projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
IPIP: Use ip-tunneling code.
[linux-2.6-block.git] / drivers / net / vxlan.c
CommitLineData
d342894c 1/*
eb5ce439 2 * VXLAN: Virtual eXtensible Local Area Network
d342894c 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>
1b13c97f 32#include <linux/ethtool.h>
e4f67add
DS		 33#include <net/arp.h>
34#include <net/ndisc.h>
d342894c 35#include <net/ip.h>
c5441932 36#include <net/ip_tunnels.h>
d342894c 37#include <net/icmp.h>
38#include <net/udp.h>
39#include <net/rtnetlink.h>
40#include <net/route.h>
41#include <net/dsfield.h>
42#include <net/inet_ecn.h>
43#include <net/net_namespace.h>
44#include <net/netns/generic.h>
45
46#define VXLAN_VERSION "0.1"
47
48#define VNI_HASH_BITS 10
49#define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
50#define FDB_HASH_BITS 8
51#define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
52#define FDB_AGE_DEFAULT 300 /* 5 min */
53#define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
54
55#define VXLAN_N_VID (1u << 24)
56#define VXLAN_VID_MASK (VXLAN_N_VID - 1)
52b702ff
AD		 57/* IP header + UDP + VXLAN + Ethernet header */
58#define VXLAN_HEADROOM (20 + 8 + 8 + 14)
d342894c 59
60#define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
61
62/* VXLAN protocol header */
63struct vxlanhdr {
64 __be32 vx_flags;
65 __be32 vx_vni;
66};
67
68/* UDP port for VXLAN traffic. */
69static unsigned int vxlan_port __read_mostly = 8472;
70module_param_named(udp_port, vxlan_port, uint, 0444);
71MODULE_PARM_DESC(udp_port, "Destination UDP port");
72
73static bool log_ecn_error = true;
74module_param(log_ecn_error, bool, 0644);
75MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
76
77/* per-net private data for this module */
78static unsigned int vxlan_net_id;
79struct vxlan_net {
80 struct socket *sock; /* UDP encap socket */
81 struct hlist_head vni_list[VNI_HASH_SIZE];
82};
83
6681712d
DS		 84struct vxlan_rdst {
85 struct rcu_head rcu;
86 __be32 remote_ip;
87 __be16 remote_port;
88 u32 remote_vni;
89 u32 remote_ifindex;
90 struct vxlan_rdst *remote_next;
91};
92
d342894c 93/* Forwarding table entry */
94struct vxlan_fdb {
95 struct hlist_node hlist; /* linked list of entries */
96 struct rcu_head rcu;
97 unsigned long updated; /* jiffies */
98 unsigned long used;
6681712d 99 struct vxlan_rdst remote;
d342894c 100 u16 state; /* see ndm_state */
101 u8 eth_addr[ETH_ALEN];
102};
103
104/* Per-cpu network traffic stats */
105struct vxlan_stats {
106 u64 rx_packets;
107 u64 rx_bytes;
108 u64 tx_packets;
109 u64 tx_bytes;
110 struct u64_stats_sync syncp;
111};
112
113/* Pseudo network device */
114struct vxlan_dev {
115 struct hlist_node hlist;
116 struct net_device *dev;
117 struct vxlan_stats __percpu *stats;
118 __u32 vni; /* virtual network id */
119 __be32 gaddr; /* multicast group */
120 __be32 saddr; /* source address */
121 unsigned int link; /* link to multicast over */
05f47d69 122 __u16 port_min; /* source port range */
123 __u16 port_max;
d342894c 124 __u8 tos; /* TOS override */
125 __u8 ttl;
e4f67add 126 u32 flags; /* VXLAN_F_* below */
d342894c 127
128 unsigned long age_interval;
129 struct timer_list age_timer;
130 spinlock_t hash_lock;
131 unsigned int addrcnt;
132 unsigned int addrmax;
d342894c 133
134 struct hlist_head fdb_head[FDB_HASH_SIZE];
135};
136
e4f67add
DS		 137#define VXLAN_F_LEARN 0x01
138#define VXLAN_F_PROXY 0x02
139#define VXLAN_F_RSC 0x04
140#define VXLAN_F_L2MISS 0x08
141#define VXLAN_F_L3MISS 0x10
142
d342894c 143/* salt for hash table */
144static u32 vxlan_salt __read_mostly;
145
146static inline struct hlist_head *vni_head(struct net *net, u32 id)
147{
148 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
149
150 return &vn->vni_list[hash_32(id, VNI_HASH_BITS)];
151}
152
153/* Look up VNI in a per net namespace table */
154static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
155{
156 struct vxlan_dev *vxlan;
d342894c 157
b67bfe0d 158 hlist_for_each_entry_rcu(vxlan, vni_head(net, id), hlist) {
d342894c 159 if (vxlan->vni == id)
160 return vxlan;
161 }
162
163 return NULL;
164}
165
166/* Fill in neighbour message in skbuff. */
167static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
168 const struct vxlan_fdb *fdb,
6681712d
DS		 169 u32 portid, u32 seq, int type, unsigned int flags,
170 const struct vxlan_rdst *rdst)
d342894c 171{
172 unsigned long now = jiffies;
173 struct nda_cacheinfo ci;
174 struct nlmsghdr *nlh;
175 struct ndmsg *ndm;
e4f67add 176 bool send_ip, send_eth;
d342894c 177
178 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
179 if (nlh == NULL)
180 return -EMSGSIZE;
181
182 ndm = nlmsg_data(nlh);
183 memset(ndm, 0, sizeof(*ndm));
e4f67add
DS		 184
185 send_eth = send_ip = true;
186
187 if (type == RTM_GETNEIGH) {
188 ndm->ndm_family = AF_INET;
6681712d 189 send_ip = rdst->remote_ip != htonl(INADDR_ANY);
e4f67add
DS		 190 send_eth = !is_zero_ether_addr(fdb->eth_addr);
191 } else
192 ndm->ndm_family = AF_BRIDGE;
d342894c 193 ndm->ndm_state = fdb->state;
194 ndm->ndm_ifindex = vxlan->dev->ifindex;
195 ndm->ndm_flags = NTF_SELF;
196 ndm->ndm_type = NDA_DST;
197
e4f67add 198 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
d342894c 199 goto nla_put_failure;
200
6681712d
DS		 201 if (send_ip && nla_put_be32(skb, NDA_DST, rdst->remote_ip))
202 goto nla_put_failure;
203
204 if (rdst->remote_port && rdst->remote_port != vxlan_port &&
205 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
206 goto nla_put_failure;
207 if (rdst->remote_vni != vxlan->vni &&
208 nla_put_be32(skb, NDA_VNI, rdst->remote_vni))
209 goto nla_put_failure;
210 if (rdst->remote_ifindex &&
211 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
d342894c 212 goto nla_put_failure;
213
214 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
215 ci.ndm_confirmed = 0;
216 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
217 ci.ndm_refcnt = 0;
218
219 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
220 goto nla_put_failure;
221
222 return nlmsg_end(skb, nlh);
223
224nla_put_failure:
225 nlmsg_cancel(skb, nlh);
226 return -EMSGSIZE;
227}
228
229static inline size_t vxlan_nlmsg_size(void)
230{
231 return NLMSG_ALIGN(sizeof(struct ndmsg))
232 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
233 + nla_total_size(sizeof(__be32)) /* NDA_DST */
6681712d
DS		 234 + nla_total_size(sizeof(__be32)) /* NDA_PORT */
235 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
236 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
d342894c 237 + nla_total_size(sizeof(struct nda_cacheinfo));
238}
239
240static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
241 const struct vxlan_fdb *fdb, int type)
242{
243 struct net *net = dev_net(vxlan->dev);
244 struct sk_buff *skb;
245 int err = -ENOBUFS;
246
247 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
248 if (skb == NULL)
249 goto errout;
250
6681712d 251 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, &fdb->remote);
d342894c 252 if (err < 0) {
253 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
254 WARN_ON(err == -EMSGSIZE);
255 kfree_skb(skb);
256 goto errout;
257 }
258
259 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
260 return;
261errout:
262 if (err < 0)
263 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
264}
265
e4f67add
DS		 266static void vxlan_ip_miss(struct net_device *dev, __be32 ipa)
267{
268 struct vxlan_dev *vxlan = netdev_priv(dev);
269 struct vxlan_fdb f;
270
271 memset(&f, 0, sizeof f);
272 f.state = NUD_STALE;
6681712d
DS		 273 f.remote.remote_ip = ipa; /* goes to NDA_DST */
274 f.remote.remote_vni = VXLAN_N_VID;
e4f67add
DS		 275
276 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
277}
278
279static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
280{
281 struct vxlan_fdb f;
282
283 memset(&f, 0, sizeof f);
284 f.state = NUD_STALE;
285 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
286
287 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
288}
289
d342894c 290/* Hash Ethernet address */
291static u32 eth_hash(const unsigned char *addr)
292{
293 u64 value = get_unaligned((u64 *)addr);
294
295 /* only want 6 bytes */
296#ifdef __BIG_ENDIAN
d342894c 297 value >>= 16;
321fb991 298#else
299 value <<= 16;
d342894c 300#endif
301 return hash_64(value, FDB_HASH_BITS);
302}
303
304/* Hash chain to use given mac address */
305static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
306 const u8 *mac)
307{
308 return &vxlan->fdb_head[eth_hash(mac)];
309}
310
311/* Look up Ethernet address in forwarding table */
312static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
313 const u8 *mac)
314
315{
316 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
317 struct vxlan_fdb *f;
d342894c 318
b67bfe0d 319 hlist_for_each_entry_rcu(f, head, hlist) {
d342894c 320 if (compare_ether_addr(mac, f->eth_addr) == 0)
321 return f;
322 }
323
324 return NULL;
325}
326
6681712d
DS		 327/* Add/update destinations for multicast */
328static int vxlan_fdb_append(struct vxlan_fdb *f,
329 __be32 ip, __u32 port, __u32 vni, __u32 ifindex)
330{
331 struct vxlan_rdst *rd_prev, *rd;
332
333 rd_prev = NULL;
334 for (rd = &f->remote; rd; rd = rd->remote_next) {
335 if (rd->remote_ip == ip &&
336 rd->remote_port == port &&
337 rd->remote_vni == vni &&
338 rd->remote_ifindex == ifindex)
339 return 0;
340 rd_prev = rd;
341 }
342 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
343 if (rd == NULL)
344 return -ENOBUFS;
345 rd->remote_ip = ip;
346 rd->remote_port = port;
347 rd->remote_vni = vni;
348 rd->remote_ifindex = ifindex;
349 rd->remote_next = NULL;
350 rd_prev->remote_next = rd;
351 return 1;
352}
353
d342894c 354/* Add new entry to forwarding table -- assumes lock held */
355static int vxlan_fdb_create(struct vxlan_dev *vxlan,
356 const u8 *mac, __be32 ip,
6681712d
DS		 357 __u16 state, __u16 flags,
358 __u32 port, __u32 vni, __u32 ifindex)
d342894c 359{
360 struct vxlan_fdb *f;
361 int notify = 0;
362
363 f = vxlan_find_mac(vxlan, mac);
364 if (f) {
365 if (flags & NLM_F_EXCL) {
366 netdev_dbg(vxlan->dev,
367 "lost race to create %pM\n", mac);
368 return -EEXIST;
369 }
370 if (f->state != state) {
371 f->state = state;
372 f->updated = jiffies;
373 notify = 1;
374 }
6681712d
DS		 375 if ((flags & NLM_F_APPEND) &&
376 is_multicast_ether_addr(f->eth_addr)) {
377 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex);
378
379 if (rc < 0)
380 return rc;
381 notify |= rc;
382 }
d342894c 383 } else {
384 if (!(flags & NLM_F_CREATE))
385 return -ENOENT;
386
387 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
388 return -ENOSPC;
389
390 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
391 f = kmalloc(sizeof(*f), GFP_ATOMIC);
392 if (!f)
393 return -ENOMEM;
394
395 notify = 1;
6681712d
DS		 396 f->remote.remote_ip = ip;
397 f->remote.remote_port = port;
398 f->remote.remote_vni = vni;
399 f->remote.remote_ifindex = ifindex;
400 f->remote.remote_next = NULL;
d342894c 401 f->state = state;
402 f->updated = f->used = jiffies;
403 memcpy(f->eth_addr, mac, ETH_ALEN);
404
405 ++vxlan->addrcnt;
406 hlist_add_head_rcu(&f->hlist,
407 vxlan_fdb_head(vxlan, mac));
408 }
409
410 if (notify)
411 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
412
413 return 0;
414}
415
6681712d
DS		 416void vxlan_fdb_free(struct rcu_head *head)
417{
418 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
419
420 while (f->remote.remote_next) {
421 struct vxlan_rdst *rd = f->remote.remote_next;
422
423 f->remote.remote_next = rd->remote_next;
424 kfree(rd);
425 }
426 kfree(f);
427}
428
d342894c 429static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
430{
431 netdev_dbg(vxlan->dev,
432 "delete %pM\n", f->eth_addr);
433
434 --vxlan->addrcnt;
435 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
436
437 hlist_del_rcu(&f->hlist);
6681712d 438 call_rcu(&f->rcu, vxlan_fdb_free);
d342894c 439}
440
441/* Add static entry (via netlink) */
442static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
443 struct net_device *dev,
444 const unsigned char *addr, u16 flags)
445{
446 struct vxlan_dev *vxlan = netdev_priv(dev);
6681712d 447 struct net *net = dev_net(vxlan->dev);
d342894c 448 __be32 ip;
6681712d 449 u32 port, vni, ifindex;
d342894c 450 int err;
451
452 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
453 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
454 ndm->ndm_state);
455 return -EINVAL;
456 }
457
458 if (tb[NDA_DST] == NULL)
459 return -EINVAL;
460
461 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
462 return -EAFNOSUPPORT;
463
464 ip = nla_get_be32(tb[NDA_DST]);
465
6681712d
DS		 466 if (tb[NDA_PORT]) {
467 if (nla_len(tb[NDA_PORT]) != sizeof(u32))
468 return -EINVAL;
469 port = nla_get_u32(tb[NDA_PORT]);
470 } else
471 port = vxlan_port;
472
473 if (tb[NDA_VNI]) {
474 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
475 return -EINVAL;
476 vni = nla_get_u32(tb[NDA_VNI]);
477 } else
478 vni = vxlan->vni;
479
480 if (tb[NDA_IFINDEX]) {
481 struct net_device *dev;
482
483 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
484 return -EINVAL;
485 ifindex = nla_get_u32(tb[NDA_IFINDEX]);
486 dev = dev_get_by_index(net, ifindex);
487 if (!dev)
488 return -EADDRNOTAVAIL;
489 dev_put(dev);
490 } else
491 ifindex = 0;
492
d342894c 493 spin_lock_bh(&vxlan->hash_lock);
6681712d
DS		 494 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags, port,
495 vni, ifindex);
d342894c 496 spin_unlock_bh(&vxlan->hash_lock);
497
498 return err;
499}
500
501/* Delete entry (via netlink) */
1690be63
VY		 502static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
503 struct net_device *dev,
d342894c 504 const unsigned char *addr)
505{
506 struct vxlan_dev *vxlan = netdev_priv(dev);
507 struct vxlan_fdb *f;
508 int err = -ENOENT;
509
510 spin_lock_bh(&vxlan->hash_lock);
511 f = vxlan_find_mac(vxlan, addr);
512 if (f) {
513 vxlan_fdb_destroy(vxlan, f);
514 err = 0;
515 }
516 spin_unlock_bh(&vxlan->hash_lock);
517
518 return err;
519}
520
521/* Dump forwarding table */
522static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
523 struct net_device *dev, int idx)
524{
525 struct vxlan_dev *vxlan = netdev_priv(dev);
526 unsigned int h;
527
528 for (h = 0; h < FDB_HASH_SIZE; ++h) {
529 struct vxlan_fdb *f;
d342894c 530 int err;
531
b67bfe0d 532 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
6681712d
DS		 533 struct vxlan_rdst *rd;
534 for (rd = &f->remote; rd; rd = rd->remote_next) {
535 if (idx < cb->args[0])
536 goto skip;
537
538 err = vxlan_fdb_info(skb, vxlan, f,
539 NETLINK_CB(cb->skb).portid,
540 cb->nlh->nlmsg_seq,
541 RTM_NEWNEIGH,
542 NLM_F_MULTI, rd);
543 if (err < 0)
544 break;
d342894c 545skip:
6681712d
DS		 546 ++idx;
547 }
d342894c 548 }
549 }
550
551 return idx;
552}
553
554/* Watch incoming packets to learn mapping between Ethernet address
555 * and Tunnel endpoint.
556 */
557static void vxlan_snoop(struct net_device *dev,
558 __be32 src_ip, const u8 *src_mac)
559{
560 struct vxlan_dev *vxlan = netdev_priv(dev);
561 struct vxlan_fdb *f;
562 int err;
563
564 f = vxlan_find_mac(vxlan, src_mac);
565 if (likely(f)) {
566 f->used = jiffies;
6681712d 567 if (likely(f->remote.remote_ip == src_ip))
d342894c 568 return;
569
570 if (net_ratelimit())
571 netdev_info(dev,
572 "%pM migrated from %pI4 to %pI4\n",
6681712d 573 src_mac, &f->remote.remote_ip, &src_ip);
d342894c 574
6681712d 575 f->remote.remote_ip = src_ip;
d342894c 576 f->updated = jiffies;
577 } else {
578 /* learned new entry */
579 spin_lock(&vxlan->hash_lock);
580 err = vxlan_fdb_create(vxlan, src_mac, src_ip,
581 NUD_REACHABLE,
6681712d
DS		 582 NLM_F_EXCL|NLM_F_CREATE,
583 vxlan_port, vxlan->vni, 0);
d342894c 584 spin_unlock(&vxlan->hash_lock);
585 }
586}
587
588
589/* See if multicast group is already in use by other ID */
590static bool vxlan_group_used(struct vxlan_net *vn,
591 const struct vxlan_dev *this)
592{
593 const struct vxlan_dev *vxlan;
d342894c 594 unsigned h;
595
596 for (h = 0; h < VNI_HASH_SIZE; ++h)
b67bfe0d 597 hlist_for_each_entry(vxlan, &vn->vni_list[h], hlist) {
d342894c 598 if (vxlan == this)
599 continue;
600
601 if (!netif_running(vxlan->dev))
602 continue;
603
604 if (vxlan->gaddr == this->gaddr)
605 return true;
606 }
607
608 return false;
609}
610
611/* kernel equivalent to IP_ADD_MEMBERSHIP */
612static int vxlan_join_group(struct net_device *dev)
613{
614 struct vxlan_dev *vxlan = netdev_priv(dev);
615 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
616 struct sock *sk = vn->sock->sk;
617 struct ip_mreqn mreq = {
af9b078e
YB		 618 .imr_multiaddr.s_addr = vxlan->gaddr,
619 .imr_ifindex = vxlan->link,
d342894c 620 };
621 int err;
622
623 /* Already a member of group */
624 if (vxlan_group_used(vn, vxlan))
625 return 0;
626
627 /* Need to drop RTNL to call multicast join */
628 rtnl_unlock();
629 lock_sock(sk);
630 err = ip_mc_join_group(sk, &mreq);
631 release_sock(sk);
632 rtnl_lock();
633
634 return err;
635}
636
637
638/* kernel equivalent to IP_DROP_MEMBERSHIP */
639static int vxlan_leave_group(struct net_device *dev)
640{
641 struct vxlan_dev *vxlan = netdev_priv(dev);
642 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
643 int err = 0;
644 struct sock *sk = vn->sock->sk;
645 struct ip_mreqn mreq = {
af9b078e
YB		 646 .imr_multiaddr.s_addr = vxlan->gaddr,
647 .imr_ifindex = vxlan->link,
d342894c 648 };
649
650 /* Only leave group when last vxlan is done. */
651 if (vxlan_group_used(vn, vxlan))
652 return 0;
653
654 /* Need to drop RTNL to call multicast leave */
655 rtnl_unlock();
656 lock_sock(sk);
657 err = ip_mc_leave_group(sk, &mreq);
658 release_sock(sk);
659 rtnl_lock();
660
661 return err;
662}
663
664/* Callback from net/ipv4/udp.c to receive packets */
665static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
666{
667 struct iphdr *oip;
668 struct vxlanhdr *vxh;
669 struct vxlan_dev *vxlan;
670 struct vxlan_stats *stats;
671 __u32 vni;
672 int err;
673
674 /* pop off outer UDP header */
675 __skb_pull(skb, sizeof(struct udphdr));
676
677 /* Need Vxlan and inner Ethernet header to be present */
678 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
679 goto error;
680
681 /* Drop packets with reserved bits set */
682 vxh = (struct vxlanhdr *) skb->data;
683 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
684 (vxh->vx_vni & htonl(0xff))) {
685 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
686 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
687 goto error;
688 }
689
690 __skb_pull(skb, sizeof(struct vxlanhdr));
d342894c 691
692 /* Is this VNI defined? */
693 vni = ntohl(vxh->vx_vni) >> 8;
694 vxlan = vxlan_find_vni(sock_net(sk), vni);
695 if (!vxlan) {
696 netdev_dbg(skb->dev, "unknown vni %d\n", vni);
697 goto drop;
698 }
699
700 if (!pskb_may_pull(skb, ETH_HLEN)) {
701 vxlan->dev->stats.rx_length_errors++;
702 vxlan->dev->stats.rx_errors++;
703 goto drop;
704 }
705
e4f67add
DS		 706 skb_reset_mac_header(skb);
707
d342894c 708 /* Re-examine inner Ethernet packet */
709 oip = ip_hdr(skb);
710 skb->protocol = eth_type_trans(skb, vxlan->dev);
d342894c 711
712 /* Ignore packet loops (and multicast echo) */
713 if (compare_ether_addr(eth_hdr(skb)->h_source,
714 vxlan->dev->dev_addr) == 0)
715 goto drop;
716
e4f67add 717 if (vxlan->flags & VXLAN_F_LEARN)
d342894c 718 vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source);
719
720 __skb_tunnel_rx(skb, vxlan->dev);
721 skb_reset_network_header(skb);
0afb1666
JG		 722
723 /* If the NIC driver gave us an encapsulated packet with
724 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
725 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
726 * for us. Otherwise force the upper layers to verify it.
727 */
728 if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
729 !(vxlan->dev->features & NETIF_F_RXCSUM))
730 skb->ip_summed = CHECKSUM_NONE;
731
732 skb->encapsulation = 0;
d342894c 733
734 err = IP_ECN_decapsulate(oip, skb);
735 if (unlikely(err)) {
736 if (log_ecn_error)
737 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
738 &oip->saddr, oip->tos);
739 if (err > 1) {
740 ++vxlan->dev->stats.rx_frame_errors;
741 ++vxlan->dev->stats.rx_errors;
742 goto drop;
743 }
744 }
745
746 stats = this_cpu_ptr(vxlan->stats);
747 u64_stats_update_begin(&stats->syncp);
748 stats->rx_packets++;
749 stats->rx_bytes += skb->len;
750 u64_stats_update_end(&stats->syncp);
751
752 netif_rx(skb);
753
754 return 0;
755error:
756 /* Put UDP header back */
757 __skb_push(skb, sizeof(struct udphdr));
758
759 return 1;
760drop:
761 /* Consume bad packet */
762 kfree_skb(skb);
763 return 0;
764}
765
e4f67add
DS		 766static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
767{
768 struct vxlan_dev *vxlan = netdev_priv(dev);
769 struct arphdr *parp;
770 u8 *arpptr, *sha;
771 __be32 sip, tip;
772 struct neighbour *n;
773
774 if (dev->flags & IFF_NOARP)
775 goto out;
776
777 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
778 dev->stats.tx_dropped++;
779 goto out;
780 }
781 parp = arp_hdr(skb);
782
783 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
784 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
785 parp->ar_pro != htons(ETH_P_IP) ||
786 parp->ar_op != htons(ARPOP_REQUEST) ||
787 parp->ar_hln != dev->addr_len ||
788 parp->ar_pln != 4)
789 goto out;
790 arpptr = (u8 *)parp + sizeof(struct arphdr);
791 sha = arpptr;
792 arpptr += dev->addr_len; /* sha */
793 memcpy(&sip, arpptr, sizeof(sip));
794 arpptr += sizeof(sip);
795 arpptr += dev->addr_len; /* tha */
796 memcpy(&tip, arpptr, sizeof(tip));
797
798 if (ipv4_is_loopback(tip) ||
799 ipv4_is_multicast(tip))
800 goto out;
801
802 n = neigh_lookup(&arp_tbl, &tip, dev);
803
804 if (n) {
805 struct vxlan_dev *vxlan = netdev_priv(dev);
806 struct vxlan_fdb *f;
807 struct sk_buff *reply;
808
809 if (!(n->nud_state & NUD_CONNECTED)) {
810 neigh_release(n);
811 goto out;
812 }
813
814 f = vxlan_find_mac(vxlan, n->ha);
6681712d 815 if (f && f->remote.remote_ip == htonl(INADDR_ANY)) {
e4f67add
DS		 816 /* bridge-local neighbor */
817 neigh_release(n);
818 goto out;
819 }
820
821 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
822 n->ha, sha);
823
824 neigh_release(n);
825
826 skb_reset_mac_header(reply);
827 __skb_pull(reply, skb_network_offset(reply));
828 reply->ip_summed = CHECKSUM_UNNECESSARY;
829 reply->pkt_type = PACKET_HOST;
830
831 if (netif_rx_ni(reply) == NET_RX_DROP)
832 dev->stats.rx_dropped++;
833 } else if (vxlan->flags & VXLAN_F_L3MISS)
834 vxlan_ip_miss(dev, tip);
835out:
836 consume_skb(skb);
837 return NETDEV_TX_OK;
838}
839
840static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
841{
842 struct vxlan_dev *vxlan = netdev_priv(dev);
843 struct neighbour *n;
844 struct iphdr *pip;
845
846 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
847 return false;
848
849 n = NULL;
850 switch (ntohs(eth_hdr(skb)->h_proto)) {
851 case ETH_P_IP:
852 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
853 return false;
854 pip = ip_hdr(skb);
855 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
856 break;
857 default:
858 return false;
859 }
860
861 if (n) {
862 bool diff;
863
864 diff = compare_ether_addr(eth_hdr(skb)->h_dest, n->ha) != 0;
865 if (diff) {
866 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
867 dev->addr_len);
868 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
869 }
870 neigh_release(n);
871 return diff;
872 } else if (vxlan->flags & VXLAN_F_L3MISS)
873 vxlan_ip_miss(dev, pip->daddr);
874 return false;
875}
876
d342894c 877/* Extract dsfield from inner protocol */
878static inline u8 vxlan_get_dsfield(const struct iphdr *iph,
879 const struct sk_buff *skb)
880{
881 if (skb->protocol == htons(ETH_P_IP))
882 return iph->tos;
883 else if (skb->protocol == htons(ETH_P_IPV6))
884 return ipv6_get_dsfield((const struct ipv6hdr *)iph);
885 else
886 return 0;
887}
888
889/* Propogate ECN bits out */
890static inline u8 vxlan_ecn_encap(u8 tos,
891 const struct iphdr *iph,
892 const struct sk_buff *skb)
893{
894 u8 inner = vxlan_get_dsfield(iph, skb);
895
896 return INET_ECN_encapsulate(tos, inner);
897}
898
1cad8715 899static void vxlan_sock_free(struct sk_buff *skb)
900{
901 sock_put(skb->sk);
902}
903
904/* On transmit, associate with the tunnel socket */
905static void vxlan_set_owner(struct net_device *dev, struct sk_buff *skb)
906{
907 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
908 struct sock *sk = vn->sock->sk;
909
910 skb_orphan(skb);
911 sock_hold(sk);
912 skb->sk = sk;
913 skb->destructor = vxlan_sock_free;
914}
915
05f47d69 916/* Compute source port for outgoing packet
917 * first choice to use L4 flow hash since it will spread
918 * better and maybe available from hardware
919 * secondary choice is to use jhash on the Ethernet header
920 */
921static u16 vxlan_src_port(const struct vxlan_dev *vxlan, struct sk_buff *skb)
922{
923 unsigned int range = (vxlan->port_max - vxlan->port_min) + 1;
924 u32 hash;
925
926 hash = skb_get_rxhash(skb);
927 if (!hash)
928 hash = jhash(skb->data, 2 * ETH_ALEN,
929 (__force u32) skb->protocol);
930
931 return (((u64) hash * range) >> 32) + vxlan->port_min;
932}
933
05c0db08
PS		 934static int handle_offloads(struct sk_buff *skb)
935{
936 if (skb_is_gso(skb)) {
937 int err = skb_unclone(skb, GFP_ATOMIC);
938 if (unlikely(err))
939 return err;
940
941 skb_shinfo(skb)->gso_type |= (SKB_GSO_UDP_TUNNEL | SKB_GSO_UDP);
942 } else if (skb->ip_summed != CHECKSUM_PARTIAL)
943 skb->ip_summed = CHECKSUM_NONE;
944
945 return 0;
946}
947
6681712d
DS		 948static netdev_tx_t vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
949 struct vxlan_rdst *rdst, bool did_rsc)
d342894c 950{
951 struct vxlan_dev *vxlan = netdev_priv(dev);
952 struct rtable *rt;
d342894c 953 const struct iphdr *old_iph;
954 struct iphdr *iph;
955 struct vxlanhdr *vxh;
956 struct udphdr *uh;
957 struct flowi4 fl4;
d342894c 958 unsigned int pkt_len = skb->len;
d342894c 959 __be32 dst;
6681712d
DS		 960 __u16 src_port, dst_port;
961 u32 vni;
d342894c 962 __be16 df = 0;
963 __u8 tos, ttl;
d342894c 964
6681712d
DS		 965 dst_port = rdst->remote_port ? rdst->remote_port : vxlan_port;
966 vni = rdst->remote_vni;
967 dst = rdst->remote_ip;
e4f67add
DS		 968
969 if (!dst) {
970 if (did_rsc) {
971 __skb_pull(skb, skb_network_offset(skb));
972 skb->ip_summed = CHECKSUM_NONE;
973 skb->pkt_type = PACKET_HOST;
974
975 /* short-circuited back to local bridge */
976 if (netif_rx(skb) == NET_RX_SUCCESS) {
977 struct vxlan_stats *stats =
978 this_cpu_ptr(vxlan->stats);
979
980 u64_stats_update_begin(&stats->syncp);
981 stats->tx_packets++;
982 stats->tx_bytes += pkt_len;
983 u64_stats_update_end(&stats->syncp);
984 } else {
985 dev->stats.tx_errors++;
986 dev->stats.tx_aborted_errors++;
987 }
988 return NETDEV_TX_OK;
989 }
ef59febe 990 goto drop;
e4f67add 991 }
ef59febe 992
d6727fe3
JG		 993 if (!skb->encapsulation) {
994 skb_reset_inner_headers(skb);
995 skb->encapsulation = 1;
996 }
997
d342894c 998 /* Need space for new headers (invalidates iph ptr) */
999 if (skb_cow_head(skb, VXLAN_HEADROOM))
1000 goto drop;
1001
d342894c 1002 old_iph = ip_hdr(skb);
1003
d342894c 1004 ttl = vxlan->ttl;
1005 if (!ttl && IN_MULTICAST(ntohl(dst)))
1006 ttl = 1;
1007
1008 tos = vxlan->tos;
1009 if (tos == 1)
1010 tos = vxlan_get_dsfield(old_iph, skb);
1011
05f47d69 1012 src_port = vxlan_src_port(vxlan, skb);
d342894c 1013
ca78f181 1014 memset(&fl4, 0, sizeof(fl4));
6681712d 1015 fl4.flowi4_oif = rdst->remote_ifindex;
ca78f181 1016 fl4.flowi4_tos = RT_TOS(tos);
1017 fl4.daddr = dst;
1018 fl4.saddr = vxlan->saddr;
1019
1020 rt = ip_route_output_key(dev_net(dev), &fl4);
d342894c 1021 if (IS_ERR(rt)) {
1022 netdev_dbg(dev, "no route to %pI4\n", &dst);
1023 dev->stats.tx_carrier_errors++;
1024 goto tx_error;
1025 }
1026
1027 if (rt->dst.dev == dev) {
1028 netdev_dbg(dev, "circular route to %pI4\n", &dst);
1029 ip_rt_put(rt);
1030 dev->stats.collisions++;
1031 goto tx_error;
1032 }
1033
1034 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1035 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1036 IPSKB_REROUTED);
1037 skb_dst_drop(skb);
1038 skb_dst_set(skb, &rt->dst);
1039
1040 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1041 vxh->vx_flags = htonl(VXLAN_FLAGS);
6681712d 1042 vxh->vx_vni = htonl(vni << 8);
d342894c 1043
1044 __skb_push(skb, sizeof(*uh));
1045 skb_reset_transport_header(skb);
1046 uh = udp_hdr(skb);
1047
6681712d 1048 uh->dest = htons(dst_port);
05f47d69 1049 uh->source = htons(src_port);
d342894c 1050
1051 uh->len = htons(skb->len);
1052 uh->check = 0;
1053
1054 __skb_push(skb, sizeof(*iph));
1055 skb_reset_network_header(skb);
1056 iph = ip_hdr(skb);
1057 iph->version = 4;
1058 iph->ihl = sizeof(struct iphdr) >> 2;
1059 iph->frag_off = df;
1060 iph->protocol = IPPROTO_UDP;
1061 iph->tos = vxlan_ecn_encap(tos, old_iph, skb);
ca78f181 1062 iph->daddr = dst;
d342894c 1063 iph->saddr = fl4.saddr;
1064 iph->ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
8dc98eb2 1065 tunnel_ip_select_ident(skb, old_iph, &rt->dst);
d342894c 1066
88c4c066
ZM		 1067 nf_reset(skb);
1068
1cad8715 1069 vxlan_set_owner(dev, skb);
1070
05c0db08
PS		 1071 if (handle_offloads(skb))
1072 goto drop;
d342894c 1073
6aed0c8b 1074 iptunnel_xmit(skb, dev);
d342894c 1075 return NETDEV_TX_OK;
1076
1077drop:
1078 dev->stats.tx_dropped++;
1079 goto tx_free;
1080
1081tx_error:
1082 dev->stats.tx_errors++;
1083tx_free:
1084 dev_kfree_skb(skb);
1085 return NETDEV_TX_OK;
1086}
1087
6681712d
DS		 1088/* Transmit local packets over Vxlan
1089 *
1090 * Outer IP header inherits ECN and DF from inner header.
1091 * Outer UDP destination is the VXLAN assigned port.
1092 * source port is based on hash of flow
1093 */
1094static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1095{
1096 struct vxlan_dev *vxlan = netdev_priv(dev);
1097 struct ethhdr *eth;
1098 bool did_rsc = false;
1099 struct vxlan_rdst group, *rdst0, *rdst;
1100 struct vxlan_fdb *f;
1101 int rc1, rc;
1102
1103 skb_reset_mac_header(skb);
1104 eth = eth_hdr(skb);
1105
1106 if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP)
1107 return arp_reduce(dev, skb);
1108 else if ((vxlan->flags&VXLAN_F_RSC) && ntohs(eth->h_proto) == ETH_P_IP)
1109 did_rsc = route_shortcircuit(dev, skb);
1110
1111 f = vxlan_find_mac(vxlan, eth->h_dest);
1112 if (f == NULL) {
1113 did_rsc = false;
1114 group.remote_port = vxlan_port;
1115 group.remote_vni = vxlan->vni;
1116 group.remote_ip = vxlan->gaddr;
1117 group.remote_ifindex = vxlan->link;
1118 group.remote_next = 0;
1119 rdst0 = &group;
1120
1121 if (group.remote_ip == htonl(INADDR_ANY) &&
1122 (vxlan->flags & VXLAN_F_L2MISS) &&
1123 !is_multicast_ether_addr(eth->h_dest))
1124 vxlan_fdb_miss(vxlan, eth->h_dest);
1125 } else
1126 rdst0 = &f->remote;
1127
1128 rc = NETDEV_TX_OK;
1129
1130 /* if there are multiple destinations, send copies */
1131 for (rdst = rdst0->remote_next; rdst; rdst = rdst->remote_next) {
1132 struct sk_buff *skb1;
1133
1134 skb1 = skb_clone(skb, GFP_ATOMIC);
1135 rc1 = vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1136 if (rc == NETDEV_TX_OK)
1137 rc = rc1;
1138 }
1139
1140 rc1 = vxlan_xmit_one(skb, dev, rdst0, did_rsc);
1141 if (rc == NETDEV_TX_OK)
1142 rc = rc1;
1143 return rc;
1144}
1145
d342894c 1146/* Walk the forwarding table and purge stale entries */
1147static void vxlan_cleanup(unsigned long arg)
1148{
1149 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1150 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1151 unsigned int h;
1152
1153 if (!netif_running(vxlan->dev))
1154 return;
1155
1156 spin_lock_bh(&vxlan->hash_lock);
1157 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1158 struct hlist_node *p, *n;
1159 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1160 struct vxlan_fdb *f
1161 = container_of(p, struct vxlan_fdb, hlist);
1162 unsigned long timeout;
1163
3c172868 1164 if (f->state & NUD_PERMANENT)
d342894c 1165 continue;
1166
1167 timeout = f->used + vxlan->age_interval * HZ;
1168 if (time_before_eq(timeout, jiffies)) {
1169 netdev_dbg(vxlan->dev,
1170 "garbage collect %pM\n",
1171 f->eth_addr);
1172 f->state = NUD_STALE;
1173 vxlan_fdb_destroy(vxlan, f);
1174 } else if (time_before(timeout, next_timer))
1175 next_timer = timeout;
1176 }
1177 }
1178 spin_unlock_bh(&vxlan->hash_lock);
1179
1180 mod_timer(&vxlan->age_timer, next_timer);
1181}
1182
1183/* Setup stats when device is created */
1184static int vxlan_init(struct net_device *dev)
1185{
1186 struct vxlan_dev *vxlan = netdev_priv(dev);
1187
1188 vxlan->stats = alloc_percpu(struct vxlan_stats);
1189 if (!vxlan->stats)
1190 return -ENOMEM;
1191
1192 return 0;
1193}
1194
1195/* Start ageing timer and join group when device is brought up */
1196static int vxlan_open(struct net_device *dev)
1197{
1198 struct vxlan_dev *vxlan = netdev_priv(dev);
1199 int err;
1200
1201 if (vxlan->gaddr) {
1202 err = vxlan_join_group(dev);
1203 if (err)
1204 return err;
1205 }
1206
1207 if (vxlan->age_interval)
1208 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
1209
1210 return 0;
1211}
1212
1213/* Purge the forwarding table */
1214static void vxlan_flush(struct vxlan_dev *vxlan)
1215{
1216 unsigned h;
1217
1218 spin_lock_bh(&vxlan->hash_lock);
1219 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1220 struct hlist_node *p, *n;
1221 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1222 struct vxlan_fdb *f
1223 = container_of(p, struct vxlan_fdb, hlist);
1224 vxlan_fdb_destroy(vxlan, f);
1225 }
1226 }
1227 spin_unlock_bh(&vxlan->hash_lock);
1228}
1229
1230/* Cleanup timer and forwarding table on shutdown */
1231static int vxlan_stop(struct net_device *dev)
1232{
1233 struct vxlan_dev *vxlan = netdev_priv(dev);
1234
1235 if (vxlan->gaddr)
1236 vxlan_leave_group(dev);
1237
1238 del_timer_sync(&vxlan->age_timer);
1239
1240 vxlan_flush(vxlan);
1241
1242 return 0;
1243}
1244
1245/* Merge per-cpu statistics */
1246static struct rtnl_link_stats64 *vxlan_stats64(struct net_device *dev,
1247 struct rtnl_link_stats64 *stats)
1248{
1249 struct vxlan_dev *vxlan = netdev_priv(dev);
1250 struct vxlan_stats tmp, sum = { 0 };
1251 unsigned int cpu;
1252
1253 for_each_possible_cpu(cpu) {
1254 unsigned int start;
1255 const struct vxlan_stats *stats
1256 = per_cpu_ptr(vxlan->stats, cpu);
1257
1258 do {
1259 start = u64_stats_fetch_begin_bh(&stats->syncp);
1260 memcpy(&tmp, stats, sizeof(tmp));
1261 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1262
1263 sum.tx_bytes += tmp.tx_bytes;
1264 sum.tx_packets += tmp.tx_packets;
1265 sum.rx_bytes += tmp.rx_bytes;
1266 sum.rx_packets += tmp.rx_packets;
1267 }
1268
1269 stats->tx_bytes = sum.tx_bytes;
1270 stats->tx_packets = sum.tx_packets;
1271 stats->rx_bytes = sum.rx_bytes;
1272 stats->rx_packets = sum.rx_packets;
1273
1274 stats->multicast = dev->stats.multicast;
1275 stats->rx_length_errors = dev->stats.rx_length_errors;
1276 stats->rx_frame_errors = dev->stats.rx_frame_errors;
1277 stats->rx_errors = dev->stats.rx_errors;
1278
1279 stats->tx_dropped = dev->stats.tx_dropped;
1280 stats->tx_carrier_errors = dev->stats.tx_carrier_errors;
1281 stats->tx_aborted_errors = dev->stats.tx_aborted_errors;
1282 stats->collisions = dev->stats.collisions;
1283 stats->tx_errors = dev->stats.tx_errors;
1284
1285 return stats;
1286}
1287
1288/* Stub, nothing needs to be done. */
1289static void vxlan_set_multicast_list(struct net_device *dev)
1290{
1291}
1292
1293static const struct net_device_ops vxlan_netdev_ops = {
1294 .ndo_init = vxlan_init,
1295 .ndo_open = vxlan_open,
1296 .ndo_stop = vxlan_stop,
1297 .ndo_start_xmit = vxlan_xmit,
1298 .ndo_get_stats64 = vxlan_stats64,
1299 .ndo_set_rx_mode = vxlan_set_multicast_list,
1300 .ndo_change_mtu = eth_change_mtu,
1301 .ndo_validate_addr = eth_validate_addr,
1302 .ndo_set_mac_address = eth_mac_addr,
1303 .ndo_fdb_add = vxlan_fdb_add,
1304 .ndo_fdb_del = vxlan_fdb_delete,
1305 .ndo_fdb_dump = vxlan_fdb_dump,
1306};
1307
1308/* Info for udev, that this is a virtual tunnel endpoint */
1309static struct device_type vxlan_type = {
1310 .name = "vxlan",
1311};
1312
1313static void vxlan_free(struct net_device *dev)
1314{
1315 struct vxlan_dev *vxlan = netdev_priv(dev);
1316
1317 free_percpu(vxlan->stats);
1318 free_netdev(dev);
1319}
1320
1321/* Initialize the device structure. */
1322static void vxlan_setup(struct net_device *dev)
1323{
1324 struct vxlan_dev *vxlan = netdev_priv(dev);
1325 unsigned h;
05f47d69 1326 int low, high;
d342894c 1327
1328 eth_hw_addr_random(dev);
1329 ether_setup(dev);
2840bf22 1330 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
d342894c 1331
1332 dev->netdev_ops = &vxlan_netdev_ops;
1333 dev->destructor = vxlan_free;
1334 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
1335
1336 dev->tx_queue_len = 0;
1337 dev->features |= NETIF_F_LLTX;
1338 dev->features |= NETIF_F_NETNS_LOCAL;
d6727fe3 1339 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
0afb1666 1340 dev->features |= NETIF_F_RXCSUM;
05c0db08 1341 dev->features |= NETIF_F_GSO_SOFTWARE;
0afb1666
JG		 1342
1343 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
05c0db08 1344 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
d342894c 1345 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
6602d007 1346 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
d342894c 1347
1348 spin_lock_init(&vxlan->hash_lock);
1349
1350 init_timer_deferrable(&vxlan->age_timer);
1351 vxlan->age_timer.function = vxlan_cleanup;
1352 vxlan->age_timer.data = (unsigned long) vxlan;
1353
05f47d69 1354 inet_get_local_port_range(&low, &high);
1355 vxlan->port_min = low;
1356 vxlan->port_max = high;
1357
d342894c 1358 vxlan->dev = dev;
1359
1360 for (h = 0; h < FDB_HASH_SIZE; ++h)
1361 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
1362}
1363
1364static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
1365 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
1366 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1367 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
1368 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1369 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
1370 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
1371 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
1372 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
1373 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
05f47d69 1374 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
e4f67add
DS		 1375 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
1376 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
1377 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
1378 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
d342894c 1379};
1380
1381static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
1382{
1383 if (tb[IFLA_ADDRESS]) {
1384 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1385 pr_debug("invalid link address (not ethernet)\n");
1386 return -EINVAL;
1387 }
1388
1389 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1390 pr_debug("invalid all zero ethernet address\n");
1391 return -EADDRNOTAVAIL;
1392 }
1393 }
1394
1395 if (!data)
1396 return -EINVAL;
1397
1398 if (data[IFLA_VXLAN_ID]) {
1399 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
1400 if (id >= VXLAN_VID_MASK)
1401 return -ERANGE;
1402 }
1403
1404 if (data[IFLA_VXLAN_GROUP]) {
1405 __be32 gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1406 if (!IN_MULTICAST(ntohl(gaddr))) {
1407 pr_debug("group address is not IPv4 multicast\n");
1408 return -EADDRNOTAVAIL;
1409 }
1410 }
05f47d69 1411
1412 if (data[IFLA_VXLAN_PORT_RANGE]) {
1413 const struct ifla_vxlan_port_range *p
1414 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1415
1416 if (ntohs(p->high) < ntohs(p->low)) {
1417 pr_debug("port range %u .. %u not valid\n",
1418 ntohs(p->low), ntohs(p->high));
1419 return -EINVAL;
1420 }
1421 }
1422
d342894c 1423 return 0;
1424}
1425
1b13c97f
YB		 1426static void vxlan_get_drvinfo(struct net_device *netdev,
1427 struct ethtool_drvinfo *drvinfo)
1428{
1429 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
1430 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
1431}
1432
1433static const struct ethtool_ops vxlan_ethtool_ops = {
1434 .get_drvinfo = vxlan_get_drvinfo,
1435 .get_link = ethtool_op_get_link,
1436};
1437
d342894c 1438static int vxlan_newlink(struct net *net, struct net_device *dev,
1439 struct nlattr *tb[], struct nlattr *data[])
1440{
1441 struct vxlan_dev *vxlan = netdev_priv(dev);
1442 __u32 vni;
1443 int err;
1444
1445 if (!data[IFLA_VXLAN_ID])
1446 return -EINVAL;
1447
1448 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1449 if (vxlan_find_vni(net, vni)) {
1450 pr_info("duplicate VNI %u\n", vni);
1451 return -EEXIST;
1452 }
1453 vxlan->vni = vni;
1454
1455 if (data[IFLA_VXLAN_GROUP])
1456 vxlan->gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1457
1458 if (data[IFLA_VXLAN_LOCAL])
1459 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1460
34e02aa1 1461 if (data[IFLA_VXLAN_LINK] &&
1462 (vxlan->link = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
1463 struct net_device *lowerdev
1464 = __dev_get_by_index(net, vxlan->link);
1465
1466 if (!lowerdev) {
1467 pr_info("ifindex %d does not exist\n", vxlan->link);
1468 return -ENODEV;
1469 }
d342894c 1470
34e02aa1 1471 if (!tb[IFLA_MTU])
d342894c 1472 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
1ba56fb4
AD		 1473
1474 /* update header length based on lower device */
1475 dev->hard_header_len = lowerdev->hard_header_len +
1476 VXLAN_HEADROOM;
d342894c 1477 }
1478
1479 if (data[IFLA_VXLAN_TOS])
1480 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1481
afb97186
VB		 1482 if (data[IFLA_VXLAN_TTL])
1483 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
1484
d342894c 1485 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
e4f67add 1486 vxlan->flags |= VXLAN_F_LEARN;
d342894c 1487
1488 if (data[IFLA_VXLAN_AGEING])
1489 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1490 else
1491 vxlan->age_interval = FDB_AGE_DEFAULT;
1492
e4f67add
DS		 1493 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
1494 vxlan->flags |= VXLAN_F_PROXY;
1495
1496 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
1497 vxlan->flags |= VXLAN_F_RSC;
1498
1499 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
1500 vxlan->flags |= VXLAN_F_L2MISS;
1501
1502 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
1503 vxlan->flags |= VXLAN_F_L3MISS;
1504
d342894c 1505 if (data[IFLA_VXLAN_LIMIT])
1506 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1507
05f47d69 1508 if (data[IFLA_VXLAN_PORT_RANGE]) {
1509 const struct ifla_vxlan_port_range *p
1510 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1511 vxlan->port_min = ntohs(p->low);
1512 vxlan->port_max = ntohs(p->high);
1513 }
1514
1b13c97f
YB		 1515 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
1516
d342894c 1517 err = register_netdevice(dev);
1518 if (!err)
1519 hlist_add_head_rcu(&vxlan->hlist, vni_head(net, vxlan->vni));
1520
1521 return err;
1522}
1523
1524static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1525{
1526 struct vxlan_dev *vxlan = netdev_priv(dev);
1527
1528 hlist_del_rcu(&vxlan->hlist);
1529
1530 unregister_netdevice_queue(dev, head);
1531}
1532
1533static size_t vxlan_get_size(const struct net_device *dev)
1534{
1535
1536 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1537 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1538 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1539 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1540 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1541 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1542 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
e4f67add
DS		 1543 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
1544 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
1545 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
1546 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
d342894c 1547 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1548 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
05f47d69 1549 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
d342894c 1550 0;
1551}
1552
1553static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1554{
1555 const struct vxlan_dev *vxlan = netdev_priv(dev);
05f47d69 1556 struct ifla_vxlan_port_range ports = {
1557 .low = htons(vxlan->port_min),
1558 .high = htons(vxlan->port_max),
1559 };
d342894c 1560
1561 if (nla_put_u32(skb, IFLA_VXLAN_ID, vxlan->vni))
1562 goto nla_put_failure;
1563
7c41c42c 1564 if (vxlan->gaddr && nla_put_be32(skb, IFLA_VXLAN_GROUP, vxlan->gaddr))
d342894c 1565 goto nla_put_failure;
1566
1567 if (vxlan->link && nla_put_u32(skb, IFLA_VXLAN_LINK, vxlan->link))
1568 goto nla_put_failure;
1569
7c41c42c 1570 if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
d342894c 1571 goto nla_put_failure;
1572
1573 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1574 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
e4f67add
DS		 1575 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
1576 !!(vxlan->flags & VXLAN_F_LEARN)) ||
1577 nla_put_u8(skb, IFLA_VXLAN_PROXY,
1578 !!(vxlan->flags & VXLAN_F_PROXY)) ||
1579 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
1580 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
1581 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
1582 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
1583 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
d342894c 1584 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1585 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax))
1586 goto nla_put_failure;
1587
05f47d69 1588 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
1589 goto nla_put_failure;
1590
d342894c 1591 return 0;
1592
1593nla_put_failure:
1594 return -EMSGSIZE;
1595}
1596
1597static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1598 .kind = "vxlan",
1599 .maxtype = IFLA_VXLAN_MAX,
1600 .policy = vxlan_policy,
1601 .priv_size = sizeof(struct vxlan_dev),
1602 .setup = vxlan_setup,
1603 .validate = vxlan_validate,
1604 .newlink = vxlan_newlink,
1605 .dellink = vxlan_dellink,
1606 .get_size = vxlan_get_size,
1607 .fill_info = vxlan_fill_info,
1608};
1609
1610static __net_init int vxlan_init_net(struct net *net)
1611{
1612 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1613 struct sock *sk;
1614 struct sockaddr_in vxlan_addr = {
1615 .sin_family = AF_INET,
1616 .sin_addr.s_addr = htonl(INADDR_ANY),
1617 };
1618 int rc;
1619 unsigned h;
1620
1621 /* Create UDP socket for encapsulation receive. */
1622 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vn->sock);
1623 if (rc < 0) {
1624 pr_debug("UDP socket create failed\n");
1625 return rc;
1626 }
bfe1b9b1 1627 /* Put in proper namespace */
1628 sk = vn->sock->sk;
1629 sk_change_net(sk, net);
d342894c 1630
1631 vxlan_addr.sin_port = htons(vxlan_port);
1632
1633 rc = kernel_bind(vn->sock, (struct sockaddr *) &vxlan_addr,
1634 sizeof(vxlan_addr));
1635 if (rc < 0) {
1636 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1637 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
bfe1b9b1 1638 sk_release_kernel(sk);
d342894c 1639 vn->sock = NULL;
1640 return rc;
1641 }
1642
1643 /* Disable multicast loopback */
d342894c 1644 inet_sk(sk)->mc_loop = 0;
1645
1646 /* Mark socket as an encapsulation socket. */
1647 udp_sk(sk)->encap_type = 1;
1648 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1649 udp_encap_enable();
1650
1651 for (h = 0; h < VNI_HASH_SIZE; ++h)
1652 INIT_HLIST_HEAD(&vn->vni_list[h]);
1653
1654 return 0;
1655}
1656
1657static __net_exit void vxlan_exit_net(struct net *net)
1658{
1659 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
9cb6cb7e
ZM		 1660 struct vxlan_dev *vxlan;
1661 unsigned h;
1662
1663 rtnl_lock();
1664 for (h = 0; h < VNI_HASH_SIZE; ++h)
1665 hlist_for_each_entry(vxlan, &vn->vni_list[h], hlist)
1666 dev_close(vxlan->dev);
1667 rtnl_unlock();
d342894c 1668
1669 if (vn->sock) {
bfe1b9b1 1670 sk_release_kernel(vn->sock->sk);
d342894c 1671 vn->sock = NULL;
1672 }
1673}
1674
1675static struct pernet_operations vxlan_net_ops = {
1676 .init = vxlan_init_net,
1677 .exit = vxlan_exit_net,
1678 .id = &vxlan_net_id,
1679 .size = sizeof(struct vxlan_net),
1680};
1681
1682static int __init vxlan_init_module(void)
1683{
1684 int rc;
1685
1686 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1687
1688 rc = register_pernet_device(&vxlan_net_ops);
1689 if (rc)
1690 goto out1;
1691
1692 rc = rtnl_link_register(&vxlan_link_ops);
1693 if (rc)
1694 goto out2;
1695
1696 return 0;
1697
1698out2:
1699 unregister_pernet_device(&vxlan_net_ops);
1700out1:
1701 return rc;
1702}
1703module_init(vxlan_init_module);
1704
1705static void __exit vxlan_cleanup_module(void)
1706{
1707 rtnl_link_unregister(&vxlan_link_ops);
1708 unregister_pernet_device(&vxlan_net_ops);
6681712d 1709 rcu_barrier();
d342894c 1710}
1711module_exit(vxlan_cleanup_module);
1712
1713MODULE_LICENSE("GPL");
1714MODULE_VERSION(VXLAN_VERSION);
1715MODULE_AUTHOR("Stephen Hemminger <shemminger@vyatta.com>");
1716MODULE_ALIAS_RTNL_LINK("vxlan");
Linux 6.x block layer and io_uring tree(s)