 vxlan_leave_group(dev);
892
893 del_timer_sync(&vxlan->age_timer);
894
895 vxlan_flush(vxlan);
896
897 return 0;
898}
899
900/* Merge per-cpu statistics */
901static struct rtnl_link_stats64 *vxlan_stats64(struct net_device *dev,
902 struct rtnl_link_stats64 *stats)
903{
904 struct vxlan_dev *vxlan = netdev_priv(dev);
905 struct vxlan_stats tmp, sum = { 0 };
906 unsigned int cpu;
907
908 for_each_possible_cpu(cpu) {
909 unsigned int start;
910 const struct vxlan_stats *stats
911 = per_cpu_ptr(vxlan->stats, cpu);
912
913 do {
914 start = u64_stats_fetch_begin_bh(&stats->syncp);
915 memcpy(&tmp, stats, sizeof(tmp));
916 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
917
918 sum.tx_bytes += tmp.tx_bytes;
919 sum.tx_packets += tmp.tx_packets;
920 sum.rx_bytes += tmp.rx_bytes;
921 sum.rx_packets += tmp.rx_packets;
922 }
923
924 stats->tx_bytes = sum.tx_bytes;
925 stats->tx_packets = sum.tx_packets;
926 stats->rx_bytes = sum.rx_bytes;
927 stats->rx_packets = sum.rx_packets;
928
929 stats->multicast = dev->stats.multicast;
930 stats->rx_length_errors = dev->stats.rx_length_errors;
931 stats->rx_frame_errors = dev->stats.rx_frame_errors;
932 stats->rx_errors = dev->stats.rx_errors;
933
934 stats->tx_dropped = dev->stats.tx_dropped;
935 stats->tx_carrier_errors = dev->stats.tx_carrier_errors;
936 stats->tx_aborted_errors = dev->stats.tx_aborted_errors;
937 stats->collisions = dev->stats.collisions;
938 stats->tx_errors = dev->stats.tx_errors;
939
940 return stats;
941}
942
943/* Stub, nothing needs to be done. */
944static void vxlan_set_multicast_list(struct net_device *dev)
945{
946}
947
948static const struct net_device_ops vxlan_netdev_ops = {
949 .ndo_init = vxlan_init,
950 .ndo_open = vxlan_open,
951 .ndo_stop = vxlan_stop,
952 .ndo_start_xmit = vxlan_xmit,
953 .ndo_get_stats64 = vxlan_stats64,
954 .ndo_set_rx_mode = vxlan_set_multicast_list,
955 .ndo_change_mtu = eth_change_mtu,
956 .ndo_validate_addr = eth_validate_addr,
957 .ndo_set_mac_address = eth_mac_addr,
958 .ndo_fdb_add = vxlan_fdb_add,
959 .ndo_fdb_del = vxlan_fdb_delete,
960 .ndo_fdb_dump = vxlan_fdb_dump,
961};
962
963/* Info for udev, that this is a virtual tunnel endpoint */
964static struct device_type vxlan_type = {
965 .name = "vxlan",
966};
967
968static void vxlan_free(struct net_device *dev)
969{
970 struct vxlan_dev *vxlan = netdev_priv(dev);
971
972 free_percpu(vxlan->stats);
973 free_netdev(dev);
974}
975
976/* Initialize the device structure. */
977static void vxlan_setup(struct net_device *dev)
978{
979 struct vxlan_dev *vxlan = netdev_priv(dev);
980 unsigned h;
05f47d69 981 int low, high;
d342894c 982
983 eth_hw_addr_random(dev);
984 ether_setup(dev);
2840bf22 985 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
d342894c 986
987 dev->netdev_ops = &vxlan_netdev_ops;
988 dev->destructor = vxlan_free;
989 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
990
991 dev->tx_queue_len = 0;
992 dev->features |= NETIF_F_LLTX;
993 dev->features |= NETIF_F_NETNS_LOCAL;
994 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
995
996 spin_lock_init(&vxlan->hash_lock);
997
998 init_timer_deferrable(&vxlan->age_timer);
999 vxlan->age_timer.function = vxlan_cleanup;
1000 vxlan->age_timer.data = (unsigned long) vxlan;
1001
05f47d69 1002 inet_get_local_port_range(&low, &high);
1003 vxlan->port_min = low;
1004 vxlan->port_max = high;
1005
d342894c 1006 vxlan->dev = dev;
1007
1008 for (h = 0; h < FDB_HASH_SIZE; ++h)
1009 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
1010}
1011
1012static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
1013 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
1014 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1015 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
1016 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1017 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
1018 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
1019 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
1020 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
1021 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
05f47d69 1022 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
d342894c 1023};
1024
1025static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
1026{
1027 if (tb[IFLA_ADDRESS]) {
1028 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1029 pr_debug("invalid link address (not ethernet)\n");
1030 return -EINVAL;
1031 }
1032
1033 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1034 pr_debug("invalid all zero ethernet address\n");
1035 return -EADDRNOTAVAIL;
1036 }
1037 }
1038
1039 if (!data)
1040 return -EINVAL;
1041
1042 if (data[IFLA_VXLAN_ID]) {
1043 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
1044 if (id >= VXLAN_VID_MASK)
1045 return -ERANGE;
1046 }
1047
1048 if (data[IFLA_VXLAN_GROUP]) {
1049 __be32 gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1050 if (!IN_MULTICAST(ntohl(gaddr))) {
1051 pr_debug("group address is not IPv4 multicast\n");
1052 return -EADDRNOTAVAIL;
1053 }
1054 }
05f47d69 1055
1056 if (data[IFLA_VXLAN_PORT_RANGE]) {
1057 const struct ifla_vxlan_port_range *p
1058 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1059
1060 if (ntohs(p->high) < ntohs(p->low)) {
1061 pr_debug("port range %u .. %u not valid\n",
1062 ntohs(p->low), ntohs(p->high));
1063 return -EINVAL;
1064 }
1065 }
1066
d342894c 1067 return 0;
1068}
1069
1070static int vxlan_newlink(struct net *net, struct net_device *dev,
1071 struct nlattr *tb[], struct nlattr *data[])
1072{
1073 struct vxlan_dev *vxlan = netdev_priv(dev);
1074 __u32 vni;
1075 int err;
1076
1077 if (!data[IFLA_VXLAN_ID])
1078 return -EINVAL;
1079
1080 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1081 if (vxlan_find_vni(net, vni)) {
1082 pr_info("duplicate VNI %u\n", vni);
1083 return -EEXIST;
1084 }
1085 vxlan->vni = vni;
1086
1087 if (data[IFLA_VXLAN_GROUP])
1088 vxlan->gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1089
1090 if (data[IFLA_VXLAN_LOCAL])
1091 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1092
34e02aa1 1093 if (data[IFLA_VXLAN_LINK] &&
1094 (vxlan->link = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
1095 struct net_device *lowerdev
1096 = __dev_get_by_index(net, vxlan->link);
1097
1098 if (!lowerdev) {
1099 pr_info("ifindex %d does not exist\n", vxlan->link);
1100 return -ENODEV;
1101 }
d342894c 1102
34e02aa1 1103 if (!tb[IFLA_MTU])
d342894c 1104 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
d342894c 1105 }
1106
1107 if (data[IFLA_VXLAN_TOS])
1108 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1109
1110 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
1111 vxlan->learn = true;
1112
1113 if (data[IFLA_VXLAN_AGEING])
1114 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1115 else
1116 vxlan->age_interval = FDB_AGE_DEFAULT;
1117
1118 if (data[IFLA_VXLAN_LIMIT])
1119 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1120
05f47d69 1121 if (data[IFLA_VXLAN_PORT_RANGE]) {
1122 const struct ifla_vxlan_port_range *p
1123 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1124 vxlan->port_min = ntohs(p->low);
1125 vxlan->port_max = ntohs(p->high);
1126 }
1127
d342894c 1128 err = register_netdevice(dev);
1129 if (!err)
1130 hlist_add_head_rcu(&vxlan->hlist, vni_head(net, vxlan->vni));
1131
1132 return err;
1133}
1134
1135static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1136{
1137 struct vxlan_dev *vxlan = netdev_priv(dev);
1138
1139 hlist_del_rcu(&vxlan->hlist);
1140
1141 unregister_netdevice_queue(dev, head);
1142}
1143
1144static size_t vxlan_get_size(const struct net_device *dev)
1145{
1146
1147 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1148 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1149 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1150 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1151 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1152 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1153 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
1154 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1155 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
05f47d69 1156 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
d342894c 1157 0;
1158}
1159
1160static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1161{
1162 const struct vxlan_dev *vxlan = netdev_priv(dev);
05f47d69 1163 struct ifla_vxlan_port_range ports = {
1164 .low = htons(vxlan->port_min),
1165 .high = htons(vxlan->port_max),
1166 };
d342894c 1167
1168 if (nla_put_u32(skb, IFLA_VXLAN_ID, vxlan->vni))
1169 goto nla_put_failure;
1170
7c41c42c 1171 if (vxlan->gaddr && nla_put_be32(skb, IFLA_VXLAN_GROUP, vxlan->gaddr))
d342894c 1172 goto nla_put_failure;
1173
1174 if (vxlan->link && nla_put_u32(skb, IFLA_VXLAN_LINK, vxlan->link))
1175 goto nla_put_failure;
1176
7c41c42c 1177 if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
d342894c 1178 goto nla_put_failure;
1179
1180 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1181 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
1182 nla_put_u8(skb, IFLA_VXLAN_LEARNING, vxlan->learn) ||
1183 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1184 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax))
1185 goto nla_put_failure;
1186
05f47d69 1187 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
1188 goto nla_put_failure;
1189
d342894c 1190 return 0;
1191
1192nla_put_failure:
1193 return -EMSGSIZE;
1194}
1195
1196static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1197 .kind = "vxlan",
1198 .maxtype = IFLA_VXLAN_MAX,
1199 .policy = vxlan_policy,
1200 .priv_size = sizeof(struct vxlan_dev),
1201 .setup = vxlan_setup,
1202 .validate = vxlan_validate,
1203 .newlink = vxlan_newlink,
1204 .dellink = vxlan_dellink,
1205 .get_size = vxlan_get_size,
1206 .fill_info = vxlan_fill_info,
1207};
1208
1209static __net_init int vxlan_init_net(struct net *net)
1210{
1211 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1212 struct sock *sk;
1213 struct sockaddr_in vxlan_addr = {
1214 .sin_family = AF_INET,
1215 .sin_addr.s_addr = htonl(INADDR_ANY),
1216 };
1217 int rc;
1218 unsigned h;
1219
1220 /* Create UDP socket for encapsulation receive. */
1221 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vn->sock);
1222 if (rc < 0) {
1223 pr_debug("UDP socket create failed\n");
1224 return rc;
1225 }
bfe1b9b1 1226 /* Put in proper namespace */
1227 sk = vn->sock->sk;
1228 sk_change_net(sk, net);
d342894c 1229
1230 vxlan_addr.sin_port = htons(vxlan_port);
1231
1232 rc = kernel_bind(vn->sock, (struct sockaddr *) &vxlan_addr,
1233 sizeof(vxlan_addr));
1234 if (rc < 0) {
1235 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1236 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
bfe1b9b1 1237 sk_release_kernel(sk);
d342894c 1238 vn->sock = NULL;
1239 return rc;
1240 }
1241
1242 /* Disable multicast loopback */
d342894c 1243 inet_sk(sk)->mc_loop = 0;
1244
1245 /* Mark socket as an encapsulation socket. */
1246 udp_sk(sk)->encap_type = 1;
1247 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1248 udp_encap_enable();
1249
1250 for (h = 0; h < VNI_HASH_SIZE; ++h)
1251 INIT_HLIST_HEAD(&vn->vni_list[h]);
1252
1253 return 0;
1254}
1255
1256static __net_exit void vxlan_exit_net(struct net *net)
1257{
1258 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1259
1260 if (vn->sock) {
bfe1b9b1 1261 sk_release_kernel(vn->sock->sk);
d342894c 1262 vn->sock = NULL;
1263 }
1264}
1265
1266static struct pernet_operations vxlan_net_ops = {
1267 .init = vxlan_init_net,
1268 .exit = vxlan_exit_net,
1269 .id = &vxlan_net_id,
1270 .size = sizeof(struct vxlan_net),
1271};
1272
1273static int __init vxlan_init_module(void)
1274{
1275 int rc;
1276
1277 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1278
1279 rc = register_pernet_device(&vxlan_net_ops);
1280 if (rc)
1281 goto out1;
1282
1283 rc = rtnl_link_register(&vxlan_link_ops);
1284 if (rc)
1285 goto out2;
1286
1287 return 0;
1288
1289out2:
1290 unregister_pernet_device(&vxlan_net_ops);
1291out1:
1292 return rc;
1293}
1294module_init(vxlan_init_module);
1295
1296static void __exit vxlan_cleanup_module(void)
1297{
1298 rtnl_link_unregister(&vxlan_link_ops);
1299 unregister_pernet_device(&vxlan_net_ops);
1300}
1301module_exit(vxlan_cleanup_module);
1302
1303MODULE_LICENSE("GPL");
1304MODULE_VERSION(VXLAN_VERSION);
1305MODULE_AUTHOR("Stephen Hemminger <shemminger@vyatta.com>");
1306MODULE_ALIAS_RTNL_LINK("vxlan");
Linux 6.x block layer and io_uring tree(s)