projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
tunneling: Add generic Tunnel segmentation.
[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>
8dc98eb2 36#include <net/ipip.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
84/* Forwarding table entry */
85struct vxlan_fdb {
86 struct hlist_node hlist; /* linked list of entries */
87 struct rcu_head rcu;
88 unsigned long updated; /* jiffies */
89 unsigned long used;
90 __be32 remote_ip;
91 u16 state; /* see ndm_state */
92 u8 eth_addr[ETH_ALEN];
93};
94
95/* Per-cpu network traffic stats */
96struct vxlan_stats {
97 u64 rx_packets;
98 u64 rx_bytes;
99 u64 tx_packets;
100 u64 tx_bytes;
101 struct u64_stats_sync syncp;
102};
103
104/* Pseudo network device */
105struct vxlan_dev {
106 struct hlist_node hlist;
107 struct net_device *dev;
108 struct vxlan_stats __percpu *stats;
109 __u32 vni; /* virtual network id */
110 __be32 gaddr; /* multicast group */
111 __be32 saddr; /* source address */
112 unsigned int link; /* link to multicast over */
05f47d69 113 __u16 port_min; /* source port range */
114 __u16 port_max;
d342894c 115 __u8 tos; /* TOS override */
116 __u8 ttl;
e4f67add 117 u32 flags; /* VXLAN_F_* below */
d342894c 118
119 unsigned long age_interval;
120 struct timer_list age_timer;
121 spinlock_t hash_lock;
122 unsigned int addrcnt;
123 unsigned int addrmax;
d342894c 124
125 struct hlist_head fdb_head[FDB_HASH_SIZE];
126};
127
e4f67add
DS		 128#define VXLAN_F_LEARN 0x01
129#define VXLAN_F_PROXY 0x02
130#define VXLAN_F_RSC 0x04
131#define VXLAN_F_L2MISS 0x08
132#define VXLAN_F_L3MISS 0x10
133
d342894c 134/* salt for hash table */
135static u32 vxlan_salt __read_mostly;
136
137static inline struct hlist_head *vni_head(struct net *net, u32 id)
138{
139 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
140
141 return &vn->vni_list[hash_32(id, VNI_HASH_BITS)];
142}
143
144/* Look up VNI in a per net namespace table */
145static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id)
146{
147 struct vxlan_dev *vxlan;
d342894c 148
b67bfe0d 149 hlist_for_each_entry_rcu(vxlan, vni_head(net, id), hlist) {
d342894c 150 if (vxlan->vni == id)
151 return vxlan;
152 }
153
154 return NULL;
155}
156
157/* Fill in neighbour message in skbuff. */
158static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
159 const struct vxlan_fdb *fdb,
160 u32 portid, u32 seq, int type, unsigned int flags)
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;
179 send_ip = fdb->remote_ip != 0;
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
e4f67add 191 if (send_ip && nla_put_be32(skb, NDA_DST, fdb->remote_ip))
d342894c 192 goto nla_put_failure;
193
194 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
195 ci.ndm_confirmed = 0;
196 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
197 ci.ndm_refcnt = 0;
198
199 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
200 goto nla_put_failure;
201
202 return nlmsg_end(skb, nlh);
203
204nla_put_failure:
205 nlmsg_cancel(skb, nlh);
206 return -EMSGSIZE;
207}
208
209static inline size_t vxlan_nlmsg_size(void)
210{
211 return NLMSG_ALIGN(sizeof(struct ndmsg))
212 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
213 + nla_total_size(sizeof(__be32)) /* NDA_DST */
214 + nla_total_size(sizeof(struct nda_cacheinfo));
215}
216
217static void vxlan_fdb_notify(struct vxlan_dev *vxlan,
218 const struct vxlan_fdb *fdb, int type)
219{
220 struct net *net = dev_net(vxlan->dev);
221 struct sk_buff *skb;
222 int err = -ENOBUFS;
223
224 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
225 if (skb == NULL)
226 goto errout;
227
228 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0);
229 if (err < 0) {
230 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
231 WARN_ON(err == -EMSGSIZE);
232 kfree_skb(skb);
233 goto errout;
234 }
235
236 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
237 return;
238errout:
239 if (err < 0)
240 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
241}
242
e4f67add
DS		 243static void vxlan_ip_miss(struct net_device *dev, __be32 ipa)
244{
245 struct vxlan_dev *vxlan = netdev_priv(dev);
246 struct vxlan_fdb f;
247
248 memset(&f, 0, sizeof f);
249 f.state = NUD_STALE;
250 f.remote_ip = ipa; /* goes to NDA_DST */
251
252 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
253}
254
255static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
256{
257 struct vxlan_fdb f;
258
259 memset(&f, 0, sizeof f);
260 f.state = NUD_STALE;
261 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
262
263 vxlan_fdb_notify(vxlan, &f, RTM_GETNEIGH);
264}
265
d342894c 266/* Hash Ethernet address */
267static u32 eth_hash(const unsigned char *addr)
268{
269 u64 value = get_unaligned((u64 *)addr);
270
271 /* only want 6 bytes */
272#ifdef __BIG_ENDIAN
d342894c 273 value >>= 16;
321fb991 274#else
275 value <<= 16;
d342894c 276#endif
277 return hash_64(value, FDB_HASH_BITS);
278}
279
280/* Hash chain to use given mac address */
281static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
282 const u8 *mac)
283{
284 return &vxlan->fdb_head[eth_hash(mac)];
285}
286
287/* Look up Ethernet address in forwarding table */
288static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
289 const u8 *mac)
290
291{
292 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
293 struct vxlan_fdb *f;
d342894c 294
b67bfe0d 295 hlist_for_each_entry_rcu(f, head, hlist) {
d342894c 296 if (compare_ether_addr(mac, f->eth_addr) == 0)
297 return f;
298 }
299
300 return NULL;
301}
302
303/* Add new entry to forwarding table -- assumes lock held */
304static int vxlan_fdb_create(struct vxlan_dev *vxlan,
305 const u8 *mac, __be32 ip,
306 __u16 state, __u16 flags)
307{
308 struct vxlan_fdb *f;
309 int notify = 0;
310
311 f = vxlan_find_mac(vxlan, mac);
312 if (f) {
313 if (flags & NLM_F_EXCL) {
314 netdev_dbg(vxlan->dev,
315 "lost race to create %pM\n", mac);
316 return -EEXIST;
317 }
318 if (f->state != state) {
319 f->state = state;
320 f->updated = jiffies;
321 notify = 1;
322 }
323 } else {
324 if (!(flags & NLM_F_CREATE))
325 return -ENOENT;
326
327 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
328 return -ENOSPC;
329
330 netdev_dbg(vxlan->dev, "add %pM -> %pI4\n", mac, &ip);
331 f = kmalloc(sizeof(*f), GFP_ATOMIC);
332 if (!f)
333 return -ENOMEM;
334
335 notify = 1;
336 f->remote_ip = ip;
337 f->state = state;
338 f->updated = f->used = jiffies;
339 memcpy(f->eth_addr, mac, ETH_ALEN);
340
341 ++vxlan->addrcnt;
342 hlist_add_head_rcu(&f->hlist,
343 vxlan_fdb_head(vxlan, mac));
344 }
345
346 if (notify)
347 vxlan_fdb_notify(vxlan, f, RTM_NEWNEIGH);
348
349 return 0;
350}
351
352static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
353{
354 netdev_dbg(vxlan->dev,
355 "delete %pM\n", f->eth_addr);
356
357 --vxlan->addrcnt;
358 vxlan_fdb_notify(vxlan, f, RTM_DELNEIGH);
359
360 hlist_del_rcu(&f->hlist);
361 kfree_rcu(f, rcu);
362}
363
364/* Add static entry (via netlink) */
365static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
366 struct net_device *dev,
367 const unsigned char *addr, u16 flags)
368{
369 struct vxlan_dev *vxlan = netdev_priv(dev);
370 __be32 ip;
371 int err;
372
373 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
374 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
375 ndm->ndm_state);
376 return -EINVAL;
377 }
378
379 if (tb[NDA_DST] == NULL)
380 return -EINVAL;
381
382 if (nla_len(tb[NDA_DST]) != sizeof(__be32))
383 return -EAFNOSUPPORT;
384
385 ip = nla_get_be32(tb[NDA_DST]);
386
387 spin_lock_bh(&vxlan->hash_lock);
388 err = vxlan_fdb_create(vxlan, addr, ip, ndm->ndm_state, flags);
389 spin_unlock_bh(&vxlan->hash_lock);
390
391 return err;
392}
393
394/* Delete entry (via netlink) */
1690be63
VY		 395static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
396 struct net_device *dev,
d342894c 397 const unsigned char *addr)
398{
399 struct vxlan_dev *vxlan = netdev_priv(dev);
400 struct vxlan_fdb *f;
401 int err = -ENOENT;
402
403 spin_lock_bh(&vxlan->hash_lock);
404 f = vxlan_find_mac(vxlan, addr);
405 if (f) {
406 vxlan_fdb_destroy(vxlan, f);
407 err = 0;
408 }
409 spin_unlock_bh(&vxlan->hash_lock);
410
411 return err;
412}
413
414/* Dump forwarding table */
415static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
416 struct net_device *dev, int idx)
417{
418 struct vxlan_dev *vxlan = netdev_priv(dev);
419 unsigned int h;
420
421 for (h = 0; h < FDB_HASH_SIZE; ++h) {
422 struct vxlan_fdb *f;
d342894c 423 int err;
424
b67bfe0d 425 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
d342894c 426 if (idx < cb->args[0])
427 goto skip;
428
429 err = vxlan_fdb_info(skb, vxlan, f,
430 NETLINK_CB(cb->skb).portid,
431 cb->nlh->nlmsg_seq,
432 RTM_NEWNEIGH,
433 NLM_F_MULTI);
434 if (err < 0)
435 break;
436skip:
437 ++idx;
438 }
439 }
440
441 return idx;
442}
443
444/* Watch incoming packets to learn mapping between Ethernet address
445 * and Tunnel endpoint.
446 */
447static void vxlan_snoop(struct net_device *dev,
448 __be32 src_ip, const u8 *src_mac)
449{
450 struct vxlan_dev *vxlan = netdev_priv(dev);
451 struct vxlan_fdb *f;
452 int err;
453
454 f = vxlan_find_mac(vxlan, src_mac);
455 if (likely(f)) {
456 f->used = jiffies;
457 if (likely(f->remote_ip == src_ip))
458 return;
459
460 if (net_ratelimit())
461 netdev_info(dev,
462 "%pM migrated from %pI4 to %pI4\n",
463 src_mac, &f->remote_ip, &src_ip);
464
465 f->remote_ip = src_ip;
466 f->updated = jiffies;
467 } else {
468 /* learned new entry */
469 spin_lock(&vxlan->hash_lock);
470 err = vxlan_fdb_create(vxlan, src_mac, src_ip,
471 NUD_REACHABLE,
472 NLM_F_EXCL|NLM_F_CREATE);
473 spin_unlock(&vxlan->hash_lock);
474 }
475}
476
477
478/* See if multicast group is already in use by other ID */
479static bool vxlan_group_used(struct vxlan_net *vn,
480 const struct vxlan_dev *this)
481{
482 const struct vxlan_dev *vxlan;
d342894c 483 unsigned h;
484
485 for (h = 0; h < VNI_HASH_SIZE; ++h)
b67bfe0d 486 hlist_for_each_entry(vxlan, &vn->vni_list[h], hlist) {
d342894c 487 if (vxlan == this)
488 continue;
489
490 if (!netif_running(vxlan->dev))
491 continue;
492
493 if (vxlan->gaddr == this->gaddr)
494 return true;
495 }
496
497 return false;
498}
499
500/* kernel equivalent to IP_ADD_MEMBERSHIP */
501static int vxlan_join_group(struct net_device *dev)
502{
503 struct vxlan_dev *vxlan = netdev_priv(dev);
504 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
505 struct sock *sk = vn->sock->sk;
506 struct ip_mreqn mreq = {
af9b078e
YB		 507 .imr_multiaddr.s_addr = vxlan->gaddr,
508 .imr_ifindex = vxlan->link,
d342894c 509 };
510 int err;
511
512 /* Already a member of group */
513 if (vxlan_group_used(vn, vxlan))
514 return 0;
515
516 /* Need to drop RTNL to call multicast join */
517 rtnl_unlock();
518 lock_sock(sk);
519 err = ip_mc_join_group(sk, &mreq);
520 release_sock(sk);
521 rtnl_lock();
522
523 return err;
524}
525
526
527/* kernel equivalent to IP_DROP_MEMBERSHIP */
528static int vxlan_leave_group(struct net_device *dev)
529{
530 struct vxlan_dev *vxlan = netdev_priv(dev);
531 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
532 int err = 0;
533 struct sock *sk = vn->sock->sk;
534 struct ip_mreqn mreq = {
af9b078e
YB		 535 .imr_multiaddr.s_addr = vxlan->gaddr,
536 .imr_ifindex = vxlan->link,
d342894c 537 };
538
539 /* Only leave group when last vxlan is done. */
540 if (vxlan_group_used(vn, vxlan))
541 return 0;
542
543 /* Need to drop RTNL to call multicast leave */
544 rtnl_unlock();
545 lock_sock(sk);
546 err = ip_mc_leave_group(sk, &mreq);
547 release_sock(sk);
548 rtnl_lock();
549
550 return err;
551}
552
553/* Callback from net/ipv4/udp.c to receive packets */
554static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
555{
556 struct iphdr *oip;
557 struct vxlanhdr *vxh;
558 struct vxlan_dev *vxlan;
559 struct vxlan_stats *stats;
560 __u32 vni;
561 int err;
562
563 /* pop off outer UDP header */
564 __skb_pull(skb, sizeof(struct udphdr));
565
566 /* Need Vxlan and inner Ethernet header to be present */
567 if (!pskb_may_pull(skb, sizeof(struct vxlanhdr)))
568 goto error;
569
570 /* Drop packets with reserved bits set */
571 vxh = (struct vxlanhdr *) skb->data;
572 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
573 (vxh->vx_vni & htonl(0xff))) {
574 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
575 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
576 goto error;
577 }
578
579 __skb_pull(skb, sizeof(struct vxlanhdr));
d342894c 580
581 /* Is this VNI defined? */
582 vni = ntohl(vxh->vx_vni) >> 8;
583 vxlan = vxlan_find_vni(sock_net(sk), vni);
584 if (!vxlan) {
585 netdev_dbg(skb->dev, "unknown vni %d\n", vni);
586 goto drop;
587 }
588
589 if (!pskb_may_pull(skb, ETH_HLEN)) {
590 vxlan->dev->stats.rx_length_errors++;
591 vxlan->dev->stats.rx_errors++;
592 goto drop;
593 }
594
e4f67add
DS		 595 skb_reset_mac_header(skb);
596
d342894c 597 /* Re-examine inner Ethernet packet */
598 oip = ip_hdr(skb);
599 skb->protocol = eth_type_trans(skb, vxlan->dev);
d342894c 600
601 /* Ignore packet loops (and multicast echo) */
602 if (compare_ether_addr(eth_hdr(skb)->h_source,
603 vxlan->dev->dev_addr) == 0)
604 goto drop;
605
e4f67add 606 if (vxlan->flags & VXLAN_F_LEARN)
d342894c 607 vxlan_snoop(skb->dev, oip->saddr, eth_hdr(skb)->h_source);
608
609 __skb_tunnel_rx(skb, vxlan->dev);
610 skb_reset_network_header(skb);
0afb1666
JG		 611
612 /* If the NIC driver gave us an encapsulated packet with
613 * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
614 * leave the CHECKSUM_UNNECESSARY, the device checksummed it
615 * for us. Otherwise force the upper layers to verify it.
616 */
617 if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
618 !(vxlan->dev->features & NETIF_F_RXCSUM))
619 skb->ip_summed = CHECKSUM_NONE;
620
621 skb->encapsulation = 0;
d342894c 622
623 err = IP_ECN_decapsulate(oip, skb);
624 if (unlikely(err)) {
625 if (log_ecn_error)
626 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
627 &oip->saddr, oip->tos);
628 if (err > 1) {
629 ++vxlan->dev->stats.rx_frame_errors;
630 ++vxlan->dev->stats.rx_errors;
631 goto drop;
632 }
633 }
634
635 stats = this_cpu_ptr(vxlan->stats);
636 u64_stats_update_begin(&stats->syncp);
637 stats->rx_packets++;
638 stats->rx_bytes += skb->len;
639 u64_stats_update_end(&stats->syncp);
640
641 netif_rx(skb);
642
643 return 0;
644error:
645 /* Put UDP header back */
646 __skb_push(skb, sizeof(struct udphdr));
647
648 return 1;
649drop:
650 /* Consume bad packet */
651 kfree_skb(skb);
652 return 0;
653}
654
e4f67add
DS		 655static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
656{
657 struct vxlan_dev *vxlan = netdev_priv(dev);
658 struct arphdr *parp;
659 u8 *arpptr, *sha;
660 __be32 sip, tip;
661 struct neighbour *n;
662
663 if (dev->flags & IFF_NOARP)
664 goto out;
665
666 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
667 dev->stats.tx_dropped++;
668 goto out;
669 }
670 parp = arp_hdr(skb);
671
672 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
673 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
674 parp->ar_pro != htons(ETH_P_IP) ||
675 parp->ar_op != htons(ARPOP_REQUEST) ||
676 parp->ar_hln != dev->addr_len ||
677 parp->ar_pln != 4)
678 goto out;
679 arpptr = (u8 *)parp + sizeof(struct arphdr);
680 sha = arpptr;
681 arpptr += dev->addr_len; /* sha */
682 memcpy(&sip, arpptr, sizeof(sip));
683 arpptr += sizeof(sip);
684 arpptr += dev->addr_len; /* tha */
685 memcpy(&tip, arpptr, sizeof(tip));
686
687 if (ipv4_is_loopback(tip) ||
688 ipv4_is_multicast(tip))
689 goto out;
690
691 n = neigh_lookup(&arp_tbl, &tip, dev);
692
693 if (n) {
694 struct vxlan_dev *vxlan = netdev_priv(dev);
695 struct vxlan_fdb *f;
696 struct sk_buff *reply;
697
698 if (!(n->nud_state & NUD_CONNECTED)) {
699 neigh_release(n);
700 goto out;
701 }
702
703 f = vxlan_find_mac(vxlan, n->ha);
704 if (f && f->remote_ip == 0) {
705 /* bridge-local neighbor */
706 neigh_release(n);
707 goto out;
708 }
709
710 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
711 n->ha, sha);
712
713 neigh_release(n);
714
715 skb_reset_mac_header(reply);
716 __skb_pull(reply, skb_network_offset(reply));
717 reply->ip_summed = CHECKSUM_UNNECESSARY;
718 reply->pkt_type = PACKET_HOST;
719
720 if (netif_rx_ni(reply) == NET_RX_DROP)
721 dev->stats.rx_dropped++;
722 } else if (vxlan->flags & VXLAN_F_L3MISS)
723 vxlan_ip_miss(dev, tip);
724out:
725 consume_skb(skb);
726 return NETDEV_TX_OK;
727}
728
729static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
730{
731 struct vxlan_dev *vxlan = netdev_priv(dev);
732 struct neighbour *n;
733 struct iphdr *pip;
734
735 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
736 return false;
737
738 n = NULL;
739 switch (ntohs(eth_hdr(skb)->h_proto)) {
740 case ETH_P_IP:
741 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
742 return false;
743 pip = ip_hdr(skb);
744 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
745 break;
746 default:
747 return false;
748 }
749
750 if (n) {
751 bool diff;
752
753 diff = compare_ether_addr(eth_hdr(skb)->h_dest, n->ha) != 0;
754 if (diff) {
755 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
756 dev->addr_len);
757 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
758 }
759 neigh_release(n);
760 return diff;
761 } else if (vxlan->flags & VXLAN_F_L3MISS)
762 vxlan_ip_miss(dev, pip->daddr);
763 return false;
764}
765
d342894c 766/* Extract dsfield from inner protocol */
767static inline u8 vxlan_get_dsfield(const struct iphdr *iph,
768 const struct sk_buff *skb)
769{
770 if (skb->protocol == htons(ETH_P_IP))
771 return iph->tos;
772 else if (skb->protocol == htons(ETH_P_IPV6))
773 return ipv6_get_dsfield((const struct ipv6hdr *)iph);
774 else
775 return 0;
776}
777
778/* Propogate ECN bits out */
779static inline u8 vxlan_ecn_encap(u8 tos,
780 const struct iphdr *iph,
781 const struct sk_buff *skb)
782{
783 u8 inner = vxlan_get_dsfield(iph, skb);
784
785 return INET_ECN_encapsulate(tos, inner);
786}
787
1cad8715 788static void vxlan_sock_free(struct sk_buff *skb)
789{
790 sock_put(skb->sk);
791}
792
793/* On transmit, associate with the tunnel socket */
794static void vxlan_set_owner(struct net_device *dev, struct sk_buff *skb)
795{
796 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
797 struct sock *sk = vn->sock->sk;
798
799 skb_orphan(skb);
800 sock_hold(sk);
801 skb->sk = sk;
802 skb->destructor = vxlan_sock_free;
803}
804
05f47d69 805/* Compute source port for outgoing packet
806 * first choice to use L4 flow hash since it will spread
807 * better and maybe available from hardware
808 * secondary choice is to use jhash on the Ethernet header
809 */
810static u16 vxlan_src_port(const struct vxlan_dev *vxlan, struct sk_buff *skb)
811{
812 unsigned int range = (vxlan->port_max - vxlan->port_min) + 1;
813 u32 hash;
814
815 hash = skb_get_rxhash(skb);
816 if (!hash)
817 hash = jhash(skb->data, 2 * ETH_ALEN,
818 (__force u32) skb->protocol);
819
820 return (((u64) hash * range) >> 32) + vxlan->port_min;
821}
822
d342894c 823/* Transmit local packets over Vxlan
824 *
825 * Outer IP header inherits ECN and DF from inner header.
826 * Outer UDP destination is the VXLAN assigned port.
05f47d69 827 * source port is based on hash of flow
d342894c 828 */
829static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
830{
831 struct vxlan_dev *vxlan = netdev_priv(dev);
832 struct rtable *rt;
d342894c 833 const struct iphdr *old_iph;
e4f67add 834 struct ethhdr *eth;
d342894c 835 struct iphdr *iph;
836 struct vxlanhdr *vxh;
837 struct udphdr *uh;
838 struct flowi4 fl4;
d342894c 839 unsigned int pkt_len = skb->len;
d342894c 840 __be32 dst;
05f47d69 841 __u16 src_port;
d342894c 842 __be16 df = 0;
843 __u8 tos, ttl;
844 int err;
e4f67add
DS		 845 bool did_rsc = false;
846 const struct vxlan_fdb *f;
847
848 skb_reset_mac_header(skb);
849 eth = eth_hdr(skb);
850
851 if ((vxlan->flags & VXLAN_F_PROXY) && ntohs(eth->h_proto) == ETH_P_ARP)
852 return arp_reduce(dev, skb);
853 else if ((vxlan->flags&VXLAN_F_RSC) && ntohs(eth->h_proto) == ETH_P_IP)
854 did_rsc = route_shortcircuit(dev, skb);
d342894c 855
e4f67add
DS		 856 f = vxlan_find_mac(vxlan, eth->h_dest);
857 if (f == NULL) {
858 did_rsc = false;
859 dst = vxlan->gaddr;
860 if (!dst && (vxlan->flags & VXLAN_F_L2MISS) &&
861 !is_multicast_ether_addr(eth->h_dest))
862 vxlan_fdb_miss(vxlan, eth->h_dest);
863 } else
864 dst = f->remote_ip;
865
866 if (!dst) {
867 if (did_rsc) {
868 __skb_pull(skb, skb_network_offset(skb));
869 skb->ip_summed = CHECKSUM_NONE;
870 skb->pkt_type = PACKET_HOST;
871
872 /* short-circuited back to local bridge */
873 if (netif_rx(skb) == NET_RX_SUCCESS) {
874 struct vxlan_stats *stats =
875 this_cpu_ptr(vxlan->stats);
876
877 u64_stats_update_begin(&stats->syncp);
878 stats->tx_packets++;
879 stats->tx_bytes += pkt_len;
880 u64_stats_update_end(&stats->syncp);
881 } else {
882 dev->stats.tx_errors++;
883 dev->stats.tx_aborted_errors++;
884 }
885 return NETDEV_TX_OK;
886 }
ef59febe 887 goto drop;
e4f67add 888 }
ef59febe 889
d6727fe3
JG		 890 if (!skb->encapsulation) {
891 skb_reset_inner_headers(skb);
892 skb->encapsulation = 1;
893 }
894
d342894c 895 /* Need space for new headers (invalidates iph ptr) */
896 if (skb_cow_head(skb, VXLAN_HEADROOM))
897 goto drop;
898
d342894c 899 old_iph = ip_hdr(skb);
900
d342894c 901 ttl = vxlan->ttl;
902 if (!ttl && IN_MULTICAST(ntohl(dst)))
903 ttl = 1;
904
905 tos = vxlan->tos;
906 if (tos == 1)
907 tos = vxlan_get_dsfield(old_iph, skb);
908
05f47d69 909 src_port = vxlan_src_port(vxlan, skb);
d342894c 910
ca78f181 911 memset(&fl4, 0, sizeof(fl4));
912 fl4.flowi4_oif = vxlan->link;
913 fl4.flowi4_tos = RT_TOS(tos);
914 fl4.daddr = dst;
915 fl4.saddr = vxlan->saddr;
916
917 rt = ip_route_output_key(dev_net(dev), &fl4);
d342894c 918 if (IS_ERR(rt)) {
919 netdev_dbg(dev, "no route to %pI4\n", &dst);
920 dev->stats.tx_carrier_errors++;
921 goto tx_error;
922 }
923
924 if (rt->dst.dev == dev) {
925 netdev_dbg(dev, "circular route to %pI4\n", &dst);
926 ip_rt_put(rt);
927 dev->stats.collisions++;
928 goto tx_error;
929 }
930
931 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
932 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
933 IPSKB_REROUTED);
934 skb_dst_drop(skb);
935 skb_dst_set(skb, &rt->dst);
936
937 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
938 vxh->vx_flags = htonl(VXLAN_FLAGS);
939 vxh->vx_vni = htonl(vxlan->vni << 8);
940
941 __skb_push(skb, sizeof(*uh));
942 skb_reset_transport_header(skb);
943 uh = udp_hdr(skb);
944
945 uh->dest = htons(vxlan_port);
05f47d69 946 uh->source = htons(src_port);
d342894c 947
948 uh->len = htons(skb->len);
949 uh->check = 0;
950
951 __skb_push(skb, sizeof(*iph));
952 skb_reset_network_header(skb);
953 iph = ip_hdr(skb);
954 iph->version = 4;
955 iph->ihl = sizeof(struct iphdr) >> 2;
956 iph->frag_off = df;
957 iph->protocol = IPPROTO_UDP;
958 iph->tos = vxlan_ecn_encap(tos, old_iph, skb);
ca78f181 959 iph->daddr = dst;
d342894c 960 iph->saddr = fl4.saddr;
961 iph->ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
8dc98eb2 962 tunnel_ip_select_ident(skb, old_iph, &rt->dst);
d342894c 963
1cad8715 964 vxlan_set_owner(dev, skb);
965
aa0010f8 966 /* See iptunnel_xmit() */
d6727fe3
JG		 967 if (skb->ip_summed != CHECKSUM_PARTIAL)
968 skb->ip_summed = CHECKSUM_NONE;
d342894c 969
970 err = ip_local_out(skb);
971 if (likely(net_xmit_eval(err) == 0)) {
972 struct vxlan_stats *stats = this_cpu_ptr(vxlan->stats);
973
974 u64_stats_update_begin(&stats->syncp);
975 stats->tx_packets++;
976 stats->tx_bytes += pkt_len;
977 u64_stats_update_end(&stats->syncp);
978 } else {
979 dev->stats.tx_errors++;
980 dev->stats.tx_aborted_errors++;
981 }
982 return NETDEV_TX_OK;
983
984drop:
985 dev->stats.tx_dropped++;
986 goto tx_free;
987
988tx_error:
989 dev->stats.tx_errors++;
990tx_free:
991 dev_kfree_skb(skb);
992 return NETDEV_TX_OK;
993}
994
995/* Walk the forwarding table and purge stale entries */
996static void vxlan_cleanup(unsigned long arg)
997{
998 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
999 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1000 unsigned int h;
1001
1002 if (!netif_running(vxlan->dev))
1003 return;
1004
1005 spin_lock_bh(&vxlan->hash_lock);
1006 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1007 struct hlist_node *p, *n;
1008 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1009 struct vxlan_fdb *f
1010 = container_of(p, struct vxlan_fdb, hlist);
1011 unsigned long timeout;
1012
3c172868 1013 if (f->state & NUD_PERMANENT)
d342894c 1014 continue;
1015
1016 timeout = f->used + vxlan->age_interval * HZ;
1017 if (time_before_eq(timeout, jiffies)) {
1018 netdev_dbg(vxlan->dev,
1019 "garbage collect %pM\n",
1020 f->eth_addr);
1021 f->state = NUD_STALE;
1022 vxlan_fdb_destroy(vxlan, f);
1023 } else if (time_before(timeout, next_timer))
1024 next_timer = timeout;
1025 }
1026 }
1027 spin_unlock_bh(&vxlan->hash_lock);
1028
1029 mod_timer(&vxlan->age_timer, next_timer);
1030}
1031
1032/* Setup stats when device is created */
1033static int vxlan_init(struct net_device *dev)
1034{
1035 struct vxlan_dev *vxlan = netdev_priv(dev);
1036
1037 vxlan->stats = alloc_percpu(struct vxlan_stats);
1038 if (!vxlan->stats)
1039 return -ENOMEM;
1040
1041 return 0;
1042}
1043
1044/* Start ageing timer and join group when device is brought up */
1045static int vxlan_open(struct net_device *dev)
1046{
1047 struct vxlan_dev *vxlan = netdev_priv(dev);
1048 int err;
1049
1050 if (vxlan->gaddr) {
1051 err = vxlan_join_group(dev);
1052 if (err)
1053 return err;
1054 }
1055
1056 if (vxlan->age_interval)
1057 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
1058
1059 return 0;
1060}
1061
1062/* Purge the forwarding table */
1063static void vxlan_flush(struct vxlan_dev *vxlan)
1064{
1065 unsigned h;
1066
1067 spin_lock_bh(&vxlan->hash_lock);
1068 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1069 struct hlist_node *p, *n;
1070 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1071 struct vxlan_fdb *f
1072 = container_of(p, struct vxlan_fdb, hlist);
1073 vxlan_fdb_destroy(vxlan, f);
1074 }
1075 }
1076 spin_unlock_bh(&vxlan->hash_lock);
1077}
1078
1079/* Cleanup timer and forwarding table on shutdown */
1080static int vxlan_stop(struct net_device *dev)
1081{
1082 struct vxlan_dev *vxlan = netdev_priv(dev);
1083
1084 if (vxlan->gaddr)
1085 vxlan_leave_group(dev);
1086
1087 del_timer_sync(&vxlan->age_timer);
1088
1089 vxlan_flush(vxlan);
1090
1091 return 0;
1092}
1093
1094/* Merge per-cpu statistics */
1095static struct rtnl_link_stats64 *vxlan_stats64(struct net_device *dev,
1096 struct rtnl_link_stats64 *stats)
1097{
1098 struct vxlan_dev *vxlan = netdev_priv(dev);
1099 struct vxlan_stats tmp, sum = { 0 };
1100 unsigned int cpu;
1101
1102 for_each_possible_cpu(cpu) {
1103 unsigned int start;
1104 const struct vxlan_stats *stats
1105 = per_cpu_ptr(vxlan->stats, cpu);
1106
1107 do {
1108 start = u64_stats_fetch_begin_bh(&stats->syncp);
1109 memcpy(&tmp, stats, sizeof(tmp));
1110 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1111
1112 sum.tx_bytes += tmp.tx_bytes;
1113 sum.tx_packets += tmp.tx_packets;
1114 sum.rx_bytes += tmp.rx_bytes;
1115 sum.rx_packets += tmp.rx_packets;
1116 }
1117
1118 stats->tx_bytes = sum.tx_bytes;
1119 stats->tx_packets = sum.tx_packets;
1120 stats->rx_bytes = sum.rx_bytes;
1121 stats->rx_packets = sum.rx_packets;
1122
1123 stats->multicast = dev->stats.multicast;
1124 stats->rx_length_errors = dev->stats.rx_length_errors;
1125 stats->rx_frame_errors = dev->stats.rx_frame_errors;
1126 stats->rx_errors = dev->stats.rx_errors;
1127
1128 stats->tx_dropped = dev->stats.tx_dropped;
1129 stats->tx_carrier_errors = dev->stats.tx_carrier_errors;
1130 stats->tx_aborted_errors = dev->stats.tx_aborted_errors;
1131 stats->collisions = dev->stats.collisions;
1132 stats->tx_errors = dev->stats.tx_errors;
1133
1134 return stats;
1135}
1136
1137/* Stub, nothing needs to be done. */
1138static void vxlan_set_multicast_list(struct net_device *dev)
1139{
1140}
1141
1142static const struct net_device_ops vxlan_netdev_ops = {
1143 .ndo_init = vxlan_init,
1144 .ndo_open = vxlan_open,
1145 .ndo_stop = vxlan_stop,
1146 .ndo_start_xmit = vxlan_xmit,
1147 .ndo_get_stats64 = vxlan_stats64,
1148 .ndo_set_rx_mode = vxlan_set_multicast_list,
1149 .ndo_change_mtu = eth_change_mtu,
1150 .ndo_validate_addr = eth_validate_addr,
1151 .ndo_set_mac_address = eth_mac_addr,
1152 .ndo_fdb_add = vxlan_fdb_add,
1153 .ndo_fdb_del = vxlan_fdb_delete,
1154 .ndo_fdb_dump = vxlan_fdb_dump,
1155};
1156
1157/* Info for udev, that this is a virtual tunnel endpoint */
1158static struct device_type vxlan_type = {
1159 .name = "vxlan",
1160};
1161
1162static void vxlan_free(struct net_device *dev)
1163{
1164 struct vxlan_dev *vxlan = netdev_priv(dev);
1165
1166 free_percpu(vxlan->stats);
1167 free_netdev(dev);
1168}
1169
1170/* Initialize the device structure. */
1171static void vxlan_setup(struct net_device *dev)
1172{
1173 struct vxlan_dev *vxlan = netdev_priv(dev);
1174 unsigned h;
05f47d69 1175 int low, high;
d342894c 1176
1177 eth_hw_addr_random(dev);
1178 ether_setup(dev);
2840bf22 1179 dev->hard_header_len = ETH_HLEN + VXLAN_HEADROOM;
d342894c 1180
1181 dev->netdev_ops = &vxlan_netdev_ops;
1182 dev->destructor = vxlan_free;
1183 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
1184
1185 dev->tx_queue_len = 0;
1186 dev->features |= NETIF_F_LLTX;
1187 dev->features |= NETIF_F_NETNS_LOCAL;
d6727fe3 1188 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
0afb1666
JG		 1189 dev->features |= NETIF_F_RXCSUM;
1190
1191 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
d342894c 1192 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
6602d007 1193 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
d342894c 1194
1195 spin_lock_init(&vxlan->hash_lock);
1196
1197 init_timer_deferrable(&vxlan->age_timer);
1198 vxlan->age_timer.function = vxlan_cleanup;
1199 vxlan->age_timer.data = (unsigned long) vxlan;
1200
05f47d69 1201 inet_get_local_port_range(&low, &high);
1202 vxlan->port_min = low;
1203 vxlan->port_max = high;
1204
d342894c 1205 vxlan->dev = dev;
1206
1207 for (h = 0; h < FDB_HASH_SIZE; ++h)
1208 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
1209}
1210
1211static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
1212 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
1213 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1214 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
1215 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1216 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
1217 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
1218 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
1219 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
1220 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
05f47d69 1221 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
e4f67add
DS		 1222 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
1223 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
1224 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
1225 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
d342894c 1226};
1227
1228static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
1229{
1230 if (tb[IFLA_ADDRESS]) {
1231 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1232 pr_debug("invalid link address (not ethernet)\n");
1233 return -EINVAL;
1234 }
1235
1236 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1237 pr_debug("invalid all zero ethernet address\n");
1238 return -EADDRNOTAVAIL;
1239 }
1240 }
1241
1242 if (!data)
1243 return -EINVAL;
1244
1245 if (data[IFLA_VXLAN_ID]) {
1246 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
1247 if (id >= VXLAN_VID_MASK)
1248 return -ERANGE;
1249 }
1250
1251 if (data[IFLA_VXLAN_GROUP]) {
1252 __be32 gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1253 if (!IN_MULTICAST(ntohl(gaddr))) {
1254 pr_debug("group address is not IPv4 multicast\n");
1255 return -EADDRNOTAVAIL;
1256 }
1257 }
05f47d69 1258
1259 if (data[IFLA_VXLAN_PORT_RANGE]) {
1260 const struct ifla_vxlan_port_range *p
1261 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1262
1263 if (ntohs(p->high) < ntohs(p->low)) {
1264 pr_debug("port range %u .. %u not valid\n",
1265 ntohs(p->low), ntohs(p->high));
1266 return -EINVAL;
1267 }
1268 }
1269
d342894c 1270 return 0;
1271}
1272
1b13c97f
YB		 1273static void vxlan_get_drvinfo(struct net_device *netdev,
1274 struct ethtool_drvinfo *drvinfo)
1275{
1276 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
1277 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
1278}
1279
1280static const struct ethtool_ops vxlan_ethtool_ops = {
1281 .get_drvinfo = vxlan_get_drvinfo,
1282 .get_link = ethtool_op_get_link,
1283};
1284
d342894c 1285static int vxlan_newlink(struct net *net, struct net_device *dev,
1286 struct nlattr *tb[], struct nlattr *data[])
1287{
1288 struct vxlan_dev *vxlan = netdev_priv(dev);
1289 __u32 vni;
1290 int err;
1291
1292 if (!data[IFLA_VXLAN_ID])
1293 return -EINVAL;
1294
1295 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
1296 if (vxlan_find_vni(net, vni)) {
1297 pr_info("duplicate VNI %u\n", vni);
1298 return -EEXIST;
1299 }
1300 vxlan->vni = vni;
1301
1302 if (data[IFLA_VXLAN_GROUP])
1303 vxlan->gaddr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
1304
1305 if (data[IFLA_VXLAN_LOCAL])
1306 vxlan->saddr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
1307
34e02aa1 1308 if (data[IFLA_VXLAN_LINK] &&
1309 (vxlan->link = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
1310 struct net_device *lowerdev
1311 = __dev_get_by_index(net, vxlan->link);
1312
1313 if (!lowerdev) {
1314 pr_info("ifindex %d does not exist\n", vxlan->link);
1315 return -ENODEV;
1316 }
d342894c 1317
34e02aa1 1318 if (!tb[IFLA_MTU])
d342894c 1319 dev->mtu = lowerdev->mtu - VXLAN_HEADROOM;
1ba56fb4
AD		 1320
1321 /* update header length based on lower device */
1322 dev->hard_header_len = lowerdev->hard_header_len +
1323 VXLAN_HEADROOM;
d342894c 1324 }
1325
1326 if (data[IFLA_VXLAN_TOS])
1327 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
1328
afb97186
VB		 1329 if (data[IFLA_VXLAN_TTL])
1330 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
1331
d342894c 1332 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
e4f67add 1333 vxlan->flags |= VXLAN_F_LEARN;
d342894c 1334
1335 if (data[IFLA_VXLAN_AGEING])
1336 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
1337 else
1338 vxlan->age_interval = FDB_AGE_DEFAULT;
1339
e4f67add
DS		 1340 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
1341 vxlan->flags |= VXLAN_F_PROXY;
1342
1343 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
1344 vxlan->flags |= VXLAN_F_RSC;
1345
1346 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
1347 vxlan->flags |= VXLAN_F_L2MISS;
1348
1349 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
1350 vxlan->flags |= VXLAN_F_L3MISS;
1351
d342894c 1352 if (data[IFLA_VXLAN_LIMIT])
1353 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
1354
05f47d69 1355 if (data[IFLA_VXLAN_PORT_RANGE]) {
1356 const struct ifla_vxlan_port_range *p
1357 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
1358 vxlan->port_min = ntohs(p->low);
1359 vxlan->port_max = ntohs(p->high);
1360 }
1361
1b13c97f
YB		 1362 SET_ETHTOOL_OPS(dev, &vxlan_ethtool_ops);
1363
d342894c 1364 err = register_netdevice(dev);
1365 if (!err)
1366 hlist_add_head_rcu(&vxlan->hlist, vni_head(net, vxlan->vni));
1367
1368 return err;
1369}
1370
1371static void vxlan_dellink(struct net_device *dev, struct list_head *head)
1372{
1373 struct vxlan_dev *vxlan = netdev_priv(dev);
1374
1375 hlist_del_rcu(&vxlan->hlist);
1376
1377 unregister_netdevice_queue(dev, head);
1378}
1379
1380static size_t vxlan_get_size(const struct net_device *dev)
1381{
1382
1383 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
1384 nla_total_size(sizeof(__be32)) +/* IFLA_VXLAN_GROUP */
1385 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
1386 nla_total_size(sizeof(__be32))+ /* IFLA_VXLAN_LOCAL */
1387 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
1388 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
1389 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
e4f67add
DS		 1390 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
1391 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
1392 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
1393 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
d342894c 1394 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
1395 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
05f47d69 1396 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
d342894c 1397 0;
1398}
1399
1400static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
1401{
1402 const struct vxlan_dev *vxlan = netdev_priv(dev);
05f47d69 1403 struct ifla_vxlan_port_range ports = {
1404 .low = htons(vxlan->port_min),
1405 .high = htons(vxlan->port_max),
1406 };
d342894c 1407
1408 if (nla_put_u32(skb, IFLA_VXLAN_ID, vxlan->vni))
1409 goto nla_put_failure;
1410
7c41c42c 1411 if (vxlan->gaddr && nla_put_be32(skb, IFLA_VXLAN_GROUP, vxlan->gaddr))
d342894c 1412 goto nla_put_failure;
1413
1414 if (vxlan->link && nla_put_u32(skb, IFLA_VXLAN_LINK, vxlan->link))
1415 goto nla_put_failure;
1416
7c41c42c 1417 if (vxlan->saddr && nla_put_be32(skb, IFLA_VXLAN_LOCAL, vxlan->saddr))
d342894c 1418 goto nla_put_failure;
1419
1420 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
1421 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
e4f67add
DS		 1422 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
1423 !!(vxlan->flags & VXLAN_F_LEARN)) ||
1424 nla_put_u8(skb, IFLA_VXLAN_PROXY,
1425 !!(vxlan->flags & VXLAN_F_PROXY)) ||
1426 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
1427 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
1428 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
1429 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
1430 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
d342894c 1431 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
1432 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax))
1433 goto nla_put_failure;
1434
05f47d69 1435 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
1436 goto nla_put_failure;
1437
d342894c 1438 return 0;
1439
1440nla_put_failure:
1441 return -EMSGSIZE;
1442}
1443
1444static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
1445 .kind = "vxlan",
1446 .maxtype = IFLA_VXLAN_MAX,
1447 .policy = vxlan_policy,
1448 .priv_size = sizeof(struct vxlan_dev),
1449 .setup = vxlan_setup,
1450 .validate = vxlan_validate,
1451 .newlink = vxlan_newlink,
1452 .dellink = vxlan_dellink,
1453 .get_size = vxlan_get_size,
1454 .fill_info = vxlan_fill_info,
1455};
1456
1457static __net_init int vxlan_init_net(struct net *net)
1458{
1459 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1460 struct sock *sk;
1461 struct sockaddr_in vxlan_addr = {
1462 .sin_family = AF_INET,
1463 .sin_addr.s_addr = htonl(INADDR_ANY),
1464 };
1465 int rc;
1466 unsigned h;
1467
1468 /* Create UDP socket for encapsulation receive. */
1469 rc = sock_create_kern(AF_INET, SOCK_DGRAM, IPPROTO_UDP, &vn->sock);
1470 if (rc < 0) {
1471 pr_debug("UDP socket create failed\n");
1472 return rc;
1473 }
bfe1b9b1 1474 /* Put in proper namespace */
1475 sk = vn->sock->sk;
1476 sk_change_net(sk, net);
d342894c 1477
1478 vxlan_addr.sin_port = htons(vxlan_port);
1479
1480 rc = kernel_bind(vn->sock, (struct sockaddr *) &vxlan_addr,
1481 sizeof(vxlan_addr));
1482 if (rc < 0) {
1483 pr_debug("bind for UDP socket %pI4:%u (%d)\n",
1484 &vxlan_addr.sin_addr, ntohs(vxlan_addr.sin_port), rc);
bfe1b9b1 1485 sk_release_kernel(sk);
d342894c 1486 vn->sock = NULL;
1487 return rc;
1488 }
1489
1490 /* Disable multicast loopback */
d342894c 1491 inet_sk(sk)->mc_loop = 0;
1492
1493 /* Mark socket as an encapsulation socket. */
1494 udp_sk(sk)->encap_type = 1;
1495 udp_sk(sk)->encap_rcv = vxlan_udp_encap_recv;
1496 udp_encap_enable();
1497
1498 for (h = 0; h < VNI_HASH_SIZE; ++h)
1499 INIT_HLIST_HEAD(&vn->vni_list[h]);
1500
1501 return 0;
1502}
1503
1504static __net_exit void vxlan_exit_net(struct net *net)
1505{
1506 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1507
1508 if (vn->sock) {
bfe1b9b1 1509 sk_release_kernel(vn->sock->sk);
d342894c 1510 vn->sock = NULL;
1511 }
1512}
1513
1514static struct pernet_operations vxlan_net_ops = {
1515 .init = vxlan_init_net,
1516 .exit = vxlan_exit_net,
1517 .id = &vxlan_net_id,
1518 .size = sizeof(struct vxlan_net),
1519};
1520
1521static int __init vxlan_init_module(void)
1522{
1523 int rc;
1524
1525 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
1526
1527 rc = register_pernet_device(&vxlan_net_ops);
1528 if (rc)
1529 goto out1;
1530
1531 rc = rtnl_link_register(&vxlan_link_ops);
1532 if (rc)
1533 goto out2;
1534
1535 return 0;
1536
1537out2:
1538 unregister_pernet_device(&vxlan_net_ops);
1539out1:
1540 return rc;
1541}
1542module_init(vxlan_init_module);
1543
1544static void __exit vxlan_cleanup_module(void)
1545{
1546 rtnl_link_unregister(&vxlan_link_ops);
1547 unregister_pernet_device(&vxlan_net_ops);
1548}
1549module_exit(vxlan_cleanup_module);
1550
1551MODULE_LICENSE("GPL");
1552MODULE_VERSION(VXLAN_VERSION);
1553MODULE_AUTHOR("Stephen Hemminger <shemminger@vyatta.com>");
1554MODULE_ALIAS_RTNL_LINK("vxlan");
Linux 6.x block layer and io_uring tree(s)