Loading...
Note: File does not exist in v6.8.
1/*
2 * VXLAN: Virtual eXtensible Local Area Network
3 *
4 * Copyright (c) 2012-2013 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
11#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12
13#include <linux/kernel.h>
14#include <linux/module.h>
15#include <linux/errno.h>
16#include <linux/slab.h>
17#include <linux/udp.h>
18#include <linux/igmp.h>
19#include <linux/if_ether.h>
20#include <linux/ethtool.h>
21#include <net/arp.h>
22#include <net/ndisc.h>
23#include <net/ip.h>
24#include <net/icmp.h>
25#include <net/rtnetlink.h>
26#include <net/inet_ecn.h>
27#include <net/net_namespace.h>
28#include <net/netns/generic.h>
29#include <net/vxlan.h>
30
31#if IS_ENABLED(CONFIG_IPV6)
32#include <net/ip6_tunnel.h>
33#include <net/ip6_checksum.h>
34#endif
35
36#define VXLAN_VERSION "0.1"
37
38#define PORT_HASH_BITS 8
39#define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
40#define FDB_AGE_DEFAULT 300 /* 5 min */
41#define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
42
43/* UDP port for VXLAN traffic.
44 * The IANA assigned port is 4789, but the Linux default is 8472
45 * for compatibility with early adopters.
46 */
47static unsigned short vxlan_port __read_mostly = 8472;
48module_param_named(udp_port, vxlan_port, ushort, 0444);
49MODULE_PARM_DESC(udp_port, "Destination UDP port");
50
51static bool log_ecn_error = true;
52module_param(log_ecn_error, bool, 0644);
53MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
54
55static unsigned int vxlan_net_id;
56static struct rtnl_link_ops vxlan_link_ops;
57
58static const u8 all_zeros_mac[ETH_ALEN + 2];
59
60static int vxlan_sock_add(struct vxlan_dev *vxlan);
61
62/* per-network namespace private data for this module */
63struct vxlan_net {
64 struct list_head vxlan_list;
65 struct hlist_head sock_list[PORT_HASH_SIZE];
66 spinlock_t sock_lock;
67};
68
69/* Forwarding table entry */
70struct vxlan_fdb {
71 struct hlist_node hlist; /* linked list of entries */
72 struct rcu_head rcu;
73 unsigned long updated; /* jiffies */
74 unsigned long used;
75 struct list_head remotes;
76 u8 eth_addr[ETH_ALEN];
77 u16 state; /* see ndm_state */
78 u8 flags; /* see ndm_flags */
79};
80
81/* salt for hash table */
82static u32 vxlan_salt __read_mostly;
83
84static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
85{
86 return vs->flags & VXLAN_F_COLLECT_METADATA ||
87 ip_tunnel_collect_metadata();
88}
89
90#if IS_ENABLED(CONFIG_IPV6)
91static inline
92bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
93{
94 if (a->sa.sa_family != b->sa.sa_family)
95 return false;
96 if (a->sa.sa_family == AF_INET6)
97 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
98 else
99 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
100}
101
102static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
103{
104 if (ipa->sa.sa_family == AF_INET6)
105 return ipv6_addr_any(&ipa->sin6.sin6_addr);
106 else
107 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
108}
109
110static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
111{
112 if (ipa->sa.sa_family == AF_INET6)
113 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
114 else
115 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
116}
117
118static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
119{
120 if (nla_len(nla) >= sizeof(struct in6_addr)) {
121 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
122 ip->sa.sa_family = AF_INET6;
123 return 0;
124 } else if (nla_len(nla) >= sizeof(__be32)) {
125 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
126 ip->sa.sa_family = AF_INET;
127 return 0;
128 } else {
129 return -EAFNOSUPPORT;
130 }
131}
132
133static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
134 const union vxlan_addr *ip)
135{
136 if (ip->sa.sa_family == AF_INET6)
137 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
138 else
139 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
140}
141
142#else /* !CONFIG_IPV6 */
143
144static inline
145bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
146{
147 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
148}
149
150static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
151{
152 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
153}
154
155static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
156{
157 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
158}
159
160static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
161{
162 if (nla_len(nla) >= sizeof(struct in6_addr)) {
163 return -EAFNOSUPPORT;
164 } else if (nla_len(nla) >= sizeof(__be32)) {
165 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
166 ip->sa.sa_family = AF_INET;
167 return 0;
168 } else {
169 return -EAFNOSUPPORT;
170 }
171}
172
173static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
174 const union vxlan_addr *ip)
175{
176 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
177}
178#endif
179
180/* Virtual Network hash table head */
181static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
182{
183 return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
184}
185
186/* Socket hash table head */
187static inline struct hlist_head *vs_head(struct net *net, __be16 port)
188{
189 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
190
191 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
192}
193
194/* First remote destination for a forwarding entry.
195 * Guaranteed to be non-NULL because remotes are never deleted.
196 */
197static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
198{
199 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
200}
201
202static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
203{
204 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
205}
206
207/* Find VXLAN socket based on network namespace, address family and UDP port
208 * and enabled unshareable flags.
209 */
210static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
211 __be16 port, u32 flags)
212{
213 struct vxlan_sock *vs;
214
215 flags &= VXLAN_F_RCV_FLAGS;
216
217 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
218 if (inet_sk(vs->sock->sk)->inet_sport == port &&
219 vxlan_get_sk_family(vs) == family &&
220 vs->flags == flags)
221 return vs;
222 }
223 return NULL;
224}
225
226static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
227{
228 struct vxlan_dev *vxlan;
229
230 /* For flow based devices, map all packets to VNI 0 */
231 if (vs->flags & VXLAN_F_COLLECT_METADATA)
232 vni = 0;
233
234 hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
235 if (vxlan->default_dst.remote_vni == vni)
236 return vxlan;
237 }
238
239 return NULL;
240}
241
242/* Look up VNI in a per net namespace table */
243static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
244 sa_family_t family, __be16 port,
245 u32 flags)
246{
247 struct vxlan_sock *vs;
248
249 vs = vxlan_find_sock(net, family, port, flags);
250 if (!vs)
251 return NULL;
252
253 return vxlan_vs_find_vni(vs, vni);
254}
255
256/* Fill in neighbour message in skbuff. */
257static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
258 const struct vxlan_fdb *fdb,
259 u32 portid, u32 seq, int type, unsigned int flags,
260 const struct vxlan_rdst *rdst)
261{
262 unsigned long now = jiffies;
263 struct nda_cacheinfo ci;
264 struct nlmsghdr *nlh;
265 struct ndmsg *ndm;
266 bool send_ip, send_eth;
267
268 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
269 if (nlh == NULL)
270 return -EMSGSIZE;
271
272 ndm = nlmsg_data(nlh);
273 memset(ndm, 0, sizeof(*ndm));
274
275 send_eth = send_ip = true;
276
277 if (type == RTM_GETNEIGH) {
278 ndm->ndm_family = AF_INET;
279 send_ip = !vxlan_addr_any(&rdst->remote_ip);
280 send_eth = !is_zero_ether_addr(fdb->eth_addr);
281 } else
282 ndm->ndm_family = AF_BRIDGE;
283 ndm->ndm_state = fdb->state;
284 ndm->ndm_ifindex = vxlan->dev->ifindex;
285 ndm->ndm_flags = fdb->flags;
286 ndm->ndm_type = RTN_UNICAST;
287
288 if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
289 nla_put_s32(skb, NDA_LINK_NETNSID,
290 peernet2id(dev_net(vxlan->dev), vxlan->net)))
291 goto nla_put_failure;
292
293 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
294 goto nla_put_failure;
295
296 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
297 goto nla_put_failure;
298
299 if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
300 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
301 goto nla_put_failure;
302 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
303 nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
304 goto nla_put_failure;
305 if (rdst->remote_ifindex &&
306 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
307 goto nla_put_failure;
308
309 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
310 ci.ndm_confirmed = 0;
311 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
312 ci.ndm_refcnt = 0;
313
314 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
315 goto nla_put_failure;
316
317 nlmsg_end(skb, nlh);
318 return 0;
319
320nla_put_failure:
321 nlmsg_cancel(skb, nlh);
322 return -EMSGSIZE;
323}
324
325static inline size_t vxlan_nlmsg_size(void)
326{
327 return NLMSG_ALIGN(sizeof(struct ndmsg))
328 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
329 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
330 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
331 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
332 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
333 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
334 + nla_total_size(sizeof(struct nda_cacheinfo));
335}
336
337static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
338 struct vxlan_rdst *rd, int type)
339{
340 struct net *net = dev_net(vxlan->dev);
341 struct sk_buff *skb;
342 int err = -ENOBUFS;
343
344 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
345 if (skb == NULL)
346 goto errout;
347
348 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
349 if (err < 0) {
350 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
351 WARN_ON(err == -EMSGSIZE);
352 kfree_skb(skb);
353 goto errout;
354 }
355
356 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
357 return;
358errout:
359 if (err < 0)
360 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
361}
362
363static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
364{
365 struct vxlan_dev *vxlan = netdev_priv(dev);
366 struct vxlan_fdb f = {
367 .state = NUD_STALE,
368 };
369 struct vxlan_rdst remote = {
370 .remote_ip = *ipa, /* goes to NDA_DST */
371 .remote_vni = cpu_to_be32(VXLAN_N_VID),
372 };
373
374 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
375}
376
377static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
378{
379 struct vxlan_fdb f = {
380 .state = NUD_STALE,
381 };
382 struct vxlan_rdst remote = { };
383
384 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
385
386 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
387}
388
389/* Hash Ethernet address */
390static u32 eth_hash(const unsigned char *addr)
391{
392 u64 value = get_unaligned((u64 *)addr);
393
394 /* only want 6 bytes */
395#ifdef __BIG_ENDIAN
396 value >>= 16;
397#else
398 value <<= 16;
399#endif
400 return hash_64(value, FDB_HASH_BITS);
401}
402
403/* Hash chain to use given mac address */
404static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
405 const u8 *mac)
406{
407 return &vxlan->fdb_head[eth_hash(mac)];
408}
409
410/* Look up Ethernet address in forwarding table */
411static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
412 const u8 *mac)
413{
414 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
415 struct vxlan_fdb *f;
416
417 hlist_for_each_entry_rcu(f, head, hlist) {
418 if (ether_addr_equal(mac, f->eth_addr))
419 return f;
420 }
421
422 return NULL;
423}
424
425static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
426 const u8 *mac)
427{
428 struct vxlan_fdb *f;
429
430 f = __vxlan_find_mac(vxlan, mac);
431 if (f)
432 f->used = jiffies;
433
434 return f;
435}
436
437/* caller should hold vxlan->hash_lock */
438static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
439 union vxlan_addr *ip, __be16 port,
440 __be32 vni, __u32 ifindex)
441{
442 struct vxlan_rdst *rd;
443
444 list_for_each_entry(rd, &f->remotes, list) {
445 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
446 rd->remote_port == port &&
447 rd->remote_vni == vni &&
448 rd->remote_ifindex == ifindex)
449 return rd;
450 }
451
452 return NULL;
453}
454
455/* Replace destination of unicast mac */
456static int vxlan_fdb_replace(struct vxlan_fdb *f,
457 union vxlan_addr *ip, __be16 port, __be32 vni,
458 __u32 ifindex)
459{
460 struct vxlan_rdst *rd;
461
462 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
463 if (rd)
464 return 0;
465
466 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
467 if (!rd)
468 return 0;
469
470 dst_cache_reset(&rd->dst_cache);
471 rd->remote_ip = *ip;
472 rd->remote_port = port;
473 rd->remote_vni = vni;
474 rd->remote_ifindex = ifindex;
475 return 1;
476}
477
478/* Add/update destinations for multicast */
479static int vxlan_fdb_append(struct vxlan_fdb *f,
480 union vxlan_addr *ip, __be16 port, __be32 vni,
481 __u32 ifindex, struct vxlan_rdst **rdp)
482{
483 struct vxlan_rdst *rd;
484
485 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
486 if (rd)
487 return 0;
488
489 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
490 if (rd == NULL)
491 return -ENOBUFS;
492
493 if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
494 kfree(rd);
495 return -ENOBUFS;
496 }
497
498 rd->remote_ip = *ip;
499 rd->remote_port = port;
500 rd->remote_vni = vni;
501 rd->remote_ifindex = ifindex;
502
503 list_add_tail_rcu(&rd->list, &f->remotes);
504
505 *rdp = rd;
506 return 1;
507}
508
509static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
510 unsigned int off,
511 struct vxlanhdr *vh, size_t hdrlen,
512 __be32 vni_field,
513 struct gro_remcsum *grc,
514 bool nopartial)
515{
516 size_t start, offset;
517
518 if (skb->remcsum_offload)
519 return vh;
520
521 if (!NAPI_GRO_CB(skb)->csum_valid)
522 return NULL;
523
524 start = vxlan_rco_start(vni_field);
525 offset = start + vxlan_rco_offset(vni_field);
526
527 vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
528 start, offset, grc, nopartial);
529
530 skb->remcsum_offload = 1;
531
532 return vh;
533}
534
535static struct sk_buff **vxlan_gro_receive(struct sock *sk,
536 struct sk_buff **head,
537 struct sk_buff *skb)
538{
539 struct sk_buff *p, **pp = NULL;
540 struct vxlanhdr *vh, *vh2;
541 unsigned int hlen, off_vx;
542 int flush = 1;
543 struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
544 __be32 flags;
545 struct gro_remcsum grc;
546
547 skb_gro_remcsum_init(&grc);
548
549 off_vx = skb_gro_offset(skb);
550 hlen = off_vx + sizeof(*vh);
551 vh = skb_gro_header_fast(skb, off_vx);
552 if (skb_gro_header_hard(skb, hlen)) {
553 vh = skb_gro_header_slow(skb, hlen, off_vx);
554 if (unlikely(!vh))
555 goto out;
556 }
557
558 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
559
560 flags = vh->vx_flags;
561
562 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
563 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
564 vh->vx_vni, &grc,
565 !!(vs->flags &
566 VXLAN_F_REMCSUM_NOPARTIAL));
567
568 if (!vh)
569 goto out;
570 }
571
572 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
573
574 for (p = *head; p; p = p->next) {
575 if (!NAPI_GRO_CB(p)->same_flow)
576 continue;
577
578 vh2 = (struct vxlanhdr *)(p->data + off_vx);
579 if (vh->vx_flags != vh2->vx_flags ||
580 vh->vx_vni != vh2->vx_vni) {
581 NAPI_GRO_CB(p)->same_flow = 0;
582 continue;
583 }
584 }
585
586 pp = call_gro_receive(eth_gro_receive, head, skb);
587 flush = 0;
588
589out:
590 skb_gro_remcsum_cleanup(skb, &grc);
591 NAPI_GRO_CB(skb)->flush |= flush;
592
593 return pp;
594}
595
596static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
597{
598 /* Sets 'skb->inner_mac_header' since we are always called with
599 * 'skb->encapsulation' set.
600 */
601 return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
602}
603
604/* Add new entry to forwarding table -- assumes lock held */
605static int vxlan_fdb_create(struct vxlan_dev *vxlan,
606 const u8 *mac, union vxlan_addr *ip,
607 __u16 state, __u16 flags,
608 __be16 port, __be32 vni, __u32 ifindex,
609 __u8 ndm_flags)
610{
611 struct vxlan_rdst *rd = NULL;
612 struct vxlan_fdb *f;
613 int notify = 0;
614 int rc;
615
616 f = __vxlan_find_mac(vxlan, mac);
617 if (f) {
618 if (flags & NLM_F_EXCL) {
619 netdev_dbg(vxlan->dev,
620 "lost race to create %pM\n", mac);
621 return -EEXIST;
622 }
623 if (f->state != state) {
624 f->state = state;
625 f->updated = jiffies;
626 notify = 1;
627 }
628 if (f->flags != ndm_flags) {
629 f->flags = ndm_flags;
630 f->updated = jiffies;
631 notify = 1;
632 }
633 if ((flags & NLM_F_REPLACE)) {
634 /* Only change unicasts */
635 if (!(is_multicast_ether_addr(f->eth_addr) ||
636 is_zero_ether_addr(f->eth_addr))) {
637 notify |= vxlan_fdb_replace(f, ip, port, vni,
638 ifindex);
639 } else
640 return -EOPNOTSUPP;
641 }
642 if ((flags & NLM_F_APPEND) &&
643 (is_multicast_ether_addr(f->eth_addr) ||
644 is_zero_ether_addr(f->eth_addr))) {
645 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
646
647 if (rc < 0)
648 return rc;
649 notify |= rc;
650 }
651 } else {
652 if (!(flags & NLM_F_CREATE))
653 return -ENOENT;
654
655 if (vxlan->cfg.addrmax &&
656 vxlan->addrcnt >= vxlan->cfg.addrmax)
657 return -ENOSPC;
658
659 /* Disallow replace to add a multicast entry */
660 if ((flags & NLM_F_REPLACE) &&
661 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
662 return -EOPNOTSUPP;
663
664 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
665 f = kmalloc(sizeof(*f), GFP_ATOMIC);
666 if (!f)
667 return -ENOMEM;
668
669 notify = 1;
670 f->state = state;
671 f->flags = ndm_flags;
672 f->updated = f->used = jiffies;
673 INIT_LIST_HEAD(&f->remotes);
674 memcpy(f->eth_addr, mac, ETH_ALEN);
675
676 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
677 if (rc < 0) {
678 kfree(f);
679 return rc;
680 }
681
682 ++vxlan->addrcnt;
683 hlist_add_head_rcu(&f->hlist,
684 vxlan_fdb_head(vxlan, mac));
685 }
686
687 if (notify) {
688 if (rd == NULL)
689 rd = first_remote_rtnl(f);
690 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
691 }
692
693 return 0;
694}
695
696static void vxlan_fdb_free(struct rcu_head *head)
697{
698 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
699 struct vxlan_rdst *rd, *nd;
700
701 list_for_each_entry_safe(rd, nd, &f->remotes, list) {
702 dst_cache_destroy(&rd->dst_cache);
703 kfree(rd);
704 }
705 kfree(f);
706}
707
708static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
709{
710 netdev_dbg(vxlan->dev,
711 "delete %pM\n", f->eth_addr);
712
713 --vxlan->addrcnt;
714 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
715
716 hlist_del_rcu(&f->hlist);
717 call_rcu(&f->rcu, vxlan_fdb_free);
718}
719
720static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
721 union vxlan_addr *ip, __be16 *port, __be32 *vni,
722 u32 *ifindex)
723{
724 struct net *net = dev_net(vxlan->dev);
725 int err;
726
727 if (tb[NDA_DST]) {
728 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
729 if (err)
730 return err;
731 } else {
732 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
733 if (remote->sa.sa_family == AF_INET) {
734 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
735 ip->sa.sa_family = AF_INET;
736#if IS_ENABLED(CONFIG_IPV6)
737 } else {
738 ip->sin6.sin6_addr = in6addr_any;
739 ip->sa.sa_family = AF_INET6;
740#endif
741 }
742 }
743
744 if (tb[NDA_PORT]) {
745 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
746 return -EINVAL;
747 *port = nla_get_be16(tb[NDA_PORT]);
748 } else {
749 *port = vxlan->cfg.dst_port;
750 }
751
752 if (tb[NDA_VNI]) {
753 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
754 return -EINVAL;
755 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
756 } else {
757 *vni = vxlan->default_dst.remote_vni;
758 }
759
760 if (tb[NDA_IFINDEX]) {
761 struct net_device *tdev;
762
763 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
764 return -EINVAL;
765 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
766 tdev = __dev_get_by_index(net, *ifindex);
767 if (!tdev)
768 return -EADDRNOTAVAIL;
769 } else {
770 *ifindex = 0;
771 }
772
773 return 0;
774}
775
776/* Add static entry (via netlink) */
777static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
778 struct net_device *dev,
779 const unsigned char *addr, u16 vid, u16 flags)
780{
781 struct vxlan_dev *vxlan = netdev_priv(dev);
782 /* struct net *net = dev_net(vxlan->dev); */
783 union vxlan_addr ip;
784 __be16 port;
785 __be32 vni;
786 u32 ifindex;
787 int err;
788
789 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
790 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
791 ndm->ndm_state);
792 return -EINVAL;
793 }
794
795 if (tb[NDA_DST] == NULL)
796 return -EINVAL;
797
798 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
799 if (err)
800 return err;
801
802 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
803 return -EAFNOSUPPORT;
804
805 spin_lock_bh(&vxlan->hash_lock);
806 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
807 port, vni, ifindex, ndm->ndm_flags);
808 spin_unlock_bh(&vxlan->hash_lock);
809
810 return err;
811}
812
813/* Delete entry (via netlink) */
814static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
815 struct net_device *dev,
816 const unsigned char *addr, u16 vid)
817{
818 struct vxlan_dev *vxlan = netdev_priv(dev);
819 struct vxlan_fdb *f;
820 struct vxlan_rdst *rd = NULL;
821 union vxlan_addr ip;
822 __be16 port;
823 __be32 vni;
824 u32 ifindex;
825 int err;
826
827 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
828 if (err)
829 return err;
830
831 err = -ENOENT;
832
833 spin_lock_bh(&vxlan->hash_lock);
834 f = vxlan_find_mac(vxlan, addr);
835 if (!f)
836 goto out;
837
838 if (!vxlan_addr_any(&ip)) {
839 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
840 if (!rd)
841 goto out;
842 }
843
844 err = 0;
845
846 /* remove a destination if it's not the only one on the list,
847 * otherwise destroy the fdb entry
848 */
849 if (rd && !list_is_singular(&f->remotes)) {
850 list_del_rcu(&rd->list);
851 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
852 kfree_rcu(rd, rcu);
853 goto out;
854 }
855
856 vxlan_fdb_destroy(vxlan, f);
857
858out:
859 spin_unlock_bh(&vxlan->hash_lock);
860
861 return err;
862}
863
864/* Dump forwarding table */
865static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
866 struct net_device *dev,
867 struct net_device *filter_dev, int *idx)
868{
869 struct vxlan_dev *vxlan = netdev_priv(dev);
870 unsigned int h;
871 int err = 0;
872
873 for (h = 0; h < FDB_HASH_SIZE; ++h) {
874 struct vxlan_fdb *f;
875
876 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
877 struct vxlan_rdst *rd;
878
879 list_for_each_entry_rcu(rd, &f->remotes, list) {
880 if (*idx < cb->args[2])
881 goto skip;
882
883 err = vxlan_fdb_info(skb, vxlan, f,
884 NETLINK_CB(cb->skb).portid,
885 cb->nlh->nlmsg_seq,
886 RTM_NEWNEIGH,
887 NLM_F_MULTI, rd);
888 if (err < 0)
889 goto out;
890skip:
891 *idx += 1;
892 }
893 }
894 }
895out:
896 return err;
897}
898
899/* Watch incoming packets to learn mapping between Ethernet address
900 * and Tunnel endpoint.
901 * Return true if packet is bogus and should be dropped.
902 */
903static bool vxlan_snoop(struct net_device *dev,
904 union vxlan_addr *src_ip, const u8 *src_mac)
905{
906 struct vxlan_dev *vxlan = netdev_priv(dev);
907 struct vxlan_fdb *f;
908
909 f = vxlan_find_mac(vxlan, src_mac);
910 if (likely(f)) {
911 struct vxlan_rdst *rdst = first_remote_rcu(f);
912
913 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
914 return false;
915
916 /* Don't migrate static entries, drop packets */
917 if (f->state & NUD_NOARP)
918 return true;
919
920 if (net_ratelimit())
921 netdev_info(dev,
922 "%pM migrated from %pIS to %pIS\n",
923 src_mac, &rdst->remote_ip.sa, &src_ip->sa);
924
925 rdst->remote_ip = *src_ip;
926 f->updated = jiffies;
927 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
928 } else {
929 /* learned new entry */
930 spin_lock(&vxlan->hash_lock);
931
932 /* close off race between vxlan_flush and incoming packets */
933 if (netif_running(dev))
934 vxlan_fdb_create(vxlan, src_mac, src_ip,
935 NUD_REACHABLE,
936 NLM_F_EXCL|NLM_F_CREATE,
937 vxlan->cfg.dst_port,
938 vxlan->default_dst.remote_vni,
939 0, NTF_SELF);
940 spin_unlock(&vxlan->hash_lock);
941 }
942
943 return false;
944}
945
946/* See if multicast group is already in use by other ID */
947static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
948{
949 struct vxlan_dev *vxlan;
950 struct vxlan_sock *sock4;
951#if IS_ENABLED(CONFIG_IPV6)
952 struct vxlan_sock *sock6;
953#endif
954 unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
955
956 sock4 = rtnl_dereference(dev->vn4_sock);
957
958 /* The vxlan_sock is only used by dev, leaving group has
959 * no effect on other vxlan devices.
960 */
961 if (family == AF_INET && sock4 && atomic_read(&sock4->refcnt) == 1)
962 return false;
963#if IS_ENABLED(CONFIG_IPV6)
964 sock6 = rtnl_dereference(dev->vn6_sock);
965 if (family == AF_INET6 && sock6 && atomic_read(&sock6->refcnt) == 1)
966 return false;
967#endif
968
969 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
970 if (!netif_running(vxlan->dev) || vxlan == dev)
971 continue;
972
973 if (family == AF_INET &&
974 rtnl_dereference(vxlan->vn4_sock) != sock4)
975 continue;
976#if IS_ENABLED(CONFIG_IPV6)
977 if (family == AF_INET6 &&
978 rtnl_dereference(vxlan->vn6_sock) != sock6)
979 continue;
980#endif
981
982 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
983 &dev->default_dst.remote_ip))
984 continue;
985
986 if (vxlan->default_dst.remote_ifindex !=
987 dev->default_dst.remote_ifindex)
988 continue;
989
990 return true;
991 }
992
993 return false;
994}
995
996static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
997{
998 struct vxlan_net *vn;
999
1000 if (!vs)
1001 return false;
1002 if (!atomic_dec_and_test(&vs->refcnt))
1003 return false;
1004
1005 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1006 spin_lock(&vn->sock_lock);
1007 hlist_del_rcu(&vs->hlist);
1008 udp_tunnel_notify_del_rx_port(vs->sock,
1009 (vs->flags & VXLAN_F_GPE) ?
1010 UDP_TUNNEL_TYPE_VXLAN_GPE :
1011 UDP_TUNNEL_TYPE_VXLAN);
1012 spin_unlock(&vn->sock_lock);
1013
1014 return true;
1015}
1016
1017static void vxlan_sock_release(struct vxlan_dev *vxlan)
1018{
1019 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1020#if IS_ENABLED(CONFIG_IPV6)
1021 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1022
1023 rcu_assign_pointer(vxlan->vn6_sock, NULL);
1024#endif
1025
1026 rcu_assign_pointer(vxlan->vn4_sock, NULL);
1027 synchronize_net();
1028
1029 if (__vxlan_sock_release_prep(sock4)) {
1030 udp_tunnel_sock_release(sock4->sock);
1031 kfree(sock4);
1032 }
1033
1034#if IS_ENABLED(CONFIG_IPV6)
1035 if (__vxlan_sock_release_prep(sock6)) {
1036 udp_tunnel_sock_release(sock6->sock);
1037 kfree(sock6);
1038 }
1039#endif
1040}
1041
1042/* Update multicast group membership when first VNI on
1043 * multicast address is brought up
1044 */
1045static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1046{
1047 struct sock *sk;
1048 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1049 int ifindex = vxlan->default_dst.remote_ifindex;
1050 int ret = -EINVAL;
1051
1052 if (ip->sa.sa_family == AF_INET) {
1053 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1054 struct ip_mreqn mreq = {
1055 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1056 .imr_ifindex = ifindex,
1057 };
1058
1059 sk = sock4->sock->sk;
1060 lock_sock(sk);
1061 ret = ip_mc_join_group(sk, &mreq);
1062 release_sock(sk);
1063#if IS_ENABLED(CONFIG_IPV6)
1064 } else {
1065 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1066
1067 sk = sock6->sock->sk;
1068 lock_sock(sk);
1069 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1070 &ip->sin6.sin6_addr);
1071 release_sock(sk);
1072#endif
1073 }
1074
1075 return ret;
1076}
1077
1078/* Inverse of vxlan_igmp_join when last VNI is brought down */
1079static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1080{
1081 struct sock *sk;
1082 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1083 int ifindex = vxlan->default_dst.remote_ifindex;
1084 int ret = -EINVAL;
1085
1086 if (ip->sa.sa_family == AF_INET) {
1087 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1088 struct ip_mreqn mreq = {
1089 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1090 .imr_ifindex = ifindex,
1091 };
1092
1093 sk = sock4->sock->sk;
1094 lock_sock(sk);
1095 ret = ip_mc_leave_group(sk, &mreq);
1096 release_sock(sk);
1097#if IS_ENABLED(CONFIG_IPV6)
1098 } else {
1099 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1100
1101 sk = sock6->sock->sk;
1102 lock_sock(sk);
1103 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1104 &ip->sin6.sin6_addr);
1105 release_sock(sk);
1106#endif
1107 }
1108
1109 return ret;
1110}
1111
1112static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1113 struct sk_buff *skb, u32 vxflags)
1114{
1115 size_t start, offset;
1116
1117 if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1118 goto out;
1119
1120 start = vxlan_rco_start(unparsed->vx_vni);
1121 offset = start + vxlan_rco_offset(unparsed->vx_vni);
1122
1123 if (!pskb_may_pull(skb, offset + sizeof(u16)))
1124 return false;
1125
1126 skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1127 !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1128out:
1129 unparsed->vx_flags &= ~VXLAN_HF_RCO;
1130 unparsed->vx_vni &= VXLAN_VNI_MASK;
1131 return true;
1132}
1133
1134static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1135 struct sk_buff *skb, u32 vxflags,
1136 struct vxlan_metadata *md)
1137{
1138 struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1139 struct metadata_dst *tun_dst;
1140
1141 if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1142 goto out;
1143
1144 md->gbp = ntohs(gbp->policy_id);
1145
1146 tun_dst = (struct metadata_dst *)skb_dst(skb);
1147 if (tun_dst) {
1148 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1149 tun_dst->u.tun_info.options_len = sizeof(*md);
1150 }
1151 if (gbp->dont_learn)
1152 md->gbp |= VXLAN_GBP_DONT_LEARN;
1153
1154 if (gbp->policy_applied)
1155 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1156
1157 /* In flow-based mode, GBP is carried in dst_metadata */
1158 if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1159 skb->mark = md->gbp;
1160out:
1161 unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1162}
1163
1164static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1165 __be16 *protocol,
1166 struct sk_buff *skb, u32 vxflags)
1167{
1168 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1169
1170 /* Need to have Next Protocol set for interfaces in GPE mode. */
1171 if (!gpe->np_applied)
1172 return false;
1173 /* "The initial version is 0. If a receiver does not support the
1174 * version indicated it MUST drop the packet.
1175 */
1176 if (gpe->version != 0)
1177 return false;
1178 /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1179 * processing MUST occur." However, we don't implement OAM
1180 * processing, thus drop the packet.
1181 */
1182 if (gpe->oam_flag)
1183 return false;
1184
1185 switch (gpe->next_protocol) {
1186 case VXLAN_GPE_NP_IPV4:
1187 *protocol = htons(ETH_P_IP);
1188 break;
1189 case VXLAN_GPE_NP_IPV6:
1190 *protocol = htons(ETH_P_IPV6);
1191 break;
1192 case VXLAN_GPE_NP_ETHERNET:
1193 *protocol = htons(ETH_P_TEB);
1194 break;
1195 default:
1196 return false;
1197 }
1198
1199 unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1200 return true;
1201}
1202
1203static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1204 struct vxlan_sock *vs,
1205 struct sk_buff *skb)
1206{
1207 union vxlan_addr saddr;
1208
1209 skb_reset_mac_header(skb);
1210 skb->protocol = eth_type_trans(skb, vxlan->dev);
1211 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1212
1213 /* Ignore packet loops (and multicast echo) */
1214 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1215 return false;
1216
1217 /* Get address from the outer IP header */
1218 if (vxlan_get_sk_family(vs) == AF_INET) {
1219 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1220 saddr.sa.sa_family = AF_INET;
1221#if IS_ENABLED(CONFIG_IPV6)
1222 } else {
1223 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1224 saddr.sa.sa_family = AF_INET6;
1225#endif
1226 }
1227
1228 if ((vxlan->flags & VXLAN_F_LEARN) &&
1229 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1230 return false;
1231
1232 return true;
1233}
1234
1235static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1236 struct sk_buff *skb)
1237{
1238 int err = 0;
1239
1240 if (vxlan_get_sk_family(vs) == AF_INET)
1241 err = IP_ECN_decapsulate(oiph, skb);
1242#if IS_ENABLED(CONFIG_IPV6)
1243 else
1244 err = IP6_ECN_decapsulate(oiph, skb);
1245#endif
1246
1247 if (unlikely(err) && log_ecn_error) {
1248 if (vxlan_get_sk_family(vs) == AF_INET)
1249 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1250 &((struct iphdr *)oiph)->saddr,
1251 ((struct iphdr *)oiph)->tos);
1252 else
1253 net_info_ratelimited("non-ECT from %pI6\n",
1254 &((struct ipv6hdr *)oiph)->saddr);
1255 }
1256 return err <= 1;
1257}
1258
1259/* Callback from net/ipv4/udp.c to receive packets */
1260static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1261{
1262 struct pcpu_sw_netstats *stats;
1263 struct vxlan_dev *vxlan;
1264 struct vxlan_sock *vs;
1265 struct vxlanhdr unparsed;
1266 struct vxlan_metadata _md;
1267 struct vxlan_metadata *md = &_md;
1268 __be16 protocol = htons(ETH_P_TEB);
1269 bool raw_proto = false;
1270 void *oiph;
1271
1272 /* Need UDP and VXLAN header to be present */
1273 if (!pskb_may_pull(skb, VXLAN_HLEN))
1274 goto drop;
1275
1276 unparsed = *vxlan_hdr(skb);
1277 /* VNI flag always required to be set */
1278 if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1279 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1280 ntohl(vxlan_hdr(skb)->vx_flags),
1281 ntohl(vxlan_hdr(skb)->vx_vni));
1282 /* Return non vxlan pkt */
1283 goto drop;
1284 }
1285 unparsed.vx_flags &= ~VXLAN_HF_VNI;
1286 unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1287
1288 vs = rcu_dereference_sk_user_data(sk);
1289 if (!vs)
1290 goto drop;
1291
1292 vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1293 if (!vxlan)
1294 goto drop;
1295
1296 /* For backwards compatibility, only allow reserved fields to be
1297 * used by VXLAN extensions if explicitly requested.
1298 */
1299 if (vs->flags & VXLAN_F_GPE) {
1300 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1301 goto drop;
1302 raw_proto = true;
1303 }
1304
1305 if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1306 !net_eq(vxlan->net, dev_net(vxlan->dev))))
1307 goto drop;
1308
1309 if (vxlan_collect_metadata(vs)) {
1310 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1311 struct metadata_dst *tun_dst;
1312
1313 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1314 key32_to_tunnel_id(vni), sizeof(*md));
1315
1316 if (!tun_dst)
1317 goto drop;
1318
1319 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1320
1321 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1322 } else {
1323 memset(md, 0, sizeof(*md));
1324 }
1325
1326 if (vs->flags & VXLAN_F_REMCSUM_RX)
1327 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1328 goto drop;
1329 if (vs->flags & VXLAN_F_GBP)
1330 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1331 /* Note that GBP and GPE can never be active together. This is
1332 * ensured in vxlan_dev_configure.
1333 */
1334
1335 if (unparsed.vx_flags || unparsed.vx_vni) {
1336 /* If there are any unprocessed flags remaining treat
1337 * this as a malformed packet. This behavior diverges from
1338 * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1339 * in reserved fields are to be ignored. The approach here
1340 * maintains compatibility with previous stack code, and also
1341 * is more robust and provides a little more security in
1342 * adding extensions to VXLAN.
1343 */
1344 goto drop;
1345 }
1346
1347 if (!raw_proto) {
1348 if (!vxlan_set_mac(vxlan, vs, skb))
1349 goto drop;
1350 } else {
1351 skb_reset_mac_header(skb);
1352 skb->dev = vxlan->dev;
1353 skb->pkt_type = PACKET_HOST;
1354 }
1355
1356 oiph = skb_network_header(skb);
1357 skb_reset_network_header(skb);
1358
1359 if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1360 ++vxlan->dev->stats.rx_frame_errors;
1361 ++vxlan->dev->stats.rx_errors;
1362 goto drop;
1363 }
1364
1365 stats = this_cpu_ptr(vxlan->dev->tstats);
1366 u64_stats_update_begin(&stats->syncp);
1367 stats->rx_packets++;
1368 stats->rx_bytes += skb->len;
1369 u64_stats_update_end(&stats->syncp);
1370
1371 gro_cells_receive(&vxlan->gro_cells, skb);
1372 return 0;
1373
1374drop:
1375 /* Consume bad packet */
1376 kfree_skb(skb);
1377 return 0;
1378}
1379
1380static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1381{
1382 struct vxlan_dev *vxlan = netdev_priv(dev);
1383 struct arphdr *parp;
1384 u8 *arpptr, *sha;
1385 __be32 sip, tip;
1386 struct neighbour *n;
1387
1388 if (dev->flags & IFF_NOARP)
1389 goto out;
1390
1391 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1392 dev->stats.tx_dropped++;
1393 goto out;
1394 }
1395 parp = arp_hdr(skb);
1396
1397 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1398 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1399 parp->ar_pro != htons(ETH_P_IP) ||
1400 parp->ar_op != htons(ARPOP_REQUEST) ||
1401 parp->ar_hln != dev->addr_len ||
1402 parp->ar_pln != 4)
1403 goto out;
1404 arpptr = (u8 *)parp + sizeof(struct arphdr);
1405 sha = arpptr;
1406 arpptr += dev->addr_len; /* sha */
1407 memcpy(&sip, arpptr, sizeof(sip));
1408 arpptr += sizeof(sip);
1409 arpptr += dev->addr_len; /* tha */
1410 memcpy(&tip, arpptr, sizeof(tip));
1411
1412 if (ipv4_is_loopback(tip) ||
1413 ipv4_is_multicast(tip))
1414 goto out;
1415
1416 n = neigh_lookup(&arp_tbl, &tip, dev);
1417
1418 if (n) {
1419 struct vxlan_fdb *f;
1420 struct sk_buff *reply;
1421
1422 if (!(n->nud_state & NUD_CONNECTED)) {
1423 neigh_release(n);
1424 goto out;
1425 }
1426
1427 f = vxlan_find_mac(vxlan, n->ha);
1428 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1429 /* bridge-local neighbor */
1430 neigh_release(n);
1431 goto out;
1432 }
1433
1434 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1435 n->ha, sha);
1436
1437 neigh_release(n);
1438
1439 if (reply == NULL)
1440 goto out;
1441
1442 skb_reset_mac_header(reply);
1443 __skb_pull(reply, skb_network_offset(reply));
1444 reply->ip_summed = CHECKSUM_UNNECESSARY;
1445 reply->pkt_type = PACKET_HOST;
1446
1447 if (netif_rx_ni(reply) == NET_RX_DROP)
1448 dev->stats.rx_dropped++;
1449 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1450 union vxlan_addr ipa = {
1451 .sin.sin_addr.s_addr = tip,
1452 .sin.sin_family = AF_INET,
1453 };
1454
1455 vxlan_ip_miss(dev, &ipa);
1456 }
1457out:
1458 consume_skb(skb);
1459 return NETDEV_TX_OK;
1460}
1461
1462#if IS_ENABLED(CONFIG_IPV6)
1463static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1464 struct neighbour *n, bool isrouter)
1465{
1466 struct net_device *dev = request->dev;
1467 struct sk_buff *reply;
1468 struct nd_msg *ns, *na;
1469 struct ipv6hdr *pip6;
1470 u8 *daddr;
1471 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1472 int ns_olen;
1473 int i, len;
1474
1475 if (dev == NULL)
1476 return NULL;
1477
1478 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1479 sizeof(*na) + na_olen + dev->needed_tailroom;
1480 reply = alloc_skb(len, GFP_ATOMIC);
1481 if (reply == NULL)
1482 return NULL;
1483
1484 reply->protocol = htons(ETH_P_IPV6);
1485 reply->dev = dev;
1486 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1487 skb_push(reply, sizeof(struct ethhdr));
1488 skb_reset_mac_header(reply);
1489
1490 ns = (struct nd_msg *)skb_transport_header(request);
1491
1492 daddr = eth_hdr(request)->h_source;
1493 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1494 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1495 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1496 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1497 break;
1498 }
1499 }
1500
1501 /* Ethernet header */
1502 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1503 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1504 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1505 reply->protocol = htons(ETH_P_IPV6);
1506
1507 skb_pull(reply, sizeof(struct ethhdr));
1508 skb_reset_network_header(reply);
1509 skb_put(reply, sizeof(struct ipv6hdr));
1510
1511 /* IPv6 header */
1512
1513 pip6 = ipv6_hdr(reply);
1514 memset(pip6, 0, sizeof(struct ipv6hdr));
1515 pip6->version = 6;
1516 pip6->priority = ipv6_hdr(request)->priority;
1517 pip6->nexthdr = IPPROTO_ICMPV6;
1518 pip6->hop_limit = 255;
1519 pip6->daddr = ipv6_hdr(request)->saddr;
1520 pip6->saddr = *(struct in6_addr *)n->primary_key;
1521
1522 skb_pull(reply, sizeof(struct ipv6hdr));
1523 skb_reset_transport_header(reply);
1524
1525 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1526
1527 /* Neighbor Advertisement */
1528 memset(na, 0, sizeof(*na)+na_olen);
1529 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1530 na->icmph.icmp6_router = isrouter;
1531 na->icmph.icmp6_override = 1;
1532 na->icmph.icmp6_solicited = 1;
1533 na->target = ns->target;
1534 ether_addr_copy(&na->opt[2], n->ha);
1535 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1536 na->opt[1] = na_olen >> 3;
1537
1538 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1539 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1540 csum_partial(na, sizeof(*na)+na_olen, 0));
1541
1542 pip6->payload_len = htons(sizeof(*na)+na_olen);
1543
1544 skb_push(reply, sizeof(struct ipv6hdr));
1545
1546 reply->ip_summed = CHECKSUM_UNNECESSARY;
1547
1548 return reply;
1549}
1550
1551static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1552{
1553 struct vxlan_dev *vxlan = netdev_priv(dev);
1554 struct nd_msg *msg;
1555 const struct ipv6hdr *iphdr;
1556 const struct in6_addr *saddr, *daddr;
1557 struct neighbour *n;
1558 struct inet6_dev *in6_dev;
1559
1560 in6_dev = __in6_dev_get(dev);
1561 if (!in6_dev)
1562 goto out;
1563
1564 iphdr = ipv6_hdr(skb);
1565 saddr = &iphdr->saddr;
1566 daddr = &iphdr->daddr;
1567
1568 msg = (struct nd_msg *)skb_transport_header(skb);
1569 if (msg->icmph.icmp6_code != 0 ||
1570 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1571 goto out;
1572
1573 if (ipv6_addr_loopback(daddr) ||
1574 ipv6_addr_is_multicast(&msg->target))
1575 goto out;
1576
1577 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1578
1579 if (n) {
1580 struct vxlan_fdb *f;
1581 struct sk_buff *reply;
1582
1583 if (!(n->nud_state & NUD_CONNECTED)) {
1584 neigh_release(n);
1585 goto out;
1586 }
1587
1588 f = vxlan_find_mac(vxlan, n->ha);
1589 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1590 /* bridge-local neighbor */
1591 neigh_release(n);
1592 goto out;
1593 }
1594
1595 reply = vxlan_na_create(skb, n,
1596 !!(f ? f->flags & NTF_ROUTER : 0));
1597
1598 neigh_release(n);
1599
1600 if (reply == NULL)
1601 goto out;
1602
1603 if (netif_rx_ni(reply) == NET_RX_DROP)
1604 dev->stats.rx_dropped++;
1605
1606 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1607 union vxlan_addr ipa = {
1608 .sin6.sin6_addr = msg->target,
1609 .sin6.sin6_family = AF_INET6,
1610 };
1611
1612 vxlan_ip_miss(dev, &ipa);
1613 }
1614
1615out:
1616 consume_skb(skb);
1617 return NETDEV_TX_OK;
1618}
1619#endif
1620
1621static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1622{
1623 struct vxlan_dev *vxlan = netdev_priv(dev);
1624 struct neighbour *n;
1625
1626 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1627 return false;
1628
1629 n = NULL;
1630 switch (ntohs(eth_hdr(skb)->h_proto)) {
1631 case ETH_P_IP:
1632 {
1633 struct iphdr *pip;
1634
1635 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1636 return false;
1637 pip = ip_hdr(skb);
1638 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1639 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1640 union vxlan_addr ipa = {
1641 .sin.sin_addr.s_addr = pip->daddr,
1642 .sin.sin_family = AF_INET,
1643 };
1644
1645 vxlan_ip_miss(dev, &ipa);
1646 return false;
1647 }
1648
1649 break;
1650 }
1651#if IS_ENABLED(CONFIG_IPV6)
1652 case ETH_P_IPV6:
1653 {
1654 struct ipv6hdr *pip6;
1655
1656 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1657 return false;
1658 pip6 = ipv6_hdr(skb);
1659 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1660 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1661 union vxlan_addr ipa = {
1662 .sin6.sin6_addr = pip6->daddr,
1663 .sin6.sin6_family = AF_INET6,
1664 };
1665
1666 vxlan_ip_miss(dev, &ipa);
1667 return false;
1668 }
1669
1670 break;
1671 }
1672#endif
1673 default:
1674 return false;
1675 }
1676
1677 if (n) {
1678 bool diff;
1679
1680 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1681 if (diff) {
1682 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1683 dev->addr_len);
1684 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1685 }
1686 neigh_release(n);
1687 return diff;
1688 }
1689
1690 return false;
1691}
1692
1693static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1694 struct vxlan_metadata *md)
1695{
1696 struct vxlanhdr_gbp *gbp;
1697
1698 if (!md->gbp)
1699 return;
1700
1701 gbp = (struct vxlanhdr_gbp *)vxh;
1702 vxh->vx_flags |= VXLAN_HF_GBP;
1703
1704 if (md->gbp & VXLAN_GBP_DONT_LEARN)
1705 gbp->dont_learn = 1;
1706
1707 if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1708 gbp->policy_applied = 1;
1709
1710 gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1711}
1712
1713static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1714 __be16 protocol)
1715{
1716 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1717
1718 gpe->np_applied = 1;
1719
1720 switch (protocol) {
1721 case htons(ETH_P_IP):
1722 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1723 return 0;
1724 case htons(ETH_P_IPV6):
1725 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1726 return 0;
1727 case htons(ETH_P_TEB):
1728 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1729 return 0;
1730 }
1731 return -EPFNOSUPPORT;
1732}
1733
1734static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1735 int iphdr_len, __be32 vni,
1736 struct vxlan_metadata *md, u32 vxflags,
1737 bool udp_sum)
1738{
1739 struct vxlanhdr *vxh;
1740 int min_headroom;
1741 int err;
1742 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1743 __be16 inner_protocol = htons(ETH_P_TEB);
1744
1745 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1746 skb->ip_summed == CHECKSUM_PARTIAL) {
1747 int csum_start = skb_checksum_start_offset(skb);
1748
1749 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1750 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1751 (skb->csum_offset == offsetof(struct udphdr, check) ||
1752 skb->csum_offset == offsetof(struct tcphdr, check)))
1753 type |= SKB_GSO_TUNNEL_REMCSUM;
1754 }
1755
1756 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1757 + VXLAN_HLEN + iphdr_len;
1758
1759 /* Need space for new headers (invalidates iph ptr) */
1760 err = skb_cow_head(skb, min_headroom);
1761 if (unlikely(err))
1762 return err;
1763
1764 err = iptunnel_handle_offloads(skb, type);
1765 if (err)
1766 return err;
1767
1768 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1769 vxh->vx_flags = VXLAN_HF_VNI;
1770 vxh->vx_vni = vxlan_vni_field(vni);
1771
1772 if (type & SKB_GSO_TUNNEL_REMCSUM) {
1773 unsigned int start;
1774
1775 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1776 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1777 vxh->vx_flags |= VXLAN_HF_RCO;
1778
1779 if (!skb_is_gso(skb)) {
1780 skb->ip_summed = CHECKSUM_NONE;
1781 skb->encapsulation = 0;
1782 }
1783 }
1784
1785 if (vxflags & VXLAN_F_GBP)
1786 vxlan_build_gbp_hdr(vxh, vxflags, md);
1787 if (vxflags & VXLAN_F_GPE) {
1788 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1789 if (err < 0)
1790 return err;
1791 inner_protocol = skb->protocol;
1792 }
1793
1794 skb_set_inner_protocol(skb, inner_protocol);
1795 return 0;
1796}
1797
1798static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1799 struct vxlan_sock *sock4,
1800 struct sk_buff *skb, int oif, u8 tos,
1801 __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1802 struct dst_cache *dst_cache,
1803 const struct ip_tunnel_info *info)
1804{
1805 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1806 struct rtable *rt = NULL;
1807 struct flowi4 fl4;
1808
1809 if (!sock4)
1810 return ERR_PTR(-EIO);
1811
1812 if (tos && !info)
1813 use_cache = false;
1814 if (use_cache) {
1815 rt = dst_cache_get_ip4(dst_cache, saddr);
1816 if (rt)
1817 return rt;
1818 }
1819
1820 memset(&fl4, 0, sizeof(fl4));
1821 fl4.flowi4_oif = oif;
1822 fl4.flowi4_tos = RT_TOS(tos);
1823 fl4.flowi4_mark = skb->mark;
1824 fl4.flowi4_proto = IPPROTO_UDP;
1825 fl4.daddr = daddr;
1826 fl4.saddr = *saddr;
1827 fl4.fl4_dport = dport;
1828 fl4.fl4_sport = sport;
1829
1830 rt = ip_route_output_key(vxlan->net, &fl4);
1831 if (likely(!IS_ERR(rt))) {
1832 if (rt->dst.dev == dev) {
1833 netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1834 ip_rt_put(rt);
1835 return ERR_PTR(-ELOOP);
1836 }
1837
1838 *saddr = fl4.saddr;
1839 if (use_cache)
1840 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1841 } else {
1842 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1843 return ERR_PTR(-ENETUNREACH);
1844 }
1845 return rt;
1846}
1847
1848#if IS_ENABLED(CONFIG_IPV6)
1849static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1850 struct net_device *dev,
1851 struct vxlan_sock *sock6,
1852 struct sk_buff *skb, int oif, u8 tos,
1853 __be32 label,
1854 const struct in6_addr *daddr,
1855 struct in6_addr *saddr,
1856 __be16 dport, __be16 sport,
1857 struct dst_cache *dst_cache,
1858 const struct ip_tunnel_info *info)
1859{
1860 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1861 struct dst_entry *ndst;
1862 struct flowi6 fl6;
1863 int err;
1864
1865 if (!sock6)
1866 return ERR_PTR(-EIO);
1867
1868 if (tos && !info)
1869 use_cache = false;
1870 if (use_cache) {
1871 ndst = dst_cache_get_ip6(dst_cache, saddr);
1872 if (ndst)
1873 return ndst;
1874 }
1875
1876 memset(&fl6, 0, sizeof(fl6));
1877 fl6.flowi6_oif = oif;
1878 fl6.daddr = *daddr;
1879 fl6.saddr = *saddr;
1880 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1881 fl6.flowi6_mark = skb->mark;
1882 fl6.flowi6_proto = IPPROTO_UDP;
1883 fl6.fl6_dport = dport;
1884 fl6.fl6_sport = sport;
1885
1886 err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1887 sock6->sock->sk,
1888 &ndst, &fl6);
1889 if (unlikely(err < 0)) {
1890 netdev_dbg(dev, "no route to %pI6\n", daddr);
1891 return ERR_PTR(-ENETUNREACH);
1892 }
1893
1894 if (unlikely(ndst->dev == dev)) {
1895 netdev_dbg(dev, "circular route to %pI6\n", daddr);
1896 dst_release(ndst);
1897 return ERR_PTR(-ELOOP);
1898 }
1899
1900 *saddr = fl6.saddr;
1901 if (use_cache)
1902 dst_cache_set_ip6(dst_cache, ndst, saddr);
1903 return ndst;
1904}
1905#endif
1906
1907/* Bypass encapsulation if the destination is local */
1908static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1909 struct vxlan_dev *dst_vxlan)
1910{
1911 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1912 union vxlan_addr loopback;
1913 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1914 struct net_device *dev = skb->dev;
1915 int len = skb->len;
1916
1917 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1918 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1919 skb->pkt_type = PACKET_HOST;
1920 skb->encapsulation = 0;
1921 skb->dev = dst_vxlan->dev;
1922 __skb_pull(skb, skb_network_offset(skb));
1923
1924 if (remote_ip->sa.sa_family == AF_INET) {
1925 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1926 loopback.sa.sa_family = AF_INET;
1927#if IS_ENABLED(CONFIG_IPV6)
1928 } else {
1929 loopback.sin6.sin6_addr = in6addr_loopback;
1930 loopback.sa.sa_family = AF_INET6;
1931#endif
1932 }
1933
1934 if (dst_vxlan->flags & VXLAN_F_LEARN)
1935 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1936
1937 u64_stats_update_begin(&tx_stats->syncp);
1938 tx_stats->tx_packets++;
1939 tx_stats->tx_bytes += len;
1940 u64_stats_update_end(&tx_stats->syncp);
1941
1942 if (netif_rx(skb) == NET_RX_SUCCESS) {
1943 u64_stats_update_begin(&rx_stats->syncp);
1944 rx_stats->rx_packets++;
1945 rx_stats->rx_bytes += len;
1946 u64_stats_update_end(&rx_stats->syncp);
1947 } else {
1948 dev->stats.rx_dropped++;
1949 }
1950}
1951
1952static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
1953 struct vxlan_dev *vxlan, union vxlan_addr *daddr,
1954 __be32 dst_port, __be32 vni, struct dst_entry *dst,
1955 u32 rt_flags)
1956{
1957#if IS_ENABLED(CONFIG_IPV6)
1958 /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
1959 * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
1960 * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
1961 */
1962 BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
1963#endif
1964 /* Bypass encapsulation if the destination is local */
1965 if (rt_flags & RTCF_LOCAL &&
1966 !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1967 struct vxlan_dev *dst_vxlan;
1968
1969 dst_release(dst);
1970 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1971 daddr->sa.sa_family, dst_port,
1972 vxlan->flags);
1973 if (!dst_vxlan) {
1974 dev->stats.tx_errors++;
1975 kfree_skb(skb);
1976
1977 return -ENOENT;
1978 }
1979 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1980 return 1;
1981 }
1982
1983 return 0;
1984}
1985
1986static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1987 struct vxlan_rdst *rdst, bool did_rsc)
1988{
1989 struct dst_cache *dst_cache;
1990 struct ip_tunnel_info *info;
1991 struct vxlan_dev *vxlan = netdev_priv(dev);
1992 const struct iphdr *old_iph = ip_hdr(skb);
1993 union vxlan_addr *dst;
1994 union vxlan_addr remote_ip, local_ip;
1995 struct vxlan_metadata _md;
1996 struct vxlan_metadata *md = &_md;
1997 __be16 src_port = 0, dst_port;
1998 struct dst_entry *ndst = NULL;
1999 __be32 vni, label;
2000 __u8 tos, ttl;
2001 int err;
2002 u32 flags = vxlan->flags;
2003 bool udp_sum = false;
2004 bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2005
2006 info = skb_tunnel_info(skb);
2007
2008 if (rdst) {
2009 dst = &rdst->remote_ip;
2010 if (vxlan_addr_any(dst)) {
2011 if (did_rsc) {
2012 /* short-circuited back to local bridge */
2013 vxlan_encap_bypass(skb, vxlan, vxlan);
2014 return;
2015 }
2016 goto drop;
2017 }
2018
2019 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2020 vni = rdst->remote_vni;
2021 local_ip = vxlan->cfg.saddr;
2022 dst_cache = &rdst->dst_cache;
2023 md->gbp = skb->mark;
2024 ttl = vxlan->cfg.ttl;
2025 if (!ttl && vxlan_addr_multicast(dst))
2026 ttl = 1;
2027
2028 tos = vxlan->cfg.tos;
2029 if (tos == 1)
2030 tos = ip_tunnel_get_dsfield(old_iph, skb);
2031
2032 if (dst->sa.sa_family == AF_INET)
2033 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2034 else
2035 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2036 label = vxlan->cfg.label;
2037 } else {
2038 if (!info) {
2039 WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2040 dev->name);
2041 goto drop;
2042 }
2043 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2044 if (remote_ip.sa.sa_family == AF_INET) {
2045 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2046 local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2047 } else {
2048 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2049 local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2050 }
2051 dst = &remote_ip;
2052 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2053 vni = tunnel_id_to_key32(info->key.tun_id);
2054 dst_cache = &info->dst_cache;
2055 if (info->options_len)
2056 md = ip_tunnel_info_opts(info);
2057 ttl = info->key.ttl;
2058 tos = info->key.tos;
2059 label = info->key.label;
2060 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2061 }
2062 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2063 vxlan->cfg.port_max, true);
2064
2065 rcu_read_lock();
2066 if (dst->sa.sa_family == AF_INET) {
2067 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2068 struct rtable *rt;
2069 __be16 df = 0;
2070
2071 rt = vxlan_get_route(vxlan, dev, sock4, skb,
2072 rdst ? rdst->remote_ifindex : 0, tos,
2073 dst->sin.sin_addr.s_addr,
2074 &local_ip.sin.sin_addr.s_addr,
2075 dst_port, src_port,
2076 dst_cache, info);
2077 if (IS_ERR(rt)) {
2078 err = PTR_ERR(rt);
2079 goto tx_error;
2080 }
2081
2082 /* Bypass encapsulation if the destination is local */
2083 if (!info) {
2084 err = encap_bypass_if_local(skb, dev, vxlan, dst,
2085 dst_port, vni, &rt->dst,
2086 rt->rt_flags);
2087 if (err)
2088 goto out_unlock;
2089 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2090 df = htons(IP_DF);
2091 }
2092
2093 ndst = &rt->dst;
2094 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2095 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2096 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2097 vni, md, flags, udp_sum);
2098 if (err < 0)
2099 goto tx_error;
2100
2101 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2102 dst->sin.sin_addr.s_addr, tos, ttl, df,
2103 src_port, dst_port, xnet, !udp_sum);
2104#if IS_ENABLED(CONFIG_IPV6)
2105 } else {
2106 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2107
2108 ndst = vxlan6_get_route(vxlan, dev, sock6, skb,
2109 rdst ? rdst->remote_ifindex : 0, tos,
2110 label, &dst->sin6.sin6_addr,
2111 &local_ip.sin6.sin6_addr,
2112 dst_port, src_port,
2113 dst_cache, info);
2114 if (IS_ERR(ndst)) {
2115 err = PTR_ERR(ndst);
2116 ndst = NULL;
2117 goto tx_error;
2118 }
2119
2120 if (!info) {
2121 u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2122
2123 err = encap_bypass_if_local(skb, dev, vxlan, dst,
2124 dst_port, vni, ndst,
2125 rt6i_flags);
2126 if (err)
2127 goto out_unlock;
2128 }
2129
2130 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2131 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2132 skb_scrub_packet(skb, xnet);
2133 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2134 vni, md, flags, udp_sum);
2135 if (err < 0)
2136 goto tx_error;
2137
2138 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2139 &local_ip.sin6.sin6_addr,
2140 &dst->sin6.sin6_addr, tos, ttl,
2141 label, src_port, dst_port, !udp_sum);
2142#endif
2143 }
2144out_unlock:
2145 rcu_read_unlock();
2146 return;
2147
2148drop:
2149 dev->stats.tx_dropped++;
2150 dev_kfree_skb(skb);
2151 return;
2152
2153tx_error:
2154 rcu_read_unlock();
2155 if (err == -ELOOP)
2156 dev->stats.collisions++;
2157 else if (err == -ENETUNREACH)
2158 dev->stats.tx_carrier_errors++;
2159 dst_release(ndst);
2160 dev->stats.tx_errors++;
2161 kfree_skb(skb);
2162}
2163
2164/* Transmit local packets over Vxlan
2165 *
2166 * Outer IP header inherits ECN and DF from inner header.
2167 * Outer UDP destination is the VXLAN assigned port.
2168 * source port is based on hash of flow
2169 */
2170static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2171{
2172 struct vxlan_dev *vxlan = netdev_priv(dev);
2173 const struct ip_tunnel_info *info;
2174 struct ethhdr *eth;
2175 bool did_rsc = false;
2176 struct vxlan_rdst *rdst, *fdst = NULL;
2177 struct vxlan_fdb *f;
2178
2179 info = skb_tunnel_info(skb);
2180
2181 skb_reset_mac_header(skb);
2182
2183 if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2184 if (info && info->mode & IP_TUNNEL_INFO_TX)
2185 vxlan_xmit_one(skb, dev, NULL, false);
2186 else
2187 kfree_skb(skb);
2188 return NETDEV_TX_OK;
2189 }
2190
2191 if (vxlan->flags & VXLAN_F_PROXY) {
2192 eth = eth_hdr(skb);
2193 if (ntohs(eth->h_proto) == ETH_P_ARP)
2194 return arp_reduce(dev, skb);
2195#if IS_ENABLED(CONFIG_IPV6)
2196 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2197 pskb_may_pull(skb, sizeof(struct ipv6hdr)
2198 + sizeof(struct nd_msg)) &&
2199 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2200 struct nd_msg *msg;
2201
2202 msg = (struct nd_msg *)skb_transport_header(skb);
2203 if (msg->icmph.icmp6_code == 0 &&
2204 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2205 return neigh_reduce(dev, skb);
2206 }
2207#endif
2208 }
2209
2210 eth = eth_hdr(skb);
2211 f = vxlan_find_mac(vxlan, eth->h_dest);
2212 did_rsc = false;
2213
2214 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2215 (ntohs(eth->h_proto) == ETH_P_IP ||
2216 ntohs(eth->h_proto) == ETH_P_IPV6)) {
2217 did_rsc = route_shortcircuit(dev, skb);
2218 if (did_rsc)
2219 f = vxlan_find_mac(vxlan, eth->h_dest);
2220 }
2221
2222 if (f == NULL) {
2223 f = vxlan_find_mac(vxlan, all_zeros_mac);
2224 if (f == NULL) {
2225 if ((vxlan->flags & VXLAN_F_L2MISS) &&
2226 !is_multicast_ether_addr(eth->h_dest))
2227 vxlan_fdb_miss(vxlan, eth->h_dest);
2228
2229 dev->stats.tx_dropped++;
2230 kfree_skb(skb);
2231 return NETDEV_TX_OK;
2232 }
2233 }
2234
2235 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2236 struct sk_buff *skb1;
2237
2238 if (!fdst) {
2239 fdst = rdst;
2240 continue;
2241 }
2242 skb1 = skb_clone(skb, GFP_ATOMIC);
2243 if (skb1)
2244 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2245 }
2246
2247 if (fdst)
2248 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2249 else
2250 kfree_skb(skb);
2251 return NETDEV_TX_OK;
2252}
2253
2254/* Walk the forwarding table and purge stale entries */
2255static void vxlan_cleanup(unsigned long arg)
2256{
2257 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2258 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2259 unsigned int h;
2260
2261 if (!netif_running(vxlan->dev))
2262 return;
2263
2264 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2265 struct hlist_node *p, *n;
2266
2267 spin_lock_bh(&vxlan->hash_lock);
2268 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2269 struct vxlan_fdb *f
2270 = container_of(p, struct vxlan_fdb, hlist);
2271 unsigned long timeout;
2272
2273 if (f->state & (NUD_PERMANENT | NUD_NOARP))
2274 continue;
2275
2276 timeout = f->used + vxlan->cfg.age_interval * HZ;
2277 if (time_before_eq(timeout, jiffies)) {
2278 netdev_dbg(vxlan->dev,
2279 "garbage collect %pM\n",
2280 f->eth_addr);
2281 f->state = NUD_STALE;
2282 vxlan_fdb_destroy(vxlan, f);
2283 } else if (time_before(timeout, next_timer))
2284 next_timer = timeout;
2285 }
2286 spin_unlock_bh(&vxlan->hash_lock);
2287 }
2288
2289 mod_timer(&vxlan->age_timer, next_timer);
2290}
2291
2292static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2293{
2294 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2295 __be32 vni = vxlan->default_dst.remote_vni;
2296
2297 spin_lock(&vn->sock_lock);
2298 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2299 spin_unlock(&vn->sock_lock);
2300}
2301
2302/* Setup stats when device is created */
2303static int vxlan_init(struct net_device *dev)
2304{
2305 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2306 if (!dev->tstats)
2307 return -ENOMEM;
2308
2309 return 0;
2310}
2311
2312static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2313{
2314 struct vxlan_fdb *f;
2315
2316 spin_lock_bh(&vxlan->hash_lock);
2317 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2318 if (f)
2319 vxlan_fdb_destroy(vxlan, f);
2320 spin_unlock_bh(&vxlan->hash_lock);
2321}
2322
2323static void vxlan_uninit(struct net_device *dev)
2324{
2325 struct vxlan_dev *vxlan = netdev_priv(dev);
2326
2327 vxlan_fdb_delete_default(vxlan);
2328
2329 free_percpu(dev->tstats);
2330}
2331
2332/* Start ageing timer and join group when device is brought up */
2333static int vxlan_open(struct net_device *dev)
2334{
2335 struct vxlan_dev *vxlan = netdev_priv(dev);
2336 int ret;
2337
2338 ret = vxlan_sock_add(vxlan);
2339 if (ret < 0)
2340 return ret;
2341
2342 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2343 ret = vxlan_igmp_join(vxlan);
2344 if (ret == -EADDRINUSE)
2345 ret = 0;
2346 if (ret) {
2347 vxlan_sock_release(vxlan);
2348 return ret;
2349 }
2350 }
2351
2352 if (vxlan->cfg.age_interval)
2353 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2354
2355 return ret;
2356}
2357
2358/* Purge the forwarding table */
2359static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2360{
2361 unsigned int h;
2362
2363 spin_lock_bh(&vxlan->hash_lock);
2364 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2365 struct hlist_node *p, *n;
2366 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2367 struct vxlan_fdb *f
2368 = container_of(p, struct vxlan_fdb, hlist);
2369 if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2370 continue;
2371 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2372 if (!is_zero_ether_addr(f->eth_addr))
2373 vxlan_fdb_destroy(vxlan, f);
2374 }
2375 }
2376 spin_unlock_bh(&vxlan->hash_lock);
2377}
2378
2379/* Cleanup timer and forwarding table on shutdown */
2380static int vxlan_stop(struct net_device *dev)
2381{
2382 struct vxlan_dev *vxlan = netdev_priv(dev);
2383 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2384 int ret = 0;
2385
2386 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2387 !vxlan_group_used(vn, vxlan))
2388 ret = vxlan_igmp_leave(vxlan);
2389
2390 del_timer_sync(&vxlan->age_timer);
2391
2392 vxlan_flush(vxlan, false);
2393 vxlan_sock_release(vxlan);
2394
2395 return ret;
2396}
2397
2398/* Stub, nothing needs to be done. */
2399static void vxlan_set_multicast_list(struct net_device *dev)
2400{
2401}
2402
2403static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2404{
2405 struct vxlan_dev *vxlan = netdev_priv(dev);
2406 struct vxlan_rdst *dst = &vxlan->default_dst;
2407 struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2408 dst->remote_ifindex);
2409 bool use_ipv6 = false;
2410
2411 if (dst->remote_ip.sa.sa_family == AF_INET6)
2412 use_ipv6 = true;
2413
2414 /* This check is different than dev->max_mtu, because it looks at
2415 * the lowerdev->mtu, rather than the static dev->max_mtu
2416 */
2417 if (lowerdev) {
2418 int max_mtu = lowerdev->mtu -
2419 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2420 if (new_mtu > max_mtu)
2421 return -EINVAL;
2422 }
2423
2424 dev->mtu = new_mtu;
2425 return 0;
2426}
2427
2428static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2429{
2430 struct vxlan_dev *vxlan = netdev_priv(dev);
2431 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2432 __be16 sport, dport;
2433
2434 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2435 vxlan->cfg.port_max, true);
2436 dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2437
2438 if (ip_tunnel_info_af(info) == AF_INET) {
2439 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2440 struct rtable *rt;
2441
2442 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2443 info->key.u.ipv4.dst,
2444 &info->key.u.ipv4.src, dport, sport,
2445 &info->dst_cache, info);
2446 if (IS_ERR(rt))
2447 return PTR_ERR(rt);
2448 ip_rt_put(rt);
2449 } else {
2450#if IS_ENABLED(CONFIG_IPV6)
2451 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2452 struct dst_entry *ndst;
2453
2454 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2455 info->key.label, &info->key.u.ipv6.dst,
2456 &info->key.u.ipv6.src, dport, sport,
2457 &info->dst_cache, info);
2458 if (IS_ERR(ndst))
2459 return PTR_ERR(ndst);
2460 dst_release(ndst);
2461#else /* !CONFIG_IPV6 */
2462 return -EPFNOSUPPORT;
2463#endif
2464 }
2465 info->key.tp_src = sport;
2466 info->key.tp_dst = dport;
2467 return 0;
2468}
2469
2470static const struct net_device_ops vxlan_netdev_ether_ops = {
2471 .ndo_init = vxlan_init,
2472 .ndo_uninit = vxlan_uninit,
2473 .ndo_open = vxlan_open,
2474 .ndo_stop = vxlan_stop,
2475 .ndo_start_xmit = vxlan_xmit,
2476 .ndo_get_stats64 = ip_tunnel_get_stats64,
2477 .ndo_set_rx_mode = vxlan_set_multicast_list,
2478 .ndo_change_mtu = vxlan_change_mtu,
2479 .ndo_validate_addr = eth_validate_addr,
2480 .ndo_set_mac_address = eth_mac_addr,
2481 .ndo_fdb_add = vxlan_fdb_add,
2482 .ndo_fdb_del = vxlan_fdb_delete,
2483 .ndo_fdb_dump = vxlan_fdb_dump,
2484 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2485};
2486
2487static const struct net_device_ops vxlan_netdev_raw_ops = {
2488 .ndo_init = vxlan_init,
2489 .ndo_uninit = vxlan_uninit,
2490 .ndo_open = vxlan_open,
2491 .ndo_stop = vxlan_stop,
2492 .ndo_start_xmit = vxlan_xmit,
2493 .ndo_get_stats64 = ip_tunnel_get_stats64,
2494 .ndo_change_mtu = vxlan_change_mtu,
2495 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2496};
2497
2498/* Info for udev, that this is a virtual tunnel endpoint */
2499static struct device_type vxlan_type = {
2500 .name = "vxlan",
2501};
2502
2503/* Calls the ndo_udp_tunnel_add of the caller in order to
2504 * supply the listening VXLAN udp ports. Callers are expected
2505 * to implement the ndo_udp_tunnel_add.
2506 */
2507static void vxlan_push_rx_ports(struct net_device *dev)
2508{
2509 struct vxlan_sock *vs;
2510 struct net *net = dev_net(dev);
2511 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2512 unsigned int i;
2513
2514 spin_lock(&vn->sock_lock);
2515 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2516 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2517 udp_tunnel_push_rx_port(dev, vs->sock,
2518 (vs->flags & VXLAN_F_GPE) ?
2519 UDP_TUNNEL_TYPE_VXLAN_GPE :
2520 UDP_TUNNEL_TYPE_VXLAN);
2521 }
2522 spin_unlock(&vn->sock_lock);
2523}
2524
2525/* Initialize the device structure. */
2526static void vxlan_setup(struct net_device *dev)
2527{
2528 struct vxlan_dev *vxlan = netdev_priv(dev);
2529 unsigned int h;
2530
2531 eth_hw_addr_random(dev);
2532 ether_setup(dev);
2533
2534 dev->destructor = free_netdev;
2535 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2536
2537 dev->features |= NETIF_F_LLTX;
2538 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2539 dev->features |= NETIF_F_RXCSUM;
2540 dev->features |= NETIF_F_GSO_SOFTWARE;
2541
2542 dev->vlan_features = dev->features;
2543 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2544 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2545 netif_keep_dst(dev);
2546 dev->priv_flags |= IFF_NO_QUEUE;
2547
2548 INIT_LIST_HEAD(&vxlan->next);
2549 spin_lock_init(&vxlan->hash_lock);
2550
2551 init_timer_deferrable(&vxlan->age_timer);
2552 vxlan->age_timer.function = vxlan_cleanup;
2553 vxlan->age_timer.data = (unsigned long) vxlan;
2554
2555 vxlan->cfg.dst_port = htons(vxlan_port);
2556
2557 vxlan->dev = dev;
2558
2559 gro_cells_init(&vxlan->gro_cells, dev);
2560
2561 for (h = 0; h < FDB_HASH_SIZE; ++h)
2562 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2563}
2564
2565static void vxlan_ether_setup(struct net_device *dev)
2566{
2567 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2568 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2569 dev->netdev_ops = &vxlan_netdev_ether_ops;
2570}
2571
2572static void vxlan_raw_setup(struct net_device *dev)
2573{
2574 dev->header_ops = NULL;
2575 dev->type = ARPHRD_NONE;
2576 dev->hard_header_len = 0;
2577 dev->addr_len = 0;
2578 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2579 dev->netdev_ops = &vxlan_netdev_raw_ops;
2580}
2581
2582static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2583 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2584 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2585 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2586 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2587 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2588 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2589 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2590 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2591 [IFLA_VXLAN_LABEL] = { .type = NLA_U32 },
2592 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2593 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2594 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2595 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2596 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2597 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2598 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2599 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2600 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 },
2601 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2602 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 },
2603 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
2604 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
2605 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2606 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2607 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, },
2608 [IFLA_VXLAN_GPE] = { .type = NLA_FLAG, },
2609 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG },
2610};
2611
2612static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2613{
2614 if (tb[IFLA_ADDRESS]) {
2615 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2616 pr_debug("invalid link address (not ethernet)\n");
2617 return -EINVAL;
2618 }
2619
2620 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2621 pr_debug("invalid all zero ethernet address\n");
2622 return -EADDRNOTAVAIL;
2623 }
2624 }
2625
2626 if (!data)
2627 return -EINVAL;
2628
2629 if (data[IFLA_VXLAN_ID]) {
2630 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2631 if (id >= VXLAN_N_VID)
2632 return -ERANGE;
2633 }
2634
2635 if (data[IFLA_VXLAN_PORT_RANGE]) {
2636 const struct ifla_vxlan_port_range *p
2637 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2638
2639 if (ntohs(p->high) < ntohs(p->low)) {
2640 pr_debug("port range %u .. %u not valid\n",
2641 ntohs(p->low), ntohs(p->high));
2642 return -EINVAL;
2643 }
2644 }
2645
2646 return 0;
2647}
2648
2649static void vxlan_get_drvinfo(struct net_device *netdev,
2650 struct ethtool_drvinfo *drvinfo)
2651{
2652 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2653 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2654}
2655
2656static const struct ethtool_ops vxlan_ethtool_ops = {
2657 .get_drvinfo = vxlan_get_drvinfo,
2658 .get_link = ethtool_op_get_link,
2659};
2660
2661static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2662 __be16 port, u32 flags)
2663{
2664 struct socket *sock;
2665 struct udp_port_cfg udp_conf;
2666 int err;
2667
2668 memset(&udp_conf, 0, sizeof(udp_conf));
2669
2670 if (ipv6) {
2671 udp_conf.family = AF_INET6;
2672 udp_conf.use_udp6_rx_checksums =
2673 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2674 udp_conf.ipv6_v6only = 1;
2675 } else {
2676 udp_conf.family = AF_INET;
2677 }
2678
2679 udp_conf.local_udp_port = port;
2680
2681 /* Open UDP socket */
2682 err = udp_sock_create(net, &udp_conf, &sock);
2683 if (err < 0)
2684 return ERR_PTR(err);
2685
2686 return sock;
2687}
2688
2689/* Create new listen socket if needed */
2690static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2691 __be16 port, u32 flags)
2692{
2693 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2694 struct vxlan_sock *vs;
2695 struct socket *sock;
2696 unsigned int h;
2697 struct udp_tunnel_sock_cfg tunnel_cfg;
2698
2699 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2700 if (!vs)
2701 return ERR_PTR(-ENOMEM);
2702
2703 for (h = 0; h < VNI_HASH_SIZE; ++h)
2704 INIT_HLIST_HEAD(&vs->vni_list[h]);
2705
2706 sock = vxlan_create_sock(net, ipv6, port, flags);
2707 if (IS_ERR(sock)) {
2708 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2709 PTR_ERR(sock));
2710 kfree(vs);
2711 return ERR_CAST(sock);
2712 }
2713
2714 vs->sock = sock;
2715 atomic_set(&vs->refcnt, 1);
2716 vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2717
2718 spin_lock(&vn->sock_lock);
2719 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2720 udp_tunnel_notify_add_rx_port(sock,
2721 (vs->flags & VXLAN_F_GPE) ?
2722 UDP_TUNNEL_TYPE_VXLAN_GPE :
2723 UDP_TUNNEL_TYPE_VXLAN);
2724 spin_unlock(&vn->sock_lock);
2725
2726 /* Mark socket as an encapsulation socket. */
2727 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2728 tunnel_cfg.sk_user_data = vs;
2729 tunnel_cfg.encap_type = 1;
2730 tunnel_cfg.encap_rcv = vxlan_rcv;
2731 tunnel_cfg.encap_destroy = NULL;
2732 tunnel_cfg.gro_receive = vxlan_gro_receive;
2733 tunnel_cfg.gro_complete = vxlan_gro_complete;
2734
2735 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2736
2737 return vs;
2738}
2739
2740static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2741{
2742 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2743 struct vxlan_sock *vs = NULL;
2744
2745 if (!vxlan->cfg.no_share) {
2746 spin_lock(&vn->sock_lock);
2747 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2748 vxlan->cfg.dst_port, vxlan->flags);
2749 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2750 spin_unlock(&vn->sock_lock);
2751 return -EBUSY;
2752 }
2753 spin_unlock(&vn->sock_lock);
2754 }
2755 if (!vs)
2756 vs = vxlan_socket_create(vxlan->net, ipv6,
2757 vxlan->cfg.dst_port, vxlan->flags);
2758 if (IS_ERR(vs))
2759 return PTR_ERR(vs);
2760#if IS_ENABLED(CONFIG_IPV6)
2761 if (ipv6)
2762 rcu_assign_pointer(vxlan->vn6_sock, vs);
2763 else
2764#endif
2765 rcu_assign_pointer(vxlan->vn4_sock, vs);
2766 vxlan_vs_add_dev(vs, vxlan);
2767 return 0;
2768}
2769
2770static int vxlan_sock_add(struct vxlan_dev *vxlan)
2771{
2772 bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2773 bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2774 int ret = 0;
2775
2776 RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2777#if IS_ENABLED(CONFIG_IPV6)
2778 RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2779 if (ipv6 || metadata)
2780 ret = __vxlan_sock_add(vxlan, true);
2781#endif
2782 if (!ret && (!ipv6 || metadata))
2783 ret = __vxlan_sock_add(vxlan, false);
2784 if (ret < 0)
2785 vxlan_sock_release(vxlan);
2786 return ret;
2787}
2788
2789static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2790 struct vxlan_config *conf)
2791{
2792 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2793 struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2794 struct vxlan_rdst *dst = &vxlan->default_dst;
2795 unsigned short needed_headroom = ETH_HLEN;
2796 int err;
2797 bool use_ipv6 = false;
2798 __be16 default_port = vxlan->cfg.dst_port;
2799 struct net_device *lowerdev = NULL;
2800
2801 if (conf->flags & VXLAN_F_GPE) {
2802 /* For now, allow GPE only together with COLLECT_METADATA.
2803 * This can be relaxed later; in such case, the other side
2804 * of the PtP link will have to be provided.
2805 */
2806 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2807 !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2808 pr_info("unsupported combination of extensions\n");
2809 return -EINVAL;
2810 }
2811
2812 vxlan_raw_setup(dev);
2813 } else {
2814 vxlan_ether_setup(dev);
2815 }
2816
2817 /* MTU range: 68 - 65535 */
2818 dev->min_mtu = ETH_MIN_MTU;
2819 dev->max_mtu = ETH_MAX_MTU;
2820
2821 vxlan->net = src_net;
2822
2823 dst->remote_vni = conf->vni;
2824
2825 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2826
2827 /* Unless IPv6 is explicitly requested, assume IPv4 */
2828 if (!dst->remote_ip.sa.sa_family)
2829 dst->remote_ip.sa.sa_family = AF_INET;
2830
2831 if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2832 vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2833 if (!IS_ENABLED(CONFIG_IPV6))
2834 return -EPFNOSUPPORT;
2835 use_ipv6 = true;
2836 vxlan->flags |= VXLAN_F_IPV6;
2837 }
2838
2839 if (conf->label && !use_ipv6) {
2840 pr_info("label only supported in use with IPv6\n");
2841 return -EINVAL;
2842 }
2843
2844 if (conf->remote_ifindex) {
2845 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2846 dst->remote_ifindex = conf->remote_ifindex;
2847
2848 if (!lowerdev) {
2849 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2850 return -ENODEV;
2851 }
2852
2853#if IS_ENABLED(CONFIG_IPV6)
2854 if (use_ipv6) {
2855 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2856 if (idev && idev->cnf.disable_ipv6) {
2857 pr_info("IPv6 is disabled via sysctl\n");
2858 return -EPERM;
2859 }
2860 }
2861#endif
2862
2863 if (!conf->mtu)
2864 dev->mtu = lowerdev->mtu -
2865 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2866
2867 needed_headroom = lowerdev->hard_header_len;
2868 } else if (vxlan_addr_multicast(&dst->remote_ip)) {
2869 pr_info("multicast destination requires interface to be specified\n");
2870 return -EINVAL;
2871 }
2872
2873 if (conf->mtu) {
2874 int max_mtu = ETH_MAX_MTU;
2875
2876 if (lowerdev)
2877 max_mtu = lowerdev->mtu;
2878
2879 max_mtu -= (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2880
2881 if (conf->mtu < dev->min_mtu || conf->mtu > dev->max_mtu)
2882 return -EINVAL;
2883
2884 dev->mtu = conf->mtu;
2885
2886 if (conf->mtu > max_mtu)
2887 dev->mtu = max_mtu;
2888 }
2889
2890 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2891 needed_headroom += VXLAN6_HEADROOM;
2892 else
2893 needed_headroom += VXLAN_HEADROOM;
2894 dev->needed_headroom = needed_headroom;
2895
2896 memcpy(&vxlan->cfg, conf, sizeof(*conf));
2897 if (!vxlan->cfg.dst_port) {
2898 if (conf->flags & VXLAN_F_GPE)
2899 vxlan->cfg.dst_port = htons(4790); /* IANA VXLAN-GPE port */
2900 else
2901 vxlan->cfg.dst_port = default_port;
2902 }
2903 vxlan->flags |= conf->flags;
2904
2905 if (!vxlan->cfg.age_interval)
2906 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2907
2908 list_for_each_entry(tmp, &vn->vxlan_list, next) {
2909 if (tmp->cfg.vni == conf->vni &&
2910 (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2911 tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2912 tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2913 (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2914 (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2915 pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2916 return -EEXIST;
2917 }
2918 }
2919
2920 dev->ethtool_ops = &vxlan_ethtool_ops;
2921
2922 /* create an fdb entry for a valid default destination */
2923 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2924 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2925 &vxlan->default_dst.remote_ip,
2926 NUD_REACHABLE|NUD_PERMANENT,
2927 NLM_F_EXCL|NLM_F_CREATE,
2928 vxlan->cfg.dst_port,
2929 vxlan->default_dst.remote_vni,
2930 vxlan->default_dst.remote_ifindex,
2931 NTF_SELF);
2932 if (err)
2933 return err;
2934 }
2935
2936 err = register_netdevice(dev);
2937 if (err) {
2938 vxlan_fdb_delete_default(vxlan);
2939 return err;
2940 }
2941
2942 list_add(&vxlan->next, &vn->vxlan_list);
2943
2944 return 0;
2945}
2946
2947static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2948 struct nlattr *tb[], struct nlattr *data[])
2949{
2950 struct vxlan_config conf;
2951
2952 memset(&conf, 0, sizeof(conf));
2953
2954 if (data[IFLA_VXLAN_ID])
2955 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2956
2957 if (data[IFLA_VXLAN_GROUP]) {
2958 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2959 } else if (data[IFLA_VXLAN_GROUP6]) {
2960 if (!IS_ENABLED(CONFIG_IPV6))
2961 return -EPFNOSUPPORT;
2962
2963 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2964 conf.remote_ip.sa.sa_family = AF_INET6;
2965 }
2966
2967 if (data[IFLA_VXLAN_LOCAL]) {
2968 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2969 conf.saddr.sa.sa_family = AF_INET;
2970 } else if (data[IFLA_VXLAN_LOCAL6]) {
2971 if (!IS_ENABLED(CONFIG_IPV6))
2972 return -EPFNOSUPPORT;
2973
2974 /* TODO: respect scope id */
2975 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2976 conf.saddr.sa.sa_family = AF_INET6;
2977 }
2978
2979 if (data[IFLA_VXLAN_LINK])
2980 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2981
2982 if (data[IFLA_VXLAN_TOS])
2983 conf.tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2984
2985 if (data[IFLA_VXLAN_TTL])
2986 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2987
2988 if (data[IFLA_VXLAN_LABEL])
2989 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
2990 IPV6_FLOWLABEL_MASK;
2991
2992 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2993 conf.flags |= VXLAN_F_LEARN;
2994
2995 if (data[IFLA_VXLAN_AGEING])
2996 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2997
2998 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2999 conf.flags |= VXLAN_F_PROXY;
3000
3001 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
3002 conf.flags |= VXLAN_F_RSC;
3003
3004 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3005 conf.flags |= VXLAN_F_L2MISS;
3006
3007 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3008 conf.flags |= VXLAN_F_L3MISS;
3009
3010 if (data[IFLA_VXLAN_LIMIT])
3011 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3012
3013 if (data[IFLA_VXLAN_COLLECT_METADATA] &&
3014 nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3015 conf.flags |= VXLAN_F_COLLECT_METADATA;
3016
3017 if (data[IFLA_VXLAN_PORT_RANGE]) {
3018 const struct ifla_vxlan_port_range *p
3019 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3020 conf.port_min = ntohs(p->low);
3021 conf.port_max = ntohs(p->high);
3022 }
3023
3024 if (data[IFLA_VXLAN_PORT])
3025 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3026
3027 if (data[IFLA_VXLAN_UDP_CSUM] &&
3028 !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3029 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3030
3031 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3032 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3033 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3034
3035 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3036 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3037 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3038
3039 if (data[IFLA_VXLAN_REMCSUM_TX] &&
3040 nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3041 conf.flags |= VXLAN_F_REMCSUM_TX;
3042
3043 if (data[IFLA_VXLAN_REMCSUM_RX] &&
3044 nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3045 conf.flags |= VXLAN_F_REMCSUM_RX;
3046
3047 if (data[IFLA_VXLAN_GBP])
3048 conf.flags |= VXLAN_F_GBP;
3049
3050 if (data[IFLA_VXLAN_GPE])
3051 conf.flags |= VXLAN_F_GPE;
3052
3053 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3054 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3055
3056 if (tb[IFLA_MTU])
3057 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3058
3059 return vxlan_dev_configure(src_net, dev, &conf);
3060}
3061
3062static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3063{
3064 struct vxlan_dev *vxlan = netdev_priv(dev);
3065 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3066
3067 vxlan_flush(vxlan, true);
3068
3069 spin_lock(&vn->sock_lock);
3070 if (!hlist_unhashed(&vxlan->hlist))
3071 hlist_del_rcu(&vxlan->hlist);
3072 spin_unlock(&vn->sock_lock);
3073
3074 gro_cells_destroy(&vxlan->gro_cells);
3075 list_del(&vxlan->next);
3076 unregister_netdevice_queue(dev, head);
3077}
3078
3079static size_t vxlan_get_size(const struct net_device *dev)
3080{
3081
3082 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
3083 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3084 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3085 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3086 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
3087 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
3088 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3089 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
3090 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
3091 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
3092 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
3093 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
3094 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */
3095 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3096 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3097 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3098 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3099 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3100 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3101 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3102 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3103 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3104 0;
3105}
3106
3107static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3108{
3109 const struct vxlan_dev *vxlan = netdev_priv(dev);
3110 const struct vxlan_rdst *dst = &vxlan->default_dst;
3111 struct ifla_vxlan_port_range ports = {
3112 .low = htons(vxlan->cfg.port_min),
3113 .high = htons(vxlan->cfg.port_max),
3114 };
3115
3116 if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3117 goto nla_put_failure;
3118
3119 if (!vxlan_addr_any(&dst->remote_ip)) {
3120 if (dst->remote_ip.sa.sa_family == AF_INET) {
3121 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3122 dst->remote_ip.sin.sin_addr.s_addr))
3123 goto nla_put_failure;
3124#if IS_ENABLED(CONFIG_IPV6)
3125 } else {
3126 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3127 &dst->remote_ip.sin6.sin6_addr))
3128 goto nla_put_failure;
3129#endif
3130 }
3131 }
3132
3133 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3134 goto nla_put_failure;
3135
3136 if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3137 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3138 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3139 vxlan->cfg.saddr.sin.sin_addr.s_addr))
3140 goto nla_put_failure;
3141#if IS_ENABLED(CONFIG_IPV6)
3142 } else {
3143 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3144 &vxlan->cfg.saddr.sin6.sin6_addr))
3145 goto nla_put_failure;
3146#endif
3147 }
3148 }
3149
3150 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3151 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3152 nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3153 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3154 !!(vxlan->flags & VXLAN_F_LEARN)) ||
3155 nla_put_u8(skb, IFLA_VXLAN_PROXY,
3156 !!(vxlan->flags & VXLAN_F_PROXY)) ||
3157 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3158 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3159 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3160 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3161 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3162 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3163 !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3164 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3165 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3166 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3167 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3168 !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3169 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3170 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3171 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3172 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3173 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3174 !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3175 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3176 !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3177 goto nla_put_failure;
3178
3179 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3180 goto nla_put_failure;
3181
3182 if (vxlan->flags & VXLAN_F_GBP &&
3183 nla_put_flag(skb, IFLA_VXLAN_GBP))
3184 goto nla_put_failure;
3185
3186 if (vxlan->flags & VXLAN_F_GPE &&
3187 nla_put_flag(skb, IFLA_VXLAN_GPE))
3188 goto nla_put_failure;
3189
3190 if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3191 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3192 goto nla_put_failure;
3193
3194 return 0;
3195
3196nla_put_failure:
3197 return -EMSGSIZE;
3198}
3199
3200static struct net *vxlan_get_link_net(const struct net_device *dev)
3201{
3202 struct vxlan_dev *vxlan = netdev_priv(dev);
3203
3204 return vxlan->net;
3205}
3206
3207static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3208 .kind = "vxlan",
3209 .maxtype = IFLA_VXLAN_MAX,
3210 .policy = vxlan_policy,
3211 .priv_size = sizeof(struct vxlan_dev),
3212 .setup = vxlan_setup,
3213 .validate = vxlan_validate,
3214 .newlink = vxlan_newlink,
3215 .dellink = vxlan_dellink,
3216 .get_size = vxlan_get_size,
3217 .fill_info = vxlan_fill_info,
3218 .get_link_net = vxlan_get_link_net,
3219};
3220
3221struct net_device *vxlan_dev_create(struct net *net, const char *name,
3222 u8 name_assign_type,
3223 struct vxlan_config *conf)
3224{
3225 struct nlattr *tb[IFLA_MAX + 1];
3226 struct net_device *dev;
3227 int err;
3228
3229 memset(&tb, 0, sizeof(tb));
3230
3231 dev = rtnl_create_link(net, name, name_assign_type,
3232 &vxlan_link_ops, tb);
3233 if (IS_ERR(dev))
3234 return dev;
3235
3236 err = vxlan_dev_configure(net, dev, conf);
3237 if (err < 0) {
3238 free_netdev(dev);
3239 return ERR_PTR(err);
3240 }
3241
3242 err = rtnl_configure_link(dev, NULL);
3243 if (err < 0) {
3244 LIST_HEAD(list_kill);
3245
3246 vxlan_dellink(dev, &list_kill);
3247 unregister_netdevice_many(&list_kill);
3248 return ERR_PTR(err);
3249 }
3250
3251 return dev;
3252}
3253EXPORT_SYMBOL_GPL(vxlan_dev_create);
3254
3255static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3256 struct net_device *dev)
3257{
3258 struct vxlan_dev *vxlan, *next;
3259 LIST_HEAD(list_kill);
3260
3261 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3262 struct vxlan_rdst *dst = &vxlan->default_dst;
3263
3264 /* In case we created vxlan device with carrier
3265 * and we loose the carrier due to module unload
3266 * we also need to remove vxlan device. In other
3267 * cases, it's not necessary and remote_ifindex
3268 * is 0 here, so no matches.
3269 */
3270 if (dst->remote_ifindex == dev->ifindex)
3271 vxlan_dellink(vxlan->dev, &list_kill);
3272 }
3273
3274 unregister_netdevice_many(&list_kill);
3275}
3276
3277static int vxlan_netdevice_event(struct notifier_block *unused,
3278 unsigned long event, void *ptr)
3279{
3280 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3281 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3282
3283 if (event == NETDEV_UNREGISTER)
3284 vxlan_handle_lowerdev_unregister(vn, dev);
3285 else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3286 vxlan_push_rx_ports(dev);
3287
3288 return NOTIFY_DONE;
3289}
3290
3291static struct notifier_block vxlan_notifier_block __read_mostly = {
3292 .notifier_call = vxlan_netdevice_event,
3293};
3294
3295static __net_init int vxlan_init_net(struct net *net)
3296{
3297 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3298 unsigned int h;
3299
3300 INIT_LIST_HEAD(&vn->vxlan_list);
3301 spin_lock_init(&vn->sock_lock);
3302
3303 for (h = 0; h < PORT_HASH_SIZE; ++h)
3304 INIT_HLIST_HEAD(&vn->sock_list[h]);
3305
3306 return 0;
3307}
3308
3309static void __net_exit vxlan_exit_net(struct net *net)
3310{
3311 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3312 struct vxlan_dev *vxlan, *next;
3313 struct net_device *dev, *aux;
3314 LIST_HEAD(list);
3315
3316 rtnl_lock();
3317 for_each_netdev_safe(net, dev, aux)
3318 if (dev->rtnl_link_ops == &vxlan_link_ops)
3319 unregister_netdevice_queue(dev, &list);
3320
3321 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3322 /* If vxlan->dev is in the same netns, it has already been added
3323 * to the list by the previous loop.
3324 */
3325 if (!net_eq(dev_net(vxlan->dev), net)) {
3326 gro_cells_destroy(&vxlan->gro_cells);
3327 unregister_netdevice_queue(vxlan->dev, &list);
3328 }
3329 }
3330
3331 unregister_netdevice_many(&list);
3332 rtnl_unlock();
3333}
3334
3335static struct pernet_operations vxlan_net_ops = {
3336 .init = vxlan_init_net,
3337 .exit = vxlan_exit_net,
3338 .id = &vxlan_net_id,
3339 .size = sizeof(struct vxlan_net),
3340};
3341
3342static int __init vxlan_init_module(void)
3343{
3344 int rc;
3345
3346 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3347
3348 rc = register_pernet_subsys(&vxlan_net_ops);
3349 if (rc)
3350 goto out1;
3351
3352 rc = register_netdevice_notifier(&vxlan_notifier_block);
3353 if (rc)
3354 goto out2;
3355
3356 rc = rtnl_link_register(&vxlan_link_ops);
3357 if (rc)
3358 goto out3;
3359
3360 return 0;
3361out3:
3362 unregister_netdevice_notifier(&vxlan_notifier_block);
3363out2:
3364 unregister_pernet_subsys(&vxlan_net_ops);
3365out1:
3366 return rc;
3367}
3368late_initcall(vxlan_init_module);
3369
3370static void __exit vxlan_cleanup_module(void)
3371{
3372 rtnl_link_unregister(&vxlan_link_ops);
3373 unregister_netdevice_notifier(&vxlan_notifier_block);
3374 unregister_pernet_subsys(&vxlan_net_ops);
3375 /* rcu_barrier() is called by netns */
3376}
3377module_exit(vxlan_cleanup_module);
3378
3379MODULE_LICENSE("GPL");
3380MODULE_VERSION(VXLAN_VERSION);
3381MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3382MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3383MODULE_ALIAS_RTNL_LINK("vxlan");