projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
packet: minor: add generic tpacket_uhdr to access packet headers
[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
d342894c 104/* Pseudo network device */
105struct vxlan_dev {
106 struct hlist_node hlist;
107 struct net_device *dev;
d342894c 108 __u32 vni; /* virtual network id */
109 __be32 gaddr; /* multicast group */
110 __be32 saddr; /* source address */
111 unsigned int link; /* link to multicast over */
05f47d69 112 __u16 port_min; /* source port range */
113 __u16 port_max;
d342894c 114 __u8 tos; /* TOS override */
115 __u8 ttl;
e4f67add 116 u32 flags; /* VXLAN_F_* below */
d342894c 117
118 unsigned long age_interval;
119 struct timer_list age_timer;
120 spinlock_t hash_lock;
121 unsigned int addrcnt;
122 unsigned int addrmax;
d342894c 123
124 struct hlist_head fdb_head[FDB_HASH_SIZE];
125};
126
e4f67add
DS		 127#define VXLAN_F_LEARN 0x01
128#define VXLAN_F_PROXY 0x02
129#define VXLAN_F_RSC 0x04
130#define VXLAN_F_L2MISS 0x08
131#define VXLAN_F_L3MISS 0x10
132
d342894c 133/* salt for hash table */
134static u32 vxlan_salt __read_mostly;
135
136static inline struct hlist_head *vni_head(struct net *net, u32 id)
137{
138 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
139
140 return &vn->vni_list[hash_32(id, VNI_HASH_BITS)];
141}
142
143/* Look up VNI in a per net namespace table */
144static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
145{
146 struct vxlan_dev *vxlan;
d342894c 147
b67bfe0d 148 hlist_for_each_entry_rcu(vxlan, vni_head(net, id), hlist) {
d342894c 149 if (vxlan->vni == id)
150 return vxlan;
151 }
152
153 return NULL;
154}
155
156/* Fill in neighbour message in skbuff. */
157static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
158 const struct vxlan_fdb *fdb,
6681712d
DS		 159 u32 portid, u32 seq, int type, unsigned int flags,
160 const struct vxlan_rdst *rdst)
d342894c 161{
162 unsigned long now = jiffies;
163 struct nda_cacheinfo ci;
164 struct nlmsghdr *nlh;
165 struct ndmsg *ndm;
e4f67add 166 bool send_ip, send_eth;
d342894c 167
168 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
169 if (nlh == NULL)
170 return -EMSGSIZE;
171
172 ndm = nlmsg_data(nlh);
173 memset(ndm, 0, sizeof(*ndm));
e4f67add
DS		 174
175 send_eth = send_ip = true;
176
177 if (type == RTM_GETNEIGH) {
178 ndm->ndm_family = AF_INET;
6681712d 179 send_ip = rdst->remote_ip != htonl(INADDR_ANY);
e4f67add
DS		 180 send_eth = !is_zero_ether_addr(fdb->eth_addr);
181 } else
182 ndm->ndm_family = AF_BRIDGE;
d342894c 183 ndm->ndm_state = fdb->state;
184 ndm->ndm_ifindex = vxlan->dev->ifindex;
185 ndm->ndm_flags = NTF_SELF;
186 ndm->ndm_type = NDA_DST;
187
e4f67add 188 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
d342894c 189 goto nla_put_failure;
190
6681712d
DS		 191 if (send_ip && nla_put_be32(skb, NDA_DST, rdst->remote_ip))
192 goto nla_put_failure;
193
194 if (rdst->remote_port && rdst->remote_port != vxlan_port &&
195 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
196 goto nla_put_failure;
197 if (rdst->remote_vni != vxlan->vni &&
198 nla_put_be32(skb, NDA_VNI, rdst->remote_vni))
199 goto nla_put_failure;
200 if (rdst->remote_ifindex &&
201 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
d342894c 202 goto nla_put_failure;
203
204 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
205 ci.ndm_confirmed = 0;
206 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
207 ci.ndm_refcnt = 0;
208
209 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
210 goto nla_put_failure;
211
212 return nlmsg_end(skb, nlh);
213
214nla_put_failure:
215 nlmsg_cancel(skb, nlh);
216 return -EMSGSIZE;
217}
218
219static inline size_t vxlan_nlmsg_size(void)
220{
221 return NLMSG_ALIGN(sizeof(struct ndmsg))
222 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
223 + nla_total_size(sizeof(__be32)) /* NDA_DST */
6681712d
DS		 224 + nla_total_size(sizeof(__be32)) /* NDA_PORT */
225 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
226 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
d342894c 227 + nla_total_size(sizeof(struct nda_cacheinfo));
228}
229
230static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
231 const struct vxlan_fdb *fdb, int type)
232{
233 struct net *net = dev_net(vxlan->dev);
234 struct sk_buff *skb;
235 int err = -ENOBUFS;
236
237 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
238 if (skb == NULL)
239 goto errout;
240
6681712d 241 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, &fdb->remote);
d342894c 242 if (err < 0) {
243 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
244 WARN_ON(err == -EMSGSIZE);
245 kfree_skb(skb);
246 goto errout;
247 }
248
249 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
250 return;
251errout:
252 if (err < 0)
253 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
254}
255
e4f67add
DS		 256static void vxlan_ip_miss(struct net_device *dev, __be32 ipa)
257{
258 struct vxlan_dev *vxlan = netdev_priv(dev);
259 struct vxlan_fdb f;
260
261 memset(&f, 0, sizeof f);
262 f.state = NUD_STALE;
6681712d
DS		 263 f.remote.remote_ip = ipa; /* goes to NDA_DST */
264 f.remote.remote_vni = VXLAN_N_VID;
e4f67add
DS		 265
266 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
267}
268
269static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
270{
271 struct vxlan_fdb f;
272
273 memset(&f, 0, sizeof f);
274 f.state = NUD_STALE;
275 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
276
277 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
278}
279
d342894c 280/* Hash Ethernet address */
281static u32 eth_hash(const unsigned char *addr)
282{
283 u64 value = get_unaligned((u64 *)addr);
284
285 /* only want 6 bytes */
286#ifdef __BIG_ENDIAN
d342894c 287 value >>= 16;
321fb991 288#else
289 value <<= 16;
d342894c 290#endif
291 return hash_64(value, FDB_HASH_BITS);
292}
293
294/* Hash chain to use given mac address */
295static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
296 const u8 *mac)
297{
298 return &vxlan->fdb_head[eth_hash(mac)];
299}
300
301/* Look up Ethernet address in forwarding table */
302static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
303 const u8 *mac)
304
305{
306 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
307 struct vxlan_fdb *f;
d342894c 308
b67bfe0d 309 hlist_for_each_entry_rcu(f, head, hlist) {
d342894c 310 if (compare_ether_addr(mac, f->eth_addr) == 0)
311 return f;
312 }
313
314 return NULL;
315}
316
6681712d
DS		 317/* Add/update destinations for multicast */
318static int vxlan_fdb_append(struct vxlan_fdb *f,
319 __be32 ip, __u32 port, __u32 vni, __u32 ifindex)
320{
321 struct vxlan_rdst *rd_prev, *rd;
322
323 rd_prev = NULL;
324 for (rd = &f->remote; rd; rd = rd->remote_next) {
325 if (rd->remote_ip == ip &&
326 rd->remote_port == port &&
327 rd->remote_vni == vni &&
328 rd->remote_ifindex == ifindex)
329 return 0;
330 rd_prev = rd;
331 }
332 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
333 if (rd == NULL)
334 return -ENOBUFS;
335 rd->remote_ip = ip;
336 rd->remote_port = port;
337 rd->remote_vni = vni;
338 rd->remote_ifindex = ifindex;
339 rd->remote_next = NULL;
340 rd_prev->remote_next = rd;
341 return 1;
342}
343
d342894c 344/* Add new entry to forwarding table -- assumes lock held */
345static int vxlan_fdb_create(struct vxlan_dev *vxlan,
346 const u8 *mac, __be32 ip,
6681712d
DS		 347 __u16 state, __u16 flags,
348 __u32 port, __u32 vni, __u32 ifindex)
d342894c 349{
350 struct vxlan_fdb *f;
351 int notify = 0;
352
353 f = vxlan_find_mac(vxlan, mac);
354 if (f) {
355 if (flags & NLM_F_EXCL) {
356 netdev_dbg(vxlan->dev,
357 "lost race to create %pM\n", mac);
358 return -EEXIST;
359 }
360 if (f->state != state) {
361 f->state = state;
362 f->updated = jiffies;
363 notify = 1;
364 }
6681712d
DS		 365 if ((flags & NLM_F_APPEND) &&
366 is_multicast_ether_addr(f->eth_addr)) {
367 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex);
368
369 if (rc < 0)
370 return rc;
371 notify |= rc;
372 }
d342894c 373 } else {
374 if (!(flags & NLM_F_CREATE))
375 return -ENOENT;
376
377 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
378 return -ENOSPC;
379
380 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
381 f = kmalloc(sizeof(*f), GFP_ATOMIC);
382 if (!f)
383 return -ENOMEM;
384
385 notify = 1;
6681712d
DS		 386 f->remote.remote_ip = ip;
387 f->remote.remote_port = port;
388 f->remote.remote_vni = vni;
389 f->remote.remote_ifindex = ifindex;
390 f->remote.remote_next = NULL;
d342894c 391 f->state = state;
392 f->updated = f->used = jiffies;
393 memcpy(f->eth_addr, mac, ETH_ALEN);
394
395 ++vxlan->addrcnt;
396 hlist_add_head_rcu(&f->hlist,
397 vxlan_fdb_head(vxlan, mac));
398 }
399
400 if (notify)
401 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
402
403 return 0;
404}
405
6706c82e 406static void vxlan_fdb_free(struct rcu_head *head)
6681712d
DS		 407{
408 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
409
410 while (f->remote.remote_next) {
411 struct vxlan_rdst *rd = f->remote.remote_next;
412
413 f->remote.remote_next = rd->remote_next;
414 kfree(rd);
415 }
416 kfree(f);
417}
418
d342894c 419static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
420{
421 netdev_dbg(vxlan->dev,
422 "delete %pM\n", f->eth_addr);
423
424 --vxlan->addrcnt;
425 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
426
427 hlist_del_rcu(&f->hlist);
6681712d 428 call_rcu(&f->rcu, vxlan_fdb_free);
d342894c 429}
430
431/* Add static entry (via netlink) */
432static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
433 struct net_device *dev,
434 const unsigned char *addr, u16 flags)
435{
436 struct vxlan_dev *vxlan = netdev_priv(dev);
6681712d 437 struct net *net = dev_net(vxlan->dev);
d342894c 438 __be32 ip;
6681712d 439 u32 port, vni, ifindex;
d342894c 440 int err;
441
442 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
443 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
444 ndm->ndm_state);
445 return -EINVAL;
446 }
447
448 if (tb[NDA_DST] == NULL)
449 return -EINVAL;
450
451 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
452 return -EAFNOSUPPORT;
453
454 ip = nla_get_be32(tb[NDA_DST]);
455
6681712d
DS		 456 if (tb[NDA_PORT]) {
457 if (nla_len(tb[NDA_PORT]) != sizeof(u32))
458 return -EINVAL;
459 port = nla_get_u32(tb[NDA_PORT]);
460 } else
461 port = vxlan_port;
462
463 if (tb[NDA_VNI]) {
464 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
465 return -EINVAL;
466 vni = nla_get_u32(tb[NDA_VNI]);
467 } else
468 vni = vxlan->vni;
469
470 if (tb[NDA_IFINDEX]) {
5abb0029 471 struct net_device *tdev;
6681712d
DS		 472
473 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
474 return -EINVAL;
475 ifindex = nla_get_u32(tb[NDA_IFINDEX]);
5abb0029
PS		 476 tdev = dev_get_by_index(net, ifindex);
477 if (!tdev)
6681712d 478 return -EADDRNOTAVAIL;
5abb0029 479 dev_put(tdev);
6681712d
DS		 480 } else
481 ifindex = 0;
482
d342894c 483 spin_lock_bh(&vxlan->hash_lock);
6681712d
DS		 484 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags, port,
485 vni, ifindex);
d342894c 486 spin_unlock_bh(&vxlan->hash_lock);
487
488 return err;
489}
490
491/* Delete entry (via netlink) */
1690be63
VY		 492static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
493 struct net_device *dev,
d342894c 494 const unsigned char *addr)
495{
496 struct vxlan_dev *vxlan = netdev_priv(dev);
497 struct vxlan_fdb *f;
498 int err = -ENOENT;
499
500 spin_lock_bh(&vxlan->hash_lock);
501 f = vxlan_find_mac(vxlan, addr);
502 if (f) {
503 vxlan_fdb_destroy(vxlan, f);
504 err = 0;
505 }
506 spin_unlock_bh(&vxlan->hash_lock);
507
508 return err;
509}
510
511/* Dump forwarding table */
512static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
513 struct net_device *dev, int idx)
514{
515 struct vxlan_dev *vxlan = netdev_priv(dev);
516 unsigned int h;
517
518 for (h = 0; h < FDB_HASH_SIZE; ++h) {
519 struct vxlan_fdb *f;
d342894c 520 int err;
521
b67bfe0d 522 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
6681712d
DS		 523 struct vxlan_rdst *rd;
524 for (rd = &f->remote; rd; rd = rd->remote_next) {
525 if (idx < cb->args[0])
526 goto skip;
527
528 err = vxlan_fdb_info(skb, vxlan, f,
529 NETLINK_CB(cb->skb).portid,
530 cb->nlh->nlmsg_seq,
531 RTM_NEWNEIGH,
532 NLM_F_MULTI, rd);
533 if (err < 0)
534 break;
d342894c 535skip:
6681712d
DS		 536 ++idx;
537 }
d342894c 538 }
539 }
540
541 return idx;
542}
543
544/* Watch incoming packets to learn mapping between Ethernet address
545 * and Tunnel endpoint.
546 */
547static void vxlan_snoop(struct net_device *dev,
548 __be32 src_ip, const u8 *src_mac)
549{
550 struct vxlan_dev *vxlan = netdev_priv(dev);
551 struct vxlan_fdb *f;
552 int err;
553
554 f = vxlan_find_mac(vxlan, src_mac);
555 if (likely(f)) {
556 f->used = jiffies;
6681712d 557 if (likely(f->remote.remote_ip == src_ip))
d342894c 558 return;
559
560 if (net_ratelimit())
561 netdev_info(dev,
562 "%pM migrated from %pI4 to %pI4\n",
6681712d 563 src_mac, &f->remote.remote_ip, &src_ip);
d342894c 564
6681712d 565 f->remote.remote_ip = src_ip;
d342894c 566 f->updated = jiffies;
567 } else {
568 /* learned new entry */
569 spin_lock(&vxlan->hash_lock);
570 err = vxlan_fdb_create(vxlan, src_mac, src_ip,
571 NUD_REACHABLE,
6681712d
DS		 572 NLM_F_EXCL|NLM_F_CREATE,
573 vxlan_port, vxlan->vni, 0);
d342894c 574 spin_unlock(&vxlan->hash_lock);
575 }
576}
577
578
579/* See if multicast group is already in use by other ID */
580static bool vxlan_group_used(struct vxlan_net *vn,
581 const struct vxlan_dev *this)
582{
583 const struct vxlan_dev *vxlan;
d342894c 584 unsigned h;
585
586 for (h = 0; h < VNI_HASH_SIZE; ++h)
b67bfe0d 587 hlist_for_each_entry(vxlan, &vn->vni_list[h], hlist) {
d342894c 588 if (vxlan == this)
589 continue;
590
591 if (!netif_running(vxlan->dev))
592 continue;
593
594 if (vxlan->gaddr == this->gaddr)
595 return true;
596 }
597
598 return false;
599}
600
601/* kernel equivalent to IP_ADD_MEMBERSHIP */
602static int vxlan_join_group(struct net_device *dev)
603{
604 struct vxlan_dev *vxlan = netdev_priv(dev);
605 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
606 struct sock *sk = vn->sock->sk;
607 struct ip_mreqn mreq = {
af9b078e
YB		 608 .imr_multiaddr.s_addr = vxlan->gaddr,
609 .imr_ifindex = vxlan->link,
d342894c 610 };
611 int err;
612
613 /* Already a member of group */
614 if (vxlan_group_used(vn, vxlan))
615 return 0;
616
617 /* Need to drop RTNL to call multicast join */
618 rtnl_unlock();
619 lock_sock(sk);
620 err = ip_mc_join_group(sk, &mreq);
621 release_sock(sk);
622 rtnl_lock();
623
624 return err;
625}
626
627
628/* kernel equivalent to IP_DROP_MEMBERSHIP */
629static int vxlan_leave_group(struct net_device *dev)
630{
631 struct vxlan_dev *vxlan = netdev_priv(dev);
632 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
633 int err = 0;
634 struct sock *sk = vn->sock->sk;
635 struct ip_mreqn mreq = {
af9b078e
YB		 636 .imr_multiaddr.s_addr = vxlan->gaddr,
637 .imr_ifindex = vxlan->link,
d342894c 638 };
639
640 /* Only leave group when last vxlan is done. */
641 if (vxlan_group_used(vn, vxlan))
642 return 0;
643
644 /* Need to drop RTNL to call multicast leave */
645 rtnl_unlock();
646 lock_sock(sk);
647 err = ip_mc_leave_group(sk, &mreq);
648 release_sock(sk);
649 rtnl_lock();
650
651 return err;
652}
653
654/* Callback from net/ipv4/udp.c to receive packets */
655static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
656{
657 struct iphdr *oip;
658 struct vxlanhdr *vxh;
659 struct vxlan_dev *vxlan;
e8171045 660 struct pcpu_tstats *stats;
d342894c 661 __u32 vni;
662 int err;
663
664 /* pop off outer UDP header */
665 __skb_pull(skb, sizeof(struct udphdr));
666
667 /* Need Vxlan and inner Ethernet header to be present */
668 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
669 goto error;
670
671 /* Drop packets with reserved bits set */
672 vxh = (struct vxlanhdr *) skb->data;
673 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
674 (vxh->vx_vni & htonl(0xff))) {
675 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
676 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
677 goto error;
678 }
679
680 __skb_pull(skb, sizeof(struct vxlanhdr));
d342894c 681
682 /* Is this VNI defined? */
683 vni = ntohl(vxh->vx_vni) >> 8;
684 vxlan = vxlan_find_vni(sock_net(sk), vni);
685 if (!vxlan) {
686 netdev_dbg(skb->dev, "unknown vni %d\n", vni);
687 goto drop;
688 }
689
690 if (!pskb_may_pull(skb, ETH_HLEN)) {
691 vxlan->dev->stats.rx_length_errors++;
692 vxlan->dev->stats.rx_errors++;
693 goto drop;
694 }
695
e4f67add
DS		 696 skb_reset_mac_header(skb);
697
d342894c 698 /* Re-examine inner Ethernet packet */
699 oip = ip_hdr(skb);
700 skb->protocol = eth_type_trans(skb, vxlan->dev);
d342894c 701
702 /* Ignore packet loops (and multicast echo) */
703 if (compare_ether_addr(eth_hdr(skb)->h_source,
704 vxlan->dev->dev_addr) == 0)
705 goto drop;
706
e4f67add 707 if (vxlan->flags & VXLAN_F_LEARN)
d342894c 708 vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source);
709
710 __skb_tunnel_rx(skb, vxlan->dev);
711 skb_reset_network_header(skb);
0afb1666
JG		 712
713 /* If the NIC driver gave us an encapsulated packet with
714 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
715 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
716 * for us. Otherwise force the upper layers to verify it.
717 */
718 if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
719 !(vxlan->dev->features & NETIF_F_RXCSUM))
720 skb->ip_summed = CHECKSUM_NONE;
721
722 skb->encapsulation = 0;
d342894c 723
724 err = IP_ECN_decapsulate(oip, skb);
725 if (unlikely(err)) {
726 if (log_ecn_error)
727 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
728 &oip->saddr, oip->tos);
729 if (err > 1) {
730 ++vxlan->dev->stats.rx_frame_errors;
731 ++vxlan->dev->stats.rx_errors;
732 goto drop;
733 }
734 }
735
e8171045 736 stats = this_cpu_ptr(vxlan->dev->tstats);
d342894c 737 u64_stats_update_begin(&stats->syncp);
738 stats->rx_packets++;
739 stats->rx_bytes += skb->len;
740 u64_stats_update_end(&stats->syncp);
741
742 netif_rx(skb);
743
744 return 0;
745error:
746 /* Put UDP header back */
747 __skb_push(skb, sizeof(struct udphdr));
748
749 return 1;
750drop:
751 /* Consume bad packet */
752 kfree_skb(skb);
753 return 0;
754}
755
e4f67add
DS		 756static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
757{
758 struct vxlan_dev *vxlan = netdev_priv(dev);
759 struct arphdr *parp;
760 u8 *arpptr, *sha;
761 __be32 sip, tip;
762 struct neighbour *n;
763
764 if (dev->flags & IFF_NOARP)
765 goto out;
766
767 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
768 dev->stats.tx_dropped++;
769 goto out;
770 }
771 parp = arp_hdr(skb);
772
773 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
774 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
775 parp->ar_pro != htons(ETH_P_IP) ||
776 parp->ar_op != htons(ARPOP_REQUEST) ||
777 parp->ar_hln != dev->addr_len ||
778 parp->ar_pln != 4)
779 goto out;
780 arpptr = (u8 *)parp + sizeof(struct arphdr);
781 sha = arpptr;
782 arpptr += dev->addr_len; /* sha */
783 memcpy(&sip, arpptr, sizeof(sip));
784 arpptr += sizeof(sip);
785 arpptr += dev->addr_len; /* tha */
786 memcpy(&tip, arpptr, sizeof(tip));
787
788 if (ipv4_is_loopback(tip) ||
789 ipv4_is_multicast(tip))
790 goto out;
791
792 n = neigh_lookup(&arp_tbl, &tip, dev);
793
794 if (n) {
e4f67add
DS		 795 struct vxlan_fdb *f;
796 struct sk_buff *reply;
797
798 if (!(n->nud_state & NUD_CONNECTED)) {
799 neigh_release(n);
800 goto out;
801 }
802
803 f = vxlan_find_mac(vxlan, n->ha);
6681712d 804 if (f && f->remote.remote_ip == htonl(INADDR_ANY)) {
e4f67add
DS		 805 /* bridge-local neighbor */
806 neigh_release(n);
807 goto out;
808 }
809
810 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
811 n->ha, sha);
812
813 neigh_release(n);
814
815 skb_reset_mac_header(reply);
816 __skb_pull(reply, skb_network_offset(reply));
817 reply->ip_summed = CHECKSUM_UNNECESSARY;
818 reply->pkt_type = PACKET_HOST;
819
820 if (netif_rx_ni(reply) == NET_RX_DROP)
821 dev->stats.rx_dropped++;
822 } else if (vxlan->flags & VXLAN_F_L3MISS)
823 vxlan_ip_miss(dev, tip);
824out:
825 consume_skb(skb);
826 return NETDEV_TX_OK;
827}
828
829static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
830{
831 struct vxlan_dev *vxlan = netdev_priv(dev);
832 struct neighbour *n;
833 struct iphdr *pip;
834
835 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
836 return false;
837
838 n = NULL;
839 switch (ntohs(eth_hdr(skb)->h_proto)) {
840 case ETH_P_IP:
841 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
842 return false;
843 pip = ip_hdr(skb);
844 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
845 break;
846 default:
847 return false;
848 }
849
850 if (n) {
851 bool diff;
852
853 diff = compare_ether_addr(eth_hdr(skb)->h_dest, n->ha) != 0;
854 if (diff) {
855 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
856 dev->addr_len);
857 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
858 }
859 neigh_release(n);
860 return diff;
861 } else if (vxlan->flags & VXLAN_F_L3MISS)
862 vxlan_ip_miss(dev, pip->daddr);
863 return false;
864}
865
1cad8715 866static void vxlan_sock_free(struct sk_buff *skb)
867{
868 sock_put(skb->sk);
869}
870
871/* On transmit, associate with the tunnel socket */
872static void vxlan_set_owner(struct net_device *dev, struct sk_buff *skb)
873{
874 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
875 struct sock *sk = vn->sock->sk;
876
877 skb_orphan(skb);
878 sock_hold(sk);
879 skb->sk = sk;
880 skb->destructor = vxlan_sock_free;
881}
882
05f47d69 883/* Compute source port for outgoing packet
884 * first choice to use L4 flow hash since it will spread
885 * better and maybe available from hardware
886 * secondary choice is to use jhash on the Ethernet header
887 */
888static u16 vxlan_src_port(const struct vxlan_dev *vxlan, struct sk_buff *skb)
889{
890 unsigned int range = (vxlan->port_max - vxlan->port_min) + 1;
891 u32 hash;
892
893 hash = skb_get_rxhash(skb);
894 if (!hash)
895 hash = jhash(skb->data, 2 * ETH_ALEN,
896 (__force u32) skb->protocol);
897
898 return (((u64) hash * range) >> 32) + vxlan->port_min;
899}
900
05c0db08
PS		 901static int handle_offloads(struct sk_buff *skb)
902{
903 if (skb_is_gso(skb)) {
904 int err = skb_unclone(skb, GFP_ATOMIC);
905 if (unlikely(err))
906 return err;
907
908 skb_shinfo(skb)->gso_type |= (SKB_GSO_UDP_TUNNEL | SKB_GSO_UDP);
909 } else if (skb->ip_summed != CHECKSUM_PARTIAL)
910 skb->ip_summed = CHECKSUM_NONE;
911
912 return 0;
913}
914
9dcc71e1
SS		 915/* Bypass encapsulation if the destination is local */
916static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
917 struct vxlan_dev *dst_vxlan)
918{
919 struct pcpu_tstats *tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
920 struct pcpu_tstats *rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
921
922 skb->pkt_type = PACKET_HOST;
923 skb->encapsulation = 0;
924 skb->dev = dst_vxlan->dev;
925 __skb_pull(skb, skb_network_offset(skb));
926
927 if (dst_vxlan->flags & VXLAN_F_LEARN)
9d9f163c
MR		 928 vxlan_snoop(skb->dev, htonl(INADDR_LOOPBACK),
929 eth_hdr(skb)->h_source);
9dcc71e1
SS		 930
931 u64_stats_update_begin(&tx_stats->syncp);
932 tx_stats->tx_packets++;
933 tx_stats->tx_bytes += skb->len;
934 u64_stats_update_end(&tx_stats->syncp);
935
936 if (netif_rx(skb) == NET_RX_SUCCESS) {
937 u64_stats_update_begin(&rx_stats->syncp);
938 rx_stats->rx_packets++;
939 rx_stats->rx_bytes += skb->len;
940 u64_stats_update_end(&rx_stats->syncp);
941 } else {
942 skb->dev->stats.rx_dropped++;
943 }
944}
945
6681712d
DS		 946static netdev_tx_t vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
947 struct vxlan_rdst *rdst, bool did_rsc)
d342894c 948{
949 struct vxlan_dev *vxlan = netdev_priv(dev);
950 struct rtable *rt;
d342894c 951 const struct iphdr *old_iph;
952 struct iphdr *iph;
953 struct vxlanhdr *vxh;
954 struct udphdr *uh;
955 struct flowi4 fl4;
d342894c 956 __be32 dst;
6681712d
DS		 957 __u16 src_port, dst_port;
958 u32 vni;
d342894c 959 __be16 df = 0;
960 __u8 tos, ttl;
d342894c 961
6681712d
DS		 962 dst_port = rdst->remote_port ? rdst->remote_port : vxlan_port;
963 vni = rdst->remote_vni;
964 dst = rdst->remote_ip;
e4f67add
DS		 965
966 if (!dst) {
967 if (did_rsc) {
e4f67add 968 /* short-circuited back to local bridge */
9dcc71e1 969 vxlan_encap_bypass(skb, vxlan, vxlan);
e4f67add
DS		 970 return NETDEV_TX_OK;
971 }
ef59febe 972 goto drop;
e4f67add 973 }
ef59febe 974
d6727fe3
JG		 975 if (!skb->encapsulation) {
976 skb_reset_inner_headers(skb);
977 skb->encapsulation = 1;
978 }
979
d342894c 980 /* Need space for new headers (invalidates iph ptr) */
981 if (skb_cow_head(skb, VXLAN_HEADROOM))
982 goto drop;
983
d342894c 984 old_iph = ip_hdr(skb);
985
d342894c 986 ttl = vxlan->ttl;
987 if (!ttl && IN_MULTICAST(ntohl(dst)))
988 ttl = 1;
989
990 tos = vxlan->tos;
991 if (tos == 1)
206aaafc 992 tos = ip_tunnel_get_dsfield(old_iph, skb);
d342894c 993
05f47d69 994 src_port = vxlan_src_port(vxlan, skb);
d342894c 995
ca78f181 996 memset(&fl4, 0, sizeof(fl4));
6681712d 997 fl4.flowi4_oif = rdst->remote_ifindex;
ca78f181 998 fl4.flowi4_tos = RT_TOS(tos);
999 fl4.daddr = dst;
1000 fl4.saddr = vxlan->saddr;
1001
1002 rt = ip_route_output_key(dev_net(dev), &fl4);
d342894c 1003 if (IS_ERR(rt)) {
1004 netdev_dbg(dev, "no route to %pI4\n", &dst);
1005 dev->stats.tx_carrier_errors++;
1006 goto tx_error;
1007 }
1008
1009 if (rt->dst.dev == dev) {
1010 netdev_dbg(dev, "circular route to %pI4\n", &dst);
1011 ip_rt_put(rt);
1012 dev->stats.collisions++;
1013 goto tx_error;
1014 }
1015
9dcc71e1 1016 /* Bypass encapsulation if the destination is local */
ab09a6d0
MR		 1017 if (rt->rt_flags & RTCF_LOCAL &&
1018 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
9dcc71e1
SS		 1019 struct vxlan_dev *dst_vxlan;
1020
1021 ip_rt_put(rt);
1022 dst_vxlan = vxlan_find_vni(dev_net(dev), vni);
1023 if (!dst_vxlan)
1024 goto tx_error;
1025 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1026 return NETDEV_TX_OK;
1027 }
1028
d342894c 1029 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1030 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1031 IPSKB_REROUTED);
1032 skb_dst_drop(skb);
1033 skb_dst_set(skb, &rt->dst);
1034
1035 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1036 vxh->vx_flags = htonl(VXLAN_FLAGS);
6681712d 1037 vxh->vx_vni = htonl(vni << 8);
d342894c 1038
1039 __skb_push(skb, sizeof(*uh));
1040 skb_reset_transport_header(skb);
1041 uh = udp_hdr(skb);
1042
6681712d 1043 uh->dest = htons(dst_port);
05f47d69 1044 uh->source = htons(src_port);
d342894c 1045
1046 uh->len = htons(skb->len);
1047 uh->check = 0;
1048
1049 __skb_push(skb, sizeof(*iph));
1050 skb_reset_network_header(skb);
1051 iph = ip_hdr(skb);
1052 iph->version = 4;
1053 iph->ihl = sizeof(struct iphdr) >> 2;
1054 iph->frag_off = df;
1055 iph->protocol = IPPROTO_UDP;
206aaafc 1056 iph->tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
ca78f181 1057 iph->daddr = dst;
d342894c 1058 iph->saddr = fl4.saddr;
1059 iph->ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
8dc98eb2 1060 tunnel_ip_select_ident(skb, old_iph, &rt->dst);
d342894c 1061
88c4c066
ZM		 1062 nf_reset(skb);
1063
1cad8715 1064 vxlan_set_owner(dev, skb);
1065
05c0db08
PS		 1066 if (handle_offloads(skb))
1067 goto drop;
d342894c 1068
6aed0c8b 1069 iptunnel_xmit(skb, dev);
d342894c 1070 return NETDEV_TX_OK;
1071
1072drop:
1073 dev->stats.tx_dropped++;
1074 goto tx_free;
1075
1076tx_error:
1077 dev->stats.tx_errors++;
1078tx_free:
1079 dev_kfree_skb(skb);
1080 return NETDEV_TX_OK;
1081}
1082
6681712d
DS		 1083/* Transmit local packets over Vxlan
1084 *
1085 * Outer IP header inherits ECN and DF from inner header.
1086 * Outer UDP destination is the VXLAN assigned port.
1087 * source port is based on hash of flow
1088 */
1089static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1090{
1091 struct vxlan_dev *vxlan = netdev_priv(dev);
1092 struct ethhdr *eth;
1093 bool did_rsc = false;
1094 struct vxlan_rdst group, *rdst0, *rdst;
1095 struct vxlan_fdb *f;
1096 int rc1, rc;
1097
1098 skb_reset_mac_header(skb);
1099 eth = eth_hdr(skb);
1100
1101 if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP)
1102 return arp_reduce(dev, skb);
1103 else if ((vxlan->flags&VXLAN_F_RSC) && ntohs(eth->h_proto) == ETH_P_IP)
1104 did_rsc = route_shortcircuit(dev, skb);
1105
1106 f = vxlan_find_mac(vxlan, eth->h_dest);
1107 if (f == NULL) {
1108 did_rsc = false;
1109 group.remote_port = vxlan_port;
1110 group.remote_vni = vxlan->vni;
1111 group.remote_ip = vxlan->gaddr;
1112 group.remote_ifindex = vxlan->link;
6706c82e 1113 group.remote_next = NULL;
6681712d
DS		 1114 rdst0 = &group;
1115
1116 if (group.remote_ip == htonl(INADDR_ANY) &&
1117 (vxlan->flags & VXLAN_F_L2MISS) &&
1118 !is_multicast_ether_addr(eth->h_dest))
1119 vxlan_fdb_miss(vxlan, eth->h_dest);
1120 } else
1121 rdst0 = &f->remote;
1122
1123 rc = NETDEV_TX_OK;
1124
1125 /* if there are multiple destinations, send copies */
1126 for (rdst = rdst0->remote_next; rdst; rdst = rdst->remote_next) {
1127 struct sk_buff *skb1;
1128
1129 skb1 = skb_clone(skb, GFP_ATOMIC);
1130 rc1 = vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1131 if (rc == NETDEV_TX_OK)
1132 rc = rc1;
1133 }
1134
1135 rc1 = vxlan_xmit_one(skb, dev, rdst0, did_rsc);
1136 if (rc == NETDEV_TX_OK)
1137 rc = rc1;
1138 return rc;
1139}
1140
d342894c 1141/* Walk the forwarding table and purge stale entries */
1142static void vxlan_cleanup(unsigned long arg)
1143{
1144 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1145 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1146 unsigned int h;
1147
1148 if (!netif_running(vxlan->dev))
1149 return;
1150
1151 spin_lock_bh(&vxlan->hash_lock);
1152 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1153 struct hlist_node *p, *n;
1154 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1155 struct vxlan_fdb *f
1156 = container_of(p, struct vxlan_fdb, hlist);
1157 unsigned long timeout;
1158
3c172868 1159 if (f->state & NUD_PERMANENT)
d342894c 1160 continue;
1161
1162 timeout = f->used + vxlan->age_interval * HZ;
1163 if (time_before_eq(timeout, jiffies)) {
1164 netdev_dbg(vxlan->dev,
1165 "garbage collect %pM\n",
1166 f->eth_addr);
1167 f->state = NUD_STALE;
1168 vxlan_fdb_destroy(vxlan, f);
1169 } else if (time_before(timeout, next_timer))
1170 next_timer = timeout;
1171 }
1172 }
1173 spin_unlock_bh(&vxlan->hash_lock);
1174
1175 mod_timer(&vxlan->age_timer, next_timer);
1176}
1177
1178/* Setup stats when device is created */
1179static int vxlan_init(struct net_device *dev)
1180{
e8171045
PS		 1181 dev->tstats = alloc_percpu(struct pcpu_tstats);
1182 if (!dev->tstats)
d342894c 1183 return -ENOMEM;
1184
1185 return 0;
1186}
1187
1188/* Start ageing timer and join group when device is brought up */
1189static int vxlan_open(struct net_device *dev)
1190{
1191 struct vxlan_dev *vxlan = netdev_priv(dev);
1192 int err;
1193
1194 if (vxlan->gaddr) {
1195 err = vxlan_join_group(dev);
1196 if (err)
1197 return err;
1198 }
1199
1200 if (vxlan->age_interval)
1201 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
1202
1203 return 0;
1204}
1205
1206/* Purge the forwarding table */
1207static void vxlan_flush(struct vxlan_dev *vxlan)
1208{
1209 unsigned h;
1210
1211 spin_lock_bh(&vxlan->hash_lock);
1212 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1213 struct hlist_node *p, *n;
1214 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1215 struct vxlan_fdb *f
1216 = container_of(p, struct vxlan_fdb, hlist);
1217 vxlan_fdb_destroy(vxlan, f);
1218 }
1219 }
1220 spin_unlock_bh(&vxlan->hash_lock);
1221}
1222
1223/* Cleanup timer and forwarding table on shutdown */
1224static int vxlan_stop(struct net_device *dev)
1225{
1226 struct vxlan_dev *vxlan = netdev_priv(dev);
1227
1228 if (vxlan->gaddr)
1229 vxlan_leave_group(dev);
1230
1231 del_timer_sync(&vxlan->age_timer);
1232
1233 vxlan_flush(vxlan);
1234
1235 return 0;
1236}
1237
d342894c 1238/* Stub, nothing needs to be done. */
1239static void vxlan_set_multicast_list(struct net_device *dev)
1240{
1241}
1242
1243static const struct net_device_ops vxlan_netdev_ops = {
1244 .ndo_init = vxlan_init,
1245 .ndo_open = vxlan_open,
1246 .ndo_stop = vxlan_stop,
1247 .ndo_start_xmit = vxlan_xmit,
e8171045 1248 .ndo_get_stats64 = ip_tunnel_get_stats64,
d342894c 1249 .ndo_set_rx_mode = vxlan_set_multicast_list,
1250 .ndo_change_mtu = eth_change_mtu,
1251 .ndo_validate_addr = eth_validate_addr,
1252 .ndo_set_mac_address = eth_mac_addr,
1253 .ndo_fdb_add = vxlan_fdb_add,
1254 .ndo_fdb_del = vxlan_fdb_delete,
1255 .ndo_fdb_dump = vxlan_fdb_dump,
1256};
1257
1258/* Info for udev, that this is a virtual tunnel endpoint */
1259static struct device_type vxlan_type = {
1260 .name = "vxlan",
1261};
1262
1263static void vxlan_free(struct net_device *dev)
1264{
e8171045 1265 free_percpu(dev->tstats);
d342894c 1266 free_netdev(dev);
1267}
1268
1269/* Initialize the device structure. */
1270static void vxlan_setup(struct net_device *dev)
1271{
1272 struct vxlan_dev *vxlan = netdev_priv(dev);
1273 unsigned h;
05f47d69 1274 int low, high;
d342894c 1275
1276 eth_hw_addr_random(dev);
1277 ether_setup(dev);
2840bf22 1278 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
d342894c 1279
1280 dev->netdev_ops = &vxlan_netdev_ops;
1281 dev->destructor = vxlan_free;
1282 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
1283
1284 dev->tx_queue_len = 0;
1285 dev->features |= NETIF_F_LLTX;
1286 dev->features |= NETIF_F_NETNS_LOCAL;
d6727fe3 1287 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
0afb1666 1288 dev->features |= NETIF_F_RXCSUM;
05c0db08 1289 dev->features |= NETIF_F_GSO_SOFTWARE;
0afb1666
JG		 1290
1291 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
05c0db08 1292 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
d342894c 1293 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
6602d007 1294 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
d342894c 1295
1296 spin_lock_init(&vxlan->hash_lock);
1297
1298 init_timer_deferrable(&vxlan->age_timer);
1299 vxlan->age_timer.function = vxlan_cleanup;
1300 vxlan->age_timer.data = (unsigned long) vxlan;
1301
05f47d69 1302 inet_get_local_port_range(&low, &high);
1303 vxlan->port_min = low;
1304 vxlan->port_max = high;
1305
d342894c 1306 vxlan->dev = dev;
1307
1308 for (h = 0; h < FDB_HASH_SIZE; ++h)
1309 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
1310}
1311
1312static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
1313 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
1314 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1315 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
1316 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1317 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
1318 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
1319 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
1320 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
1321 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
05f47d69 1322 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
e4f67add
DS		 1323 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
1324 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
1325 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
1326 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
d342894c 1327};
1328
1329static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
1330{
1331 if (tb[IFLA_ADDRESS]) {
1332 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1333 pr_debug("invalid link address (not ethernet)\n");
1334 return -EINVAL;
1335 }
1336
1337 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1338 pr_debug("invalid all zero ethernet address\n");
1339 return -EADDRNOTAVAIL;
1340 }
1341 }
1342
1343 if (!data)
1344 return -EINVAL;
1345
1346 if (data[IFLA_VXLAN_ID]) {
1347 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
1348 if (id >= VXLAN_VID_MASK)
1349 return -ERANGE;
1350 }
1351
1352 if (data[IFLA_VXLAN_GROUP]) {
1353 __be32 gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1354 if (!IN_MULTICAST(ntohl(gaddr))) {
1355 pr_debug("group address is not IPv4 multicast\n");
1356 return -EADDRNOTAVAIL;
1357 }
1358 }
05f47d69 1359
1360 if (data[IFLA_VXLAN_PORT_RANGE]) {
1361 const struct ifla_vxlan_port_range *p
1362 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1363
1364 if (ntohs(p->high) < ntohs(p->low)) {
1365 pr_debug("port range %u .. %u not valid\n",
1366 ntohs(p->low), ntohs(p->high));
1367 return -EINVAL;
1368 }
1369 }
1370
d342894c 1371 return 0;
1372}
1373
1b13c97f
YB		 1374static void vxlan_get_drvinfo(struct net_device *netdev,
1375 struct ethtool_drvinfo *drvinfo)
1376{
1377 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
1378 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
1379}
1380
1381static const struct ethtool_ops vxlan_ethtool_ops = {
1382 .get_drvinfo = vxlan_get_drvinfo,
1383 .get_link = ethtool_op_get_link,
1384};
1385
d342894c 1386static int vxlan_newlink(struct net *net, struct net_device *dev,
1387 struct nlattr *tb[], struct nlattr *data[])
1388{
1389 struct vxlan_dev *vxlan = netdev_priv(dev);
1390 __u32 vni;
1391 int err;
1392
1393 if (!data[IFLA_VXLAN_ID])
1394 return -EINVAL;
1395
1396 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1397 if (vxlan_find_vni(net, vni)) {
1398 pr_info("duplicate VNI %u\n", vni);
1399 return -EEXIST;
1400 }
1401 vxlan->vni = vni;
1402
1403 if (data[IFLA_VXLAN_GROUP])
1404 vxlan->gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1405
1406 if (data[IFLA_VXLAN_LOCAL])
1407 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1408
34e02aa1 1409 if (data[IFLA_VXLAN_LINK] &&
1410 (vxlan->link = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
1411 struct net_device *lowerdev
1412 = __dev_get_by_index(net, vxlan->link);
1413
1414 if (!lowerdev) {
1415 pr_info("ifindex %d does not exist\n", vxlan->link);
1416 return -ENODEV;
1417 }
d342894c 1418
34e02aa1 1419 if (!tb[IFLA_MTU])
d342894c 1420 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
1ba56fb4
AD		 1421
1422 /* update header length based on lower device */
1423 dev->hard_header_len = lowerdev->hard_header_len +
1424 VXLAN_HEADROOM;
d342894c 1425 }
1426
1427 if (data[IFLA_VXLAN_TOS])
1428 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1429
afb97186
VB		 1430 if (data[IFLA_VXLAN_TTL])
1431 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
1432
d342894c 1433 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
e4f67add 1434 vxlan->flags |= VXLAN_F_LEARN;
d342894c 1435
1436 if (data[IFLA_VXLAN_AGEING])
1437 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1438 else
1439 vxlan->age_interval = FDB_AGE_DEFAULT;
1440
e4f67add
DS		 1441 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
1442 vxlan->flags |= VXLAN_F_PROXY;
1443
1444 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
1445 vxlan->flags |= VXLAN_F_RSC;
1446
1447 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
1448 vxlan->flags |= VXLAN_F_L2MISS;
1449
1450 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
1451 vxlan->flags |= VXLAN_F_L3MISS;
1452
d342894c 1453 if (data[IFLA_VXLAN_LIMIT])
1454 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1455
05f47d69 1456 if (data[IFLA_VXLAN_PORT_RANGE]) {
1457 const struct ifla_vxlan_port_range *p
1458 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1459 vxlan->port_min = ntohs(p->low);
1460 vxlan->port_max = ntohs(p->high);
1461 }
1462
1b13c97f
YB		 1463 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
1464
d342894c 1465 err = register_netdevice(dev);
1466 if (!err)
1467 hlist_add_head_rcu(&vxlan->hlist, vni_head(net, vxlan->vni));
1468
1469 return err;
1470}
1471
1472static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1473{
1474 struct vxlan_dev *vxlan = netdev_priv(dev);
1475
1476 hlist_del_rcu(&vxlan->hlist);
1477
1478 unregister_netdevice_queue(dev, head);
1479}
1480
1481static size_t vxlan_get_size(const struct net_device *dev)
1482{
1483
1484 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1485 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1486 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1487 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1488 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1489 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1490 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
e4f67add
DS		 1491 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
1492 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
1493 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
1494 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
d342894c 1495 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1496 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
05f47d69 1497 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
d342894c 1498 0;
1499}
1500
1501static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1502{
1503 const struct vxlan_dev *vxlan = netdev_priv(dev);
05f47d69 1504 struct ifla_vxlan_port_range ports = {
1505 .low = htons(vxlan->port_min),
1506 .high = htons(vxlan->port_max),
1507 };
d342894c 1508
1509 if (nla_put_u32(skb, IFLA_VXLAN_ID, vxlan->vni))
1510 goto nla_put_failure;
1511
7c41c42c 1512 if (vxlan->gaddr && nla_put_be32(skb, IFLA_VXLAN_GROUP, vxlan->gaddr))
d342894c 1513 goto nla_put_failure;
1514
1515 if (vxlan->link && nla_put_u32(skb, IFLA_VXLAN_LINK, vxlan->link))
1516 goto nla_put_failure;
1517
7c41c42c 1518 if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
d342894c 1519 goto nla_put_failure;
1520
1521 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1522 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
e4f67add
DS		 1523 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
1524 !!(vxlan->flags & VXLAN_F_LEARN)) ||
1525 nla_put_u8(skb, IFLA_VXLAN_PROXY,
1526 !!(vxlan->flags & VXLAN_F_PROXY)) ||
1527 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
1528 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
1529 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
1530 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
1531 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
d342894c 1532 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1533 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax))
1534 goto nla_put_failure;
1535
05f47d69 1536 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
1537 goto nla_put_failure;
1538
d342894c 1539 return 0;
1540
1541nla_put_failure:
1542 return -EMSGSIZE;
1543}
1544
1545static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1546 .kind = "vxlan",
1547 .maxtype = IFLA_VXLAN_MAX,
1548 .policy = vxlan_policy,
1549 .priv_size = sizeof(struct vxlan_dev),
1550 .setup = vxlan_setup,
1551 .validate = vxlan_validate,
1552 .newlink = vxlan_newlink,
1553 .dellink = vxlan_dellink,
1554 .get_size = vxlan_get_size,
1555 .fill_info = vxlan_fill_info,
1556};
1557
1558static __net_init int vxlan_init_net(struct net *net)
1559{
1560 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1561 struct sock *sk;
1562 struct sockaddr_in vxlan_addr = {
1563 .sin_family = AF_INET,
1564 .sin_addr.s_addr = htonl(INADDR_ANY),
1565 };
1566 int rc;
1567 unsigned h;
1568
1569 /* Create UDP socket for encapsulation receive. */
1570 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vn->sock);
1571 if (rc < 0) {
1572 pr_debug("UDP socket create failed\n");
1573 return rc;
1574 }
bfe1b9b1 1575 /* Put in proper namespace */
1576 sk = vn->sock->sk;
1577 sk_change_net(sk, net);
d342894c 1578
1579 vxlan_addr.sin_port = htons(vxlan_port);
1580
1581 rc = kernel_bind(vn->sock, (struct sockaddr *) &vxlan_addr,
1582 sizeof(vxlan_addr));
1583 if (rc < 0) {
1584 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1585 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
bfe1b9b1 1586 sk_release_kernel(sk);
d342894c 1587 vn->sock = NULL;
1588 return rc;
1589 }
1590
1591 /* Disable multicast loopback */
d342894c 1592 inet_sk(sk)->mc_loop = 0;
1593
1594 /* Mark socket as an encapsulation socket. */
1595 udp_sk(sk)->encap_type = 1;
1596 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1597 udp_encap_enable();
1598
1599 for (h = 0; h < VNI_HASH_SIZE; ++h)
1600 INIT_HLIST_HEAD(&vn->vni_list[h]);
1601
1602 return 0;
1603}
1604
1605static __net_exit void vxlan_exit_net(struct net *net)
1606{
1607 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
9cb6cb7e
ZM		 1608 struct vxlan_dev *vxlan;
1609 unsigned h;
1610
1611 rtnl_lock();
1612 for (h = 0; h < VNI_HASH_SIZE; ++h)
1613 hlist_for_each_entry(vxlan, &vn->vni_list[h], hlist)
1614 dev_close(vxlan->dev);
1615 rtnl_unlock();
d342894c 1616
1617 if (vn->sock) {
bfe1b9b1 1618 sk_release_kernel(vn->sock->sk);
d342894c 1619 vn->sock = NULL;
1620 }
1621}
1622
1623static struct pernet_operations vxlan_net_ops = {
1624 .init = vxlan_init_net,
1625 .exit = vxlan_exit_net,
1626 .id = &vxlan_net_id,
1627 .size = sizeof(struct vxlan_net),
1628};
1629
1630static int __init vxlan_init_module(void)
1631{
1632 int rc;
1633
1634 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1635
1636 rc = register_pernet_device(&vxlan_net_ops);
1637 if (rc)
1638 goto out1;
1639
1640 rc = rtnl_link_register(&vxlan_link_ops);
1641 if (rc)
1642 goto out2;
1643
1644 return 0;
1645
1646out2:
1647 unregister_pernet_device(&vxlan_net_ops);
1648out1:
1649 return rc;
1650}
1651module_init(vxlan_init_module);
1652
1653static void __exit vxlan_cleanup_module(void)
1654{
1655 rtnl_link_unregister(&vxlan_link_ops);
1656 unregister_pernet_device(&vxlan_net_ops);
6681712d 1657 rcu_barrier();
d342894c 1658}
1659module_exit(vxlan_cleanup_module);
1660
1661MODULE_LICENSE("GPL");
1662MODULE_VERSION(VXLAN_VERSION);
1663MODULE_AUTHOR("Stephen Hemminger <shemminger@vyatta.com>");
1664MODULE_ALIAS_RTNL_LINK("vxlan");
Linux 6.x block layer and io_uring tree(s)