]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame_incremental - drivers/net/vxlan.c
projects / mirror_ubuntu-artful-kernel.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
rtnl: simplify error return path in rtnl_create_link()
[mirror_ubuntu-artful-kernel.git] / drivers / net / vxlan.c
... / ...
CommitLineData
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 __be32 vni;
79 u8 flags; /* see ndm_flags */
80};
81
82/* salt for hash table */
83static u32 vxlan_salt __read_mostly;
84
85static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
86{
87 return vs->flags & VXLAN_F_COLLECT_METADATA ||
88 ip_tunnel_collect_metadata();
89}
90
91#if IS_ENABLED(CONFIG_IPV6)
92static inline
93bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
94{
95 if (a->sa.sa_family != b->sa.sa_family)
96 return false;
97 if (a->sa.sa_family == AF_INET6)
98 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
99 else
100 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
101}
102
103static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
104{
105 if (ipa->sa.sa_family == AF_INET6)
106 return ipv6_addr_any(&ipa->sin6.sin6_addr);
107 else
108 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
109}
110
111static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
112{
113 if (ipa->sa.sa_family == AF_INET6)
114 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
115 else
116 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
117}
118
119static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
120{
121 if (nla_len(nla) >= sizeof(struct in6_addr)) {
122 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
123 ip->sa.sa_family = AF_INET6;
124 return 0;
125 } else if (nla_len(nla) >= sizeof(__be32)) {
126 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
127 ip->sa.sa_family = AF_INET;
128 return 0;
129 } else {
130 return -EAFNOSUPPORT;
131 }
132}
133
134static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
135 const union vxlan_addr *ip)
136{
137 if (ip->sa.sa_family == AF_INET6)
138 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
139 else
140 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
141}
142
143#else /* !CONFIG_IPV6 */
144
145static inline
146bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
147{
148 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
149}
150
151static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
152{
153 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
154}
155
156static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
157{
158 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
159}
160
161static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
162{
163 if (nla_len(nla) >= sizeof(struct in6_addr)) {
164 return -EAFNOSUPPORT;
165 } else if (nla_len(nla) >= sizeof(__be32)) {
166 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
167 ip->sa.sa_family = AF_INET;
168 return 0;
169 } else {
170 return -EAFNOSUPPORT;
171 }
172}
173
174static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
175 const union vxlan_addr *ip)
176{
177 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
178}
179#endif
180
181/* Virtual Network hash table head */
182static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
183{
184 return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
185}
186
187/* Socket hash table head */
188static inline struct hlist_head *vs_head(struct net *net, __be16 port)
189{
190 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
191
192 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
193}
194
195/* First remote destination for a forwarding entry.
196 * Guaranteed to be non-NULL because remotes are never deleted.
197 */
198static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
199{
200 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
201}
202
203static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
204{
205 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
206}
207
208/* Find VXLAN socket based on network namespace, address family and UDP port
209 * and enabled unshareable flags.
210 */
211static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
212 __be16 port, u32 flags)
213{
214 struct vxlan_sock *vs;
215
216 flags &= VXLAN_F_RCV_FLAGS;
217
218 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
219 if (inet_sk(vs->sock->sk)->inet_sport == port &&
220 vxlan_get_sk_family(vs) == family &&
221 vs->flags == flags)
222 return vs;
223 }
224 return NULL;
225}
226
227static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
228{
229 struct vxlan_dev *vxlan;
230
231 /* For flow based devices, map all packets to VNI 0 */
232 if (vs->flags & VXLAN_F_COLLECT_METADATA)
233 vni = 0;
234
235 hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
236 if (vxlan->default_dst.remote_vni == vni)
237 return vxlan;
238 }
239
240 return NULL;
241}
242
243/* Look up VNI in a per net namespace table */
244static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
245 sa_family_t family, __be16 port,
246 u32 flags)
247{
248 struct vxlan_sock *vs;
249
250 vs = vxlan_find_sock(net, family, port, flags);
251 if (!vs)
252 return NULL;
253
254 return vxlan_vs_find_vni(vs, vni);
255}
256
257/* Fill in neighbour message in skbuff. */
258static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
259 const struct vxlan_fdb *fdb,
260 u32 portid, u32 seq, int type, unsigned int flags,
261 const struct vxlan_rdst *rdst)
262{
263 unsigned long now = jiffies;
264 struct nda_cacheinfo ci;
265 struct nlmsghdr *nlh;
266 struct ndmsg *ndm;
267 bool send_ip, send_eth;
268
269 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
270 if (nlh == NULL)
271 return -EMSGSIZE;
272
273 ndm = nlmsg_data(nlh);
274 memset(ndm, 0, sizeof(*ndm));
275
276 send_eth = send_ip = true;
277
278 if (type == RTM_GETNEIGH) {
279 ndm->ndm_family = AF_INET;
280 send_ip = !vxlan_addr_any(&rdst->remote_ip);
281 send_eth = !is_zero_ether_addr(fdb->eth_addr);
282 } else
283 ndm->ndm_family = AF_BRIDGE;
284 ndm->ndm_state = fdb->state;
285 ndm->ndm_ifindex = vxlan->dev->ifindex;
286 ndm->ndm_flags = fdb->flags;
287 ndm->ndm_type = RTN_UNICAST;
288
289 if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
290 nla_put_s32(skb, NDA_LINK_NETNSID,
291 peernet2id(dev_net(vxlan->dev), vxlan->net)))
292 goto nla_put_failure;
293
294 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
295 goto nla_put_failure;
296
297 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
298 goto nla_put_failure;
299
300 if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
301 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
302 goto nla_put_failure;
303 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
304 nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
305 goto nla_put_failure;
306 if ((vxlan->flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
307 nla_put_u32(skb, NDA_SRC_VNI,
308 be32_to_cpu(fdb->vni)))
309 goto nla_put_failure;
310 if (rdst->remote_ifindex &&
311 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
312 goto nla_put_failure;
313
314 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
315 ci.ndm_confirmed = 0;
316 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
317 ci.ndm_refcnt = 0;
318
319 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
320 goto nla_put_failure;
321
322 nlmsg_end(skb, nlh);
323 return 0;
324
325nla_put_failure:
326 nlmsg_cancel(skb, nlh);
327 return -EMSGSIZE;
328}
329
330static inline size_t vxlan_nlmsg_size(void)
331{
332 return NLMSG_ALIGN(sizeof(struct ndmsg))
333 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
334 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
335 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
336 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
337 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
338 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
339 + nla_total_size(sizeof(struct nda_cacheinfo));
340}
341
342static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
343 struct vxlan_rdst *rd, int type)
344{
345 struct net *net = dev_net(vxlan->dev);
346 struct sk_buff *skb;
347 int err = -ENOBUFS;
348
349 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
350 if (skb == NULL)
351 goto errout;
352
353 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
354 if (err < 0) {
355 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
356 WARN_ON(err == -EMSGSIZE);
357 kfree_skb(skb);
358 goto errout;
359 }
360
361 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
362 return;
363errout:
364 if (err < 0)
365 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
366}
367
368static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
369{
370 struct vxlan_dev *vxlan = netdev_priv(dev);
371 struct vxlan_fdb f = {
372 .state = NUD_STALE,
373 };
374 struct vxlan_rdst remote = {
375 .remote_ip = *ipa, /* goes to NDA_DST */
376 .remote_vni = cpu_to_be32(VXLAN_N_VID),
377 };
378
379 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
380}
381
382static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
383{
384 struct vxlan_fdb f = {
385 .state = NUD_STALE,
386 };
387 struct vxlan_rdst remote = { };
388
389 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
390
391 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
392}
393
394/* Hash Ethernet address */
395static u32 eth_hash(const unsigned char *addr)
396{
397 u64 value = get_unaligned((u64 *)addr);
398
399 /* only want 6 bytes */
400#ifdef __BIG_ENDIAN
401 value >>= 16;
402#else
403 value <<= 16;
404#endif
405 return hash_64(value, FDB_HASH_BITS);
406}
407
408static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
409{
410 /* use 1 byte of OUI and 3 bytes of NIC */
411 u32 key = get_unaligned((u32 *)(addr + 2));
412
413 return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
414}
415
416/* Hash chain to use given mac address */
417static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
418 const u8 *mac, __be32 vni)
419{
420 if (vxlan->flags & VXLAN_F_COLLECT_METADATA)
421 return &vxlan->fdb_head[eth_vni_hash(mac, vni)];
422 else
423 return &vxlan->fdb_head[eth_hash(mac)];
424}
425
426/* Look up Ethernet address in forwarding table */
427static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
428 const u8 *mac, __be32 vni)
429{
430 struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
431 struct vxlan_fdb *f;
432
433 hlist_for_each_entry_rcu(f, head, hlist) {
434 if (ether_addr_equal(mac, f->eth_addr)) {
435 if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
436 if (vni == f->vni)
437 return f;
438 } else {
439 return f;
440 }
441 }
442 }
443
444 return NULL;
445}
446
447static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
448 const u8 *mac, __be32 vni)
449{
450 struct vxlan_fdb *f;
451
452 f = __vxlan_find_mac(vxlan, mac, vni);
453 if (f)
454 f->used = jiffies;
455
456 return f;
457}
458
459/* caller should hold vxlan->hash_lock */
460static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
461 union vxlan_addr *ip, __be16 port,
462 __be32 vni, __u32 ifindex)
463{
464 struct vxlan_rdst *rd;
465
466 list_for_each_entry(rd, &f->remotes, list) {
467 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
468 rd->remote_port == port &&
469 rd->remote_vni == vni &&
470 rd->remote_ifindex == ifindex)
471 return rd;
472 }
473
474 return NULL;
475}
476
477/* Replace destination of unicast mac */
478static int vxlan_fdb_replace(struct vxlan_fdb *f,
479 union vxlan_addr *ip, __be16 port, __be32 vni,
480 __u32 ifindex)
481{
482 struct vxlan_rdst *rd;
483
484 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
485 if (rd)
486 return 0;
487
488 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
489 if (!rd)
490 return 0;
491
492 dst_cache_reset(&rd->dst_cache);
493 rd->remote_ip = *ip;
494 rd->remote_port = port;
495 rd->remote_vni = vni;
496 rd->remote_ifindex = ifindex;
497 return 1;
498}
499
500/* Add/update destinations for multicast */
501static int vxlan_fdb_append(struct vxlan_fdb *f,
502 union vxlan_addr *ip, __be16 port, __be32 vni,
503 __u32 ifindex, struct vxlan_rdst **rdp)
504{
505 struct vxlan_rdst *rd;
506
507 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
508 if (rd)
509 return 0;
510
511 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
512 if (rd == NULL)
513 return -ENOBUFS;
514
515 if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
516 kfree(rd);
517 return -ENOBUFS;
518 }
519
520 rd->remote_ip = *ip;
521 rd->remote_port = port;
522 rd->remote_vni = vni;
523 rd->remote_ifindex = ifindex;
524
525 list_add_tail_rcu(&rd->list, &f->remotes);
526
527 *rdp = rd;
528 return 1;
529}
530
531static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
532 unsigned int off,
533 struct vxlanhdr *vh, size_t hdrlen,
534 __be32 vni_field,
535 struct gro_remcsum *grc,
536 bool nopartial)
537{
538 size_t start, offset;
539
540 if (skb->remcsum_offload)
541 return vh;
542
543 if (!NAPI_GRO_CB(skb)->csum_valid)
544 return NULL;
545
546 start = vxlan_rco_start(vni_field);
547 offset = start + vxlan_rco_offset(vni_field);
548
549 vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
550 start, offset, grc, nopartial);
551
552 skb->remcsum_offload = 1;
553
554 return vh;
555}
556
557static struct sk_buff **vxlan_gro_receive(struct sock *sk,
558 struct sk_buff **head,
559 struct sk_buff *skb)
560{
561 struct sk_buff *p, **pp = NULL;
562 struct vxlanhdr *vh, *vh2;
563 unsigned int hlen, off_vx;
564 int flush = 1;
565 struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
566 __be32 flags;
567 struct gro_remcsum grc;
568
569 skb_gro_remcsum_init(&grc);
570
571 off_vx = skb_gro_offset(skb);
572 hlen = off_vx + sizeof(*vh);
573 vh = skb_gro_header_fast(skb, off_vx);
574 if (skb_gro_header_hard(skb, hlen)) {
575 vh = skb_gro_header_slow(skb, hlen, off_vx);
576 if (unlikely(!vh))
577 goto out;
578 }
579
580 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
581
582 flags = vh->vx_flags;
583
584 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
585 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
586 vh->vx_vni, &grc,
587 !!(vs->flags &
588 VXLAN_F_REMCSUM_NOPARTIAL));
589
590 if (!vh)
591 goto out;
592 }
593
594 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
595
596 for (p = *head; p; p = p->next) {
597 if (!NAPI_GRO_CB(p)->same_flow)
598 continue;
599
600 vh2 = (struct vxlanhdr *)(p->data + off_vx);
601 if (vh->vx_flags != vh2->vx_flags ||
602 vh->vx_vni != vh2->vx_vni) {
603 NAPI_GRO_CB(p)->same_flow = 0;
604 continue;
605 }
606 }
607
608 pp = call_gro_receive(eth_gro_receive, head, skb);
609 flush = 0;
610
611out:
612 skb_gro_remcsum_cleanup(skb, &grc);
613 NAPI_GRO_CB(skb)->flush |= flush;
614
615 return pp;
616}
617
618static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
619{
620 /* Sets 'skb->inner_mac_header' since we are always called with
621 * 'skb->encapsulation' set.
622 */
623 return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
624}
625
626/* Add new entry to forwarding table -- assumes lock held */
627static int vxlan_fdb_create(struct vxlan_dev *vxlan,
628 const u8 *mac, union vxlan_addr *ip,
629 __u16 state, __u16 flags,
630 __be16 port, __be32 src_vni, __be32 vni,
631 __u32 ifindex, __u8 ndm_flags)
632{
633 struct vxlan_rdst *rd = NULL;
634 struct vxlan_fdb *f;
635 int notify = 0;
636 int rc;
637
638 f = __vxlan_find_mac(vxlan, mac, src_vni);
639 if (f) {
640 if (flags & NLM_F_EXCL) {
641 netdev_dbg(vxlan->dev,
642 "lost race to create %pM\n", mac);
643 return -EEXIST;
644 }
645 if (f->state != state) {
646 f->state = state;
647 f->updated = jiffies;
648 notify = 1;
649 }
650 if (f->flags != ndm_flags) {
651 f->flags = ndm_flags;
652 f->updated = jiffies;
653 notify = 1;
654 }
655 if ((flags & NLM_F_REPLACE)) {
656 /* Only change unicasts */
657 if (!(is_multicast_ether_addr(f->eth_addr) ||
658 is_zero_ether_addr(f->eth_addr))) {
659 notify |= vxlan_fdb_replace(f, ip, port, vni,
660 ifindex);
661 } else
662 return -EOPNOTSUPP;
663 }
664 if ((flags & NLM_F_APPEND) &&
665 (is_multicast_ether_addr(f->eth_addr) ||
666 is_zero_ether_addr(f->eth_addr))) {
667 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
668
669 if (rc < 0)
670 return rc;
671 notify |= rc;
672 }
673 } else {
674 if (!(flags & NLM_F_CREATE))
675 return -ENOENT;
676
677 if (vxlan->cfg.addrmax &&
678 vxlan->addrcnt >= vxlan->cfg.addrmax)
679 return -ENOSPC;
680
681 /* Disallow replace to add a multicast entry */
682 if ((flags & NLM_F_REPLACE) &&
683 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
684 return -EOPNOTSUPP;
685
686 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
687 f = kmalloc(sizeof(*f), GFP_ATOMIC);
688 if (!f)
689 return -ENOMEM;
690
691 notify = 1;
692 f->state = state;
693 f->flags = ndm_flags;
694 f->updated = f->used = jiffies;
695 f->vni = src_vni;
696 INIT_LIST_HEAD(&f->remotes);
697 memcpy(f->eth_addr, mac, ETH_ALEN);
698
699 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
700 if (rc < 0) {
701 kfree(f);
702 return rc;
703 }
704
705 ++vxlan->addrcnt;
706 hlist_add_head_rcu(&f->hlist,
707 vxlan_fdb_head(vxlan, mac, src_vni));
708 }
709
710 if (notify) {
711 if (rd == NULL)
712 rd = first_remote_rtnl(f);
713 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
714 }
715
716 return 0;
717}
718
719static void vxlan_fdb_free(struct rcu_head *head)
720{
721 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
722 struct vxlan_rdst *rd, *nd;
723
724 list_for_each_entry_safe(rd, nd, &f->remotes, list) {
725 dst_cache_destroy(&rd->dst_cache);
726 kfree(rd);
727 }
728 kfree(f);
729}
730
731static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
732{
733 netdev_dbg(vxlan->dev,
734 "delete %pM\n", f->eth_addr);
735
736 --vxlan->addrcnt;
737 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
738
739 hlist_del_rcu(&f->hlist);
740 call_rcu(&f->rcu, vxlan_fdb_free);
741}
742
743static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
744 union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
745 __be32 *vni, u32 *ifindex)
746{
747 struct net *net = dev_net(vxlan->dev);
748 int err;
749
750 if (tb[NDA_DST]) {
751 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
752 if (err)
753 return err;
754 } else {
755 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
756 if (remote->sa.sa_family == AF_INET) {
757 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
758 ip->sa.sa_family = AF_INET;
759#if IS_ENABLED(CONFIG_IPV6)
760 } else {
761 ip->sin6.sin6_addr = in6addr_any;
762 ip->sa.sa_family = AF_INET6;
763#endif
764 }
765 }
766
767 if (tb[NDA_PORT]) {
768 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
769 return -EINVAL;
770 *port = nla_get_be16(tb[NDA_PORT]);
771 } else {
772 *port = vxlan->cfg.dst_port;
773 }
774
775 if (tb[NDA_VNI]) {
776 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
777 return -EINVAL;
778 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
779 } else {
780 *vni = vxlan->default_dst.remote_vni;
781 }
782
783 if (tb[NDA_SRC_VNI]) {
784 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
785 return -EINVAL;
786 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
787 } else {
788 *src_vni = vxlan->default_dst.remote_vni;
789 }
790
791 if (tb[NDA_IFINDEX]) {
792 struct net_device *tdev;
793
794 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
795 return -EINVAL;
796 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
797 tdev = __dev_get_by_index(net, *ifindex);
798 if (!tdev)
799 return -EADDRNOTAVAIL;
800 } else {
801 *ifindex = 0;
802 }
803
804 return 0;
805}
806
807/* Add static entry (via netlink) */
808static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
809 struct net_device *dev,
810 const unsigned char *addr, u16 vid, u16 flags)
811{
812 struct vxlan_dev *vxlan = netdev_priv(dev);
813 /* struct net *net = dev_net(vxlan->dev); */
814 union vxlan_addr ip;
815 __be16 port;
816 __be32 src_vni, vni;
817 u32 ifindex;
818 int err;
819
820 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
821 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
822 ndm->ndm_state);
823 return -EINVAL;
824 }
825
826 if (tb[NDA_DST] == NULL)
827 return -EINVAL;
828
829 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
830 if (err)
831 return err;
832
833 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
834 return -EAFNOSUPPORT;
835
836 spin_lock_bh(&vxlan->hash_lock);
837 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
838 port, src_vni, vni, ifindex, ndm->ndm_flags);
839 spin_unlock_bh(&vxlan->hash_lock);
840
841 return err;
842}
843
844static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
845 const unsigned char *addr, union vxlan_addr ip,
846 __be16 port, __be32 src_vni, u32 vni, u32 ifindex,
847 u16 vid)
848{
849 struct vxlan_fdb *f;
850 struct vxlan_rdst *rd = NULL;
851 int err = -ENOENT;
852
853 f = vxlan_find_mac(vxlan, addr, src_vni);
854 if (!f)
855 return err;
856
857 if (!vxlan_addr_any(&ip)) {
858 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
859 if (!rd)
860 goto out;
861 }
862
863 /* remove a destination if it's not the only one on the list,
864 * otherwise destroy the fdb entry
865 */
866 if (rd && !list_is_singular(&f->remotes)) {
867 list_del_rcu(&rd->list);
868 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
869 kfree_rcu(rd, rcu);
870 goto out;
871 }
872
873 vxlan_fdb_destroy(vxlan, f);
874
875out:
876 return 0;
877}
878
879/* Delete entry (via netlink) */
880static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
881 struct net_device *dev,
882 const unsigned char *addr, u16 vid)
883{
884 struct vxlan_dev *vxlan = netdev_priv(dev);
885 union vxlan_addr ip;
886 __be32 src_vni, vni;
887 __be16 port;
888 u32 ifindex;
889 int err;
890
891 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
892 if (err)
893 return err;
894
895 spin_lock_bh(&vxlan->hash_lock);
896 err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
897 vid);
898 spin_unlock_bh(&vxlan->hash_lock);
899
900 return err;
901}
902
903/* Dump forwarding table */
904static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
905 struct net_device *dev,
906 struct net_device *filter_dev, int *idx)
907{
908 struct vxlan_dev *vxlan = netdev_priv(dev);
909 unsigned int h;
910 int err = 0;
911
912 for (h = 0; h < FDB_HASH_SIZE; ++h) {
913 struct vxlan_fdb *f;
914
915 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
916 struct vxlan_rdst *rd;
917
918 list_for_each_entry_rcu(rd, &f->remotes, list) {
919 if (*idx < cb->args[2])
920 goto skip;
921
922 err = vxlan_fdb_info(skb, vxlan, f,
923 NETLINK_CB(cb->skb).portid,
924 cb->nlh->nlmsg_seq,
925 RTM_NEWNEIGH,
926 NLM_F_MULTI, rd);
927 if (err < 0)
928 goto out;
929skip:
930 *idx += 1;
931 }
932 }
933 }
934out:
935 return err;
936}
937
938/* Watch incoming packets to learn mapping between Ethernet address
939 * and Tunnel endpoint.
940 * Return true if packet is bogus and should be dropped.
941 */
942static bool vxlan_snoop(struct net_device *dev,
943 union vxlan_addr *src_ip, const u8 *src_mac,
944 __be32 vni)
945{
946 struct vxlan_dev *vxlan = netdev_priv(dev);
947 struct vxlan_fdb *f;
948
949 f = vxlan_find_mac(vxlan, src_mac, vni);
950 if (likely(f)) {
951 struct vxlan_rdst *rdst = first_remote_rcu(f);
952
953 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
954 return false;
955
956 /* Don't migrate static entries, drop packets */
957 if (f->state & NUD_NOARP)
958 return true;
959
960 if (net_ratelimit())
961 netdev_info(dev,
962 "%pM migrated from %pIS to %pIS\n",
963 src_mac, &rdst->remote_ip.sa, &src_ip->sa);
964
965 rdst->remote_ip = *src_ip;
966 f->updated = jiffies;
967 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
968 } else {
969 /* learned new entry */
970 spin_lock(&vxlan->hash_lock);
971
972 /* close off race between vxlan_flush and incoming packets */
973 if (netif_running(dev))
974 vxlan_fdb_create(vxlan, src_mac, src_ip,
975 NUD_REACHABLE,
976 NLM_F_EXCL|NLM_F_CREATE,
977 vxlan->cfg.dst_port,
978 vni,
979 vxlan->default_dst.remote_vni,
980 0, NTF_SELF);
981 spin_unlock(&vxlan->hash_lock);
982 }
983
984 return false;
985}
986
987/* See if multicast group is already in use by other ID */
988static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
989{
990 struct vxlan_dev *vxlan;
991 struct vxlan_sock *sock4;
992#if IS_ENABLED(CONFIG_IPV6)
993 struct vxlan_sock *sock6;
994#endif
995 unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
996
997 sock4 = rtnl_dereference(dev->vn4_sock);
998
999 /* The vxlan_sock is only used by dev, leaving group has
1000 * no effect on other vxlan devices.
1001 */
1002 if (family == AF_INET && sock4 && atomic_read(&sock4->refcnt) == 1)
1003 return false;
1004#if IS_ENABLED(CONFIG_IPV6)
1005 sock6 = rtnl_dereference(dev->vn6_sock);
1006 if (family == AF_INET6 && sock6 && atomic_read(&sock6->refcnt) == 1)
1007 return false;
1008#endif
1009
1010 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1011 if (!netif_running(vxlan->dev) || vxlan == dev)
1012 continue;
1013
1014 if (family == AF_INET &&
1015 rtnl_dereference(vxlan->vn4_sock) != sock4)
1016 continue;
1017#if IS_ENABLED(CONFIG_IPV6)
1018 if (family == AF_INET6 &&
1019 rtnl_dereference(vxlan->vn6_sock) != sock6)
1020 continue;
1021#endif
1022
1023 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1024 &dev->default_dst.remote_ip))
1025 continue;
1026
1027 if (vxlan->default_dst.remote_ifindex !=
1028 dev->default_dst.remote_ifindex)
1029 continue;
1030
1031 return true;
1032 }
1033
1034 return false;
1035}
1036
1037static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1038{
1039 struct vxlan_net *vn;
1040
1041 if (!vs)
1042 return false;
1043 if (!atomic_dec_and_test(&vs->refcnt))
1044 return false;
1045
1046 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1047 spin_lock(&vn->sock_lock);
1048 hlist_del_rcu(&vs->hlist);
1049 udp_tunnel_notify_del_rx_port(vs->sock,
1050 (vs->flags & VXLAN_F_GPE) ?
1051 UDP_TUNNEL_TYPE_VXLAN_GPE :
1052 UDP_TUNNEL_TYPE_VXLAN);
1053 spin_unlock(&vn->sock_lock);
1054
1055 return true;
1056}
1057
1058static void vxlan_sock_release(struct vxlan_dev *vxlan)
1059{
1060 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1061#if IS_ENABLED(CONFIG_IPV6)
1062 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1063
1064 rcu_assign_pointer(vxlan->vn6_sock, NULL);
1065#endif
1066
1067 rcu_assign_pointer(vxlan->vn4_sock, NULL);
1068 synchronize_net();
1069
1070 if (__vxlan_sock_release_prep(sock4)) {
1071 udp_tunnel_sock_release(sock4->sock);
1072 kfree(sock4);
1073 }
1074
1075#if IS_ENABLED(CONFIG_IPV6)
1076 if (__vxlan_sock_release_prep(sock6)) {
1077 udp_tunnel_sock_release(sock6->sock);
1078 kfree(sock6);
1079 }
1080#endif
1081}
1082
1083/* Update multicast group membership when first VNI on
1084 * multicast address is brought up
1085 */
1086static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1087{
1088 struct sock *sk;
1089 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1090 int ifindex = vxlan->default_dst.remote_ifindex;
1091 int ret = -EINVAL;
1092
1093 if (ip->sa.sa_family == AF_INET) {
1094 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1095 struct ip_mreqn mreq = {
1096 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1097 .imr_ifindex = ifindex,
1098 };
1099
1100 sk = sock4->sock->sk;
1101 lock_sock(sk);
1102 ret = ip_mc_join_group(sk, &mreq);
1103 release_sock(sk);
1104#if IS_ENABLED(CONFIG_IPV6)
1105 } else {
1106 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1107
1108 sk = sock6->sock->sk;
1109 lock_sock(sk);
1110 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1111 &ip->sin6.sin6_addr);
1112 release_sock(sk);
1113#endif
1114 }
1115
1116 return ret;
1117}
1118
1119/* Inverse of vxlan_igmp_join when last VNI is brought down */
1120static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1121{
1122 struct sock *sk;
1123 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1124 int ifindex = vxlan->default_dst.remote_ifindex;
1125 int ret = -EINVAL;
1126
1127 if (ip->sa.sa_family == AF_INET) {
1128 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1129 struct ip_mreqn mreq = {
1130 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1131 .imr_ifindex = ifindex,
1132 };
1133
1134 sk = sock4->sock->sk;
1135 lock_sock(sk);
1136 ret = ip_mc_leave_group(sk, &mreq);
1137 release_sock(sk);
1138#if IS_ENABLED(CONFIG_IPV6)
1139 } else {
1140 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1141
1142 sk = sock6->sock->sk;
1143 lock_sock(sk);
1144 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1145 &ip->sin6.sin6_addr);
1146 release_sock(sk);
1147#endif
1148 }
1149
1150 return ret;
1151}
1152
1153static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1154 struct sk_buff *skb, u32 vxflags)
1155{
1156 size_t start, offset;
1157
1158 if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1159 goto out;
1160
1161 start = vxlan_rco_start(unparsed->vx_vni);
1162 offset = start + vxlan_rco_offset(unparsed->vx_vni);
1163
1164 if (!pskb_may_pull(skb, offset + sizeof(u16)))
1165 return false;
1166
1167 skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1168 !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1169out:
1170 unparsed->vx_flags &= ~VXLAN_HF_RCO;
1171 unparsed->vx_vni &= VXLAN_VNI_MASK;
1172 return true;
1173}
1174
1175static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1176 struct sk_buff *skb, u32 vxflags,
1177 struct vxlan_metadata *md)
1178{
1179 struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1180 struct metadata_dst *tun_dst;
1181
1182 if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1183 goto out;
1184
1185 md->gbp = ntohs(gbp->policy_id);
1186
1187 tun_dst = (struct metadata_dst *)skb_dst(skb);
1188 if (tun_dst) {
1189 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1190 tun_dst->u.tun_info.options_len = sizeof(*md);
1191 }
1192 if (gbp->dont_learn)
1193 md->gbp |= VXLAN_GBP_DONT_LEARN;
1194
1195 if (gbp->policy_applied)
1196 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1197
1198 /* In flow-based mode, GBP is carried in dst_metadata */
1199 if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1200 skb->mark = md->gbp;
1201out:
1202 unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1203}
1204
1205static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1206 __be16 *protocol,
1207 struct sk_buff *skb, u32 vxflags)
1208{
1209 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1210
1211 /* Need to have Next Protocol set for interfaces in GPE mode. */
1212 if (!gpe->np_applied)
1213 return false;
1214 /* "The initial version is 0. If a receiver does not support the
1215 * version indicated it MUST drop the packet.
1216 */
1217 if (gpe->version != 0)
1218 return false;
1219 /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1220 * processing MUST occur." However, we don't implement OAM
1221 * processing, thus drop the packet.
1222 */
1223 if (gpe->oam_flag)
1224 return false;
1225
1226 switch (gpe->next_protocol) {
1227 case VXLAN_GPE_NP_IPV4:
1228 *protocol = htons(ETH_P_IP);
1229 break;
1230 case VXLAN_GPE_NP_IPV6:
1231 *protocol = htons(ETH_P_IPV6);
1232 break;
1233 case VXLAN_GPE_NP_ETHERNET:
1234 *protocol = htons(ETH_P_TEB);
1235 break;
1236 default:
1237 return false;
1238 }
1239
1240 unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1241 return true;
1242}
1243
1244static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1245 struct vxlan_sock *vs,
1246 struct sk_buff *skb, __be32 vni)
1247{
1248 union vxlan_addr saddr;
1249
1250 skb_reset_mac_header(skb);
1251 skb->protocol = eth_type_trans(skb, vxlan->dev);
1252 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1253
1254 /* Ignore packet loops (and multicast echo) */
1255 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1256 return false;
1257
1258 /* Get address from the outer IP header */
1259 if (vxlan_get_sk_family(vs) == AF_INET) {
1260 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1261 saddr.sa.sa_family = AF_INET;
1262#if IS_ENABLED(CONFIG_IPV6)
1263 } else {
1264 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1265 saddr.sa.sa_family = AF_INET6;
1266#endif
1267 }
1268
1269 if ((vxlan->flags & VXLAN_F_LEARN) &&
1270 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, vni))
1271 return false;
1272
1273 return true;
1274}
1275
1276static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1277 struct sk_buff *skb)
1278{
1279 int err = 0;
1280
1281 if (vxlan_get_sk_family(vs) == AF_INET)
1282 err = IP_ECN_decapsulate(oiph, skb);
1283#if IS_ENABLED(CONFIG_IPV6)
1284 else
1285 err = IP6_ECN_decapsulate(oiph, skb);
1286#endif
1287
1288 if (unlikely(err) && log_ecn_error) {
1289 if (vxlan_get_sk_family(vs) == AF_INET)
1290 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1291 &((struct iphdr *)oiph)->saddr,
1292 ((struct iphdr *)oiph)->tos);
1293 else
1294 net_info_ratelimited("non-ECT from %pI6\n",
1295 &((struct ipv6hdr *)oiph)->saddr);
1296 }
1297 return err <= 1;
1298}
1299
1300/* Callback from net/ipv4/udp.c to receive packets */
1301static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1302{
1303 struct pcpu_sw_netstats *stats;
1304 struct vxlan_dev *vxlan;
1305 struct vxlan_sock *vs;
1306 struct vxlanhdr unparsed;
1307 struct vxlan_metadata _md;
1308 struct vxlan_metadata *md = &_md;
1309 __be16 protocol = htons(ETH_P_TEB);
1310 bool raw_proto = false;
1311 void *oiph;
1312 __be32 vni = 0;
1313
1314 /* Need UDP and VXLAN header to be present */
1315 if (!pskb_may_pull(skb, VXLAN_HLEN))
1316 goto drop;
1317
1318 unparsed = *vxlan_hdr(skb);
1319 /* VNI flag always required to be set */
1320 if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1321 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1322 ntohl(vxlan_hdr(skb)->vx_flags),
1323 ntohl(vxlan_hdr(skb)->vx_vni));
1324 /* Return non vxlan pkt */
1325 goto drop;
1326 }
1327 unparsed.vx_flags &= ~VXLAN_HF_VNI;
1328 unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1329
1330 vs = rcu_dereference_sk_user_data(sk);
1331 if (!vs)
1332 goto drop;
1333
1334 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1335
1336 vxlan = vxlan_vs_find_vni(vs, vni);
1337 if (!vxlan)
1338 goto drop;
1339
1340 /* For backwards compatibility, only allow reserved fields to be
1341 * used by VXLAN extensions if explicitly requested.
1342 */
1343 if (vs->flags & VXLAN_F_GPE) {
1344 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1345 goto drop;
1346 raw_proto = true;
1347 }
1348
1349 if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1350 !net_eq(vxlan->net, dev_net(vxlan->dev))))
1351 goto drop;
1352
1353 if (vxlan_collect_metadata(vs)) {
1354 struct metadata_dst *tun_dst;
1355
1356 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1357 key32_to_tunnel_id(vni), sizeof(*md));
1358
1359 if (!tun_dst)
1360 goto drop;
1361
1362 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1363
1364 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1365 } else {
1366 memset(md, 0, sizeof(*md));
1367 }
1368
1369 if (vs->flags & VXLAN_F_REMCSUM_RX)
1370 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1371 goto drop;
1372 if (vs->flags & VXLAN_F_GBP)
1373 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1374 /* Note that GBP and GPE can never be active together. This is
1375 * ensured in vxlan_dev_configure.
1376 */
1377
1378 if (unparsed.vx_flags || unparsed.vx_vni) {
1379 /* If there are any unprocessed flags remaining treat
1380 * this as a malformed packet. This behavior diverges from
1381 * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1382 * in reserved fields are to be ignored. The approach here
1383 * maintains compatibility with previous stack code, and also
1384 * is more robust and provides a little more security in
1385 * adding extensions to VXLAN.
1386 */
1387 goto drop;
1388 }
1389
1390 if (!raw_proto) {
1391 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1392 goto drop;
1393 } else {
1394 skb_reset_mac_header(skb);
1395 skb->dev = vxlan->dev;
1396 skb->pkt_type = PACKET_HOST;
1397 }
1398
1399 oiph = skb_network_header(skb);
1400 skb_reset_network_header(skb);
1401
1402 if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1403 ++vxlan->dev->stats.rx_frame_errors;
1404 ++vxlan->dev->stats.rx_errors;
1405 goto drop;
1406 }
1407
1408 stats = this_cpu_ptr(vxlan->dev->tstats);
1409 u64_stats_update_begin(&stats->syncp);
1410 stats->rx_packets++;
1411 stats->rx_bytes += skb->len;
1412 u64_stats_update_end(&stats->syncp);
1413
1414 gro_cells_receive(&vxlan->gro_cells, skb);
1415 return 0;
1416
1417drop:
1418 /* Consume bad packet */
1419 kfree_skb(skb);
1420 return 0;
1421}
1422
1423static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1424{
1425 struct vxlan_dev *vxlan = netdev_priv(dev);
1426 struct arphdr *parp;
1427 u8 *arpptr, *sha;
1428 __be32 sip, tip;
1429 struct neighbour *n;
1430
1431 if (dev->flags & IFF_NOARP)
1432 goto out;
1433
1434 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1435 dev->stats.tx_dropped++;
1436 goto out;
1437 }
1438 parp = arp_hdr(skb);
1439
1440 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1441 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1442 parp->ar_pro != htons(ETH_P_IP) ||
1443 parp->ar_op != htons(ARPOP_REQUEST) ||
1444 parp->ar_hln != dev->addr_len ||
1445 parp->ar_pln != 4)
1446 goto out;
1447 arpptr = (u8 *)parp + sizeof(struct arphdr);
1448 sha = arpptr;
1449 arpptr += dev->addr_len; /* sha */
1450 memcpy(&sip, arpptr, sizeof(sip));
1451 arpptr += sizeof(sip);
1452 arpptr += dev->addr_len; /* tha */
1453 memcpy(&tip, arpptr, sizeof(tip));
1454
1455 if (ipv4_is_loopback(tip) ||
1456 ipv4_is_multicast(tip))
1457 goto out;
1458
1459 n = neigh_lookup(&arp_tbl, &tip, dev);
1460
1461 if (n) {
1462 struct vxlan_fdb *f;
1463 struct sk_buff *reply;
1464
1465 if (!(n->nud_state & NUD_CONNECTED)) {
1466 neigh_release(n);
1467 goto out;
1468 }
1469
1470 f = vxlan_find_mac(vxlan, n->ha, vni);
1471 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1472 /* bridge-local neighbor */
1473 neigh_release(n);
1474 goto out;
1475 }
1476
1477 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1478 n->ha, sha);
1479
1480 neigh_release(n);
1481
1482 if (reply == NULL)
1483 goto out;
1484
1485 skb_reset_mac_header(reply);
1486 __skb_pull(reply, skb_network_offset(reply));
1487 reply->ip_summed = CHECKSUM_UNNECESSARY;
1488 reply->pkt_type = PACKET_HOST;
1489
1490 if (netif_rx_ni(reply) == NET_RX_DROP)
1491 dev->stats.rx_dropped++;
1492 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1493 union vxlan_addr ipa = {
1494 .sin.sin_addr.s_addr = tip,
1495 .sin.sin_family = AF_INET,
1496 };
1497
1498 vxlan_ip_miss(dev, &ipa);
1499 }
1500out:
1501 consume_skb(skb);
1502 return NETDEV_TX_OK;
1503}
1504
1505#if IS_ENABLED(CONFIG_IPV6)
1506static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1507 struct neighbour *n, bool isrouter)
1508{
1509 struct net_device *dev = request->dev;
1510 struct sk_buff *reply;
1511 struct nd_msg *ns, *na;
1512 struct ipv6hdr *pip6;
1513 u8 *daddr;
1514 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1515 int ns_olen;
1516 int i, len;
1517
1518 if (dev == NULL)
1519 return NULL;
1520
1521 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1522 sizeof(*na) + na_olen + dev->needed_tailroom;
1523 reply = alloc_skb(len, GFP_ATOMIC);
1524 if (reply == NULL)
1525 return NULL;
1526
1527 reply->protocol = htons(ETH_P_IPV6);
1528 reply->dev = dev;
1529 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1530 skb_push(reply, sizeof(struct ethhdr));
1531 skb_reset_mac_header(reply);
1532
1533 ns = (struct nd_msg *)skb_transport_header(request);
1534
1535 daddr = eth_hdr(request)->h_source;
1536 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1537 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1538 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1539 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1540 break;
1541 }
1542 }
1543
1544 /* Ethernet header */
1545 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1546 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1547 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1548 reply->protocol = htons(ETH_P_IPV6);
1549
1550 skb_pull(reply, sizeof(struct ethhdr));
1551 skb_reset_network_header(reply);
1552 skb_put(reply, sizeof(struct ipv6hdr));
1553
1554 /* IPv6 header */
1555
1556 pip6 = ipv6_hdr(reply);
1557 memset(pip6, 0, sizeof(struct ipv6hdr));
1558 pip6->version = 6;
1559 pip6->priority = ipv6_hdr(request)->priority;
1560 pip6->nexthdr = IPPROTO_ICMPV6;
1561 pip6->hop_limit = 255;
1562 pip6->daddr = ipv6_hdr(request)->saddr;
1563 pip6->saddr = *(struct in6_addr *)n->primary_key;
1564
1565 skb_pull(reply, sizeof(struct ipv6hdr));
1566 skb_reset_transport_header(reply);
1567
1568 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1569
1570 /* Neighbor Advertisement */
1571 memset(na, 0, sizeof(*na)+na_olen);
1572 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1573 na->icmph.icmp6_router = isrouter;
1574 na->icmph.icmp6_override = 1;
1575 na->icmph.icmp6_solicited = 1;
1576 na->target = ns->target;
1577 ether_addr_copy(&na->opt[2], n->ha);
1578 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1579 na->opt[1] = na_olen >> 3;
1580
1581 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1582 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1583 csum_partial(na, sizeof(*na)+na_olen, 0));
1584
1585 pip6->payload_len = htons(sizeof(*na)+na_olen);
1586
1587 skb_push(reply, sizeof(struct ipv6hdr));
1588
1589 reply->ip_summed = CHECKSUM_UNNECESSARY;
1590
1591 return reply;
1592}
1593
1594static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1595{
1596 struct vxlan_dev *vxlan = netdev_priv(dev);
1597 struct nd_msg *msg;
1598 const struct ipv6hdr *iphdr;
1599 const struct in6_addr *saddr, *daddr;
1600 struct neighbour *n;
1601 struct inet6_dev *in6_dev;
1602
1603 in6_dev = __in6_dev_get(dev);
1604 if (!in6_dev)
1605 goto out;
1606
1607 iphdr = ipv6_hdr(skb);
1608 saddr = &iphdr->saddr;
1609 daddr = &iphdr->daddr;
1610
1611 msg = (struct nd_msg *)skb_transport_header(skb);
1612 if (msg->icmph.icmp6_code != 0 ||
1613 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1614 goto out;
1615
1616 if (ipv6_addr_loopback(daddr) ||
1617 ipv6_addr_is_multicast(&msg->target))
1618 goto out;
1619
1620 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1621
1622 if (n) {
1623 struct vxlan_fdb *f;
1624 struct sk_buff *reply;
1625
1626 if (!(n->nud_state & NUD_CONNECTED)) {
1627 neigh_release(n);
1628 goto out;
1629 }
1630
1631 f = vxlan_find_mac(vxlan, n->ha, vni);
1632 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1633 /* bridge-local neighbor */
1634 neigh_release(n);
1635 goto out;
1636 }
1637
1638 reply = vxlan_na_create(skb, n,
1639 !!(f ? f->flags & NTF_ROUTER : 0));
1640
1641 neigh_release(n);
1642
1643 if (reply == NULL)
1644 goto out;
1645
1646 if (netif_rx_ni(reply) == NET_RX_DROP)
1647 dev->stats.rx_dropped++;
1648
1649 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1650 union vxlan_addr ipa = {
1651 .sin6.sin6_addr = msg->target,
1652 .sin6.sin6_family = AF_INET6,
1653 };
1654
1655 vxlan_ip_miss(dev, &ipa);
1656 }
1657
1658out:
1659 consume_skb(skb);
1660 return NETDEV_TX_OK;
1661}
1662#endif
1663
1664static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1665{
1666 struct vxlan_dev *vxlan = netdev_priv(dev);
1667 struct neighbour *n;
1668
1669 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1670 return false;
1671
1672 n = NULL;
1673 switch (ntohs(eth_hdr(skb)->h_proto)) {
1674 case ETH_P_IP:
1675 {
1676 struct iphdr *pip;
1677
1678 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1679 return false;
1680 pip = ip_hdr(skb);
1681 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1682 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1683 union vxlan_addr ipa = {
1684 .sin.sin_addr.s_addr = pip->daddr,
1685 .sin.sin_family = AF_INET,
1686 };
1687
1688 vxlan_ip_miss(dev, &ipa);
1689 return false;
1690 }
1691
1692 break;
1693 }
1694#if IS_ENABLED(CONFIG_IPV6)
1695 case ETH_P_IPV6:
1696 {
1697 struct ipv6hdr *pip6;
1698
1699 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1700 return false;
1701 pip6 = ipv6_hdr(skb);
1702 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1703 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1704 union vxlan_addr ipa = {
1705 .sin6.sin6_addr = pip6->daddr,
1706 .sin6.sin6_family = AF_INET6,
1707 };
1708
1709 vxlan_ip_miss(dev, &ipa);
1710 return false;
1711 }
1712
1713 break;
1714 }
1715#endif
1716 default:
1717 return false;
1718 }
1719
1720 if (n) {
1721 bool diff;
1722
1723 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1724 if (diff) {
1725 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1726 dev->addr_len);
1727 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1728 }
1729 neigh_release(n);
1730 return diff;
1731 }
1732
1733 return false;
1734}
1735
1736static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1737 struct vxlan_metadata *md)
1738{
1739 struct vxlanhdr_gbp *gbp;
1740
1741 if (!md->gbp)
1742 return;
1743
1744 gbp = (struct vxlanhdr_gbp *)vxh;
1745 vxh->vx_flags |= VXLAN_HF_GBP;
1746
1747 if (md->gbp & VXLAN_GBP_DONT_LEARN)
1748 gbp->dont_learn = 1;
1749
1750 if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1751 gbp->policy_applied = 1;
1752
1753 gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1754}
1755
1756static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1757 __be16 protocol)
1758{
1759 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1760
1761 gpe->np_applied = 1;
1762
1763 switch (protocol) {
1764 case htons(ETH_P_IP):
1765 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1766 return 0;
1767 case htons(ETH_P_IPV6):
1768 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1769 return 0;
1770 case htons(ETH_P_TEB):
1771 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1772 return 0;
1773 }
1774 return -EPFNOSUPPORT;
1775}
1776
1777static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1778 int iphdr_len, __be32 vni,
1779 struct vxlan_metadata *md, u32 vxflags,
1780 bool udp_sum)
1781{
1782 struct vxlanhdr *vxh;
1783 int min_headroom;
1784 int err;
1785 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1786 __be16 inner_protocol = htons(ETH_P_TEB);
1787
1788 if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1789 skb->ip_summed == CHECKSUM_PARTIAL) {
1790 int csum_start = skb_checksum_start_offset(skb);
1791
1792 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1793 !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1794 (skb->csum_offset == offsetof(struct udphdr, check) ||
1795 skb->csum_offset == offsetof(struct tcphdr, check)))
1796 type |= SKB_GSO_TUNNEL_REMCSUM;
1797 }
1798
1799 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1800 + VXLAN_HLEN + iphdr_len;
1801
1802 /* Need space for new headers (invalidates iph ptr) */
1803 err = skb_cow_head(skb, min_headroom);
1804 if (unlikely(err))
1805 return err;
1806
1807 err = iptunnel_handle_offloads(skb, type);
1808 if (err)
1809 return err;
1810
1811 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1812 vxh->vx_flags = VXLAN_HF_VNI;
1813 vxh->vx_vni = vxlan_vni_field(vni);
1814
1815 if (type & SKB_GSO_TUNNEL_REMCSUM) {
1816 unsigned int start;
1817
1818 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1819 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1820 vxh->vx_flags |= VXLAN_HF_RCO;
1821
1822 if (!skb_is_gso(skb)) {
1823 skb->ip_summed = CHECKSUM_NONE;
1824 skb->encapsulation = 0;
1825 }
1826 }
1827
1828 if (vxflags & VXLAN_F_GBP)
1829 vxlan_build_gbp_hdr(vxh, vxflags, md);
1830 if (vxflags & VXLAN_F_GPE) {
1831 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1832 if (err < 0)
1833 return err;
1834 inner_protocol = skb->protocol;
1835 }
1836
1837 skb_set_inner_protocol(skb, inner_protocol);
1838 return 0;
1839}
1840
1841static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1842 struct vxlan_sock *sock4,
1843 struct sk_buff *skb, int oif, u8 tos,
1844 __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
1845 struct dst_cache *dst_cache,
1846 const struct ip_tunnel_info *info)
1847{
1848 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1849 struct rtable *rt = NULL;
1850 struct flowi4 fl4;
1851
1852 if (!sock4)
1853 return ERR_PTR(-EIO);
1854
1855 if (tos && !info)
1856 use_cache = false;
1857 if (use_cache) {
1858 rt = dst_cache_get_ip4(dst_cache, saddr);
1859 if (rt)
1860 return rt;
1861 }
1862
1863 memset(&fl4, 0, sizeof(fl4));
1864 fl4.flowi4_oif = oif;
1865 fl4.flowi4_tos = RT_TOS(tos);
1866 fl4.flowi4_mark = skb->mark;
1867 fl4.flowi4_proto = IPPROTO_UDP;
1868 fl4.daddr = daddr;
1869 fl4.saddr = *saddr;
1870 fl4.fl4_dport = dport;
1871 fl4.fl4_sport = sport;
1872
1873 rt = ip_route_output_key(vxlan->net, &fl4);
1874 if (likely(!IS_ERR(rt))) {
1875 if (rt->dst.dev == dev) {
1876 netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1877 ip_rt_put(rt);
1878 return ERR_PTR(-ELOOP);
1879 }
1880
1881 *saddr = fl4.saddr;
1882 if (use_cache)
1883 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1884 } else {
1885 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1886 return ERR_PTR(-ENETUNREACH);
1887 }
1888 return rt;
1889}
1890
1891#if IS_ENABLED(CONFIG_IPV6)
1892static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1893 struct net_device *dev,
1894 struct vxlan_sock *sock6,
1895 struct sk_buff *skb, int oif, u8 tos,
1896 __be32 label,
1897 const struct in6_addr *daddr,
1898 struct in6_addr *saddr,
1899 __be16 dport, __be16 sport,
1900 struct dst_cache *dst_cache,
1901 const struct ip_tunnel_info *info)
1902{
1903 bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1904 struct dst_entry *ndst;
1905 struct flowi6 fl6;
1906 int err;
1907
1908 if (!sock6)
1909 return ERR_PTR(-EIO);
1910
1911 if (tos && !info)
1912 use_cache = false;
1913 if (use_cache) {
1914 ndst = dst_cache_get_ip6(dst_cache, saddr);
1915 if (ndst)
1916 return ndst;
1917 }
1918
1919 memset(&fl6, 0, sizeof(fl6));
1920 fl6.flowi6_oif = oif;
1921 fl6.daddr = *daddr;
1922 fl6.saddr = *saddr;
1923 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1924 fl6.flowi6_mark = skb->mark;
1925 fl6.flowi6_proto = IPPROTO_UDP;
1926 fl6.fl6_dport = dport;
1927 fl6.fl6_sport = sport;
1928
1929 err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1930 sock6->sock->sk,
1931 &ndst, &fl6);
1932 if (unlikely(err < 0)) {
1933 netdev_dbg(dev, "no route to %pI6\n", daddr);
1934 return ERR_PTR(-ENETUNREACH);
1935 }
1936
1937 if (unlikely(ndst->dev == dev)) {
1938 netdev_dbg(dev, "circular route to %pI6\n", daddr);
1939 dst_release(ndst);
1940 return ERR_PTR(-ELOOP);
1941 }
1942
1943 *saddr = fl6.saddr;
1944 if (use_cache)
1945 dst_cache_set_ip6(dst_cache, ndst, saddr);
1946 return ndst;
1947}
1948#endif
1949
1950/* Bypass encapsulation if the destination is local */
1951static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1952 struct vxlan_dev *dst_vxlan, __be32 vni)
1953{
1954 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1955 union vxlan_addr loopback;
1956 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1957 struct net_device *dev = skb->dev;
1958 int len = skb->len;
1959
1960 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1961 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1962 skb->pkt_type = PACKET_HOST;
1963 skb->encapsulation = 0;
1964 skb->dev = dst_vxlan->dev;
1965 __skb_pull(skb, skb_network_offset(skb));
1966
1967 if (remote_ip->sa.sa_family == AF_INET) {
1968 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1969 loopback.sa.sa_family = AF_INET;
1970#if IS_ENABLED(CONFIG_IPV6)
1971 } else {
1972 loopback.sin6.sin6_addr = in6addr_loopback;
1973 loopback.sa.sa_family = AF_INET6;
1974#endif
1975 }
1976
1977 if (dst_vxlan->flags & VXLAN_F_LEARN)
1978 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source, vni);
1979
1980 u64_stats_update_begin(&tx_stats->syncp);
1981 tx_stats->tx_packets++;
1982 tx_stats->tx_bytes += len;
1983 u64_stats_update_end(&tx_stats->syncp);
1984
1985 if (netif_rx(skb) == NET_RX_SUCCESS) {
1986 u64_stats_update_begin(&rx_stats->syncp);
1987 rx_stats->rx_packets++;
1988 rx_stats->rx_bytes += len;
1989 u64_stats_update_end(&rx_stats->syncp);
1990 } else {
1991 dev->stats.rx_dropped++;
1992 }
1993}
1994
1995static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
1996 struct vxlan_dev *vxlan, union vxlan_addr *daddr,
1997 __be16 dst_port, __be32 vni, struct dst_entry *dst,
1998 u32 rt_flags)
1999{
2000#if IS_ENABLED(CONFIG_IPV6)
2001 /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2002 * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2003 * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2004 */
2005 BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2006#endif
2007 /* Bypass encapsulation if the destination is local */
2008 if (rt_flags & RTCF_LOCAL &&
2009 !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2010 struct vxlan_dev *dst_vxlan;
2011
2012 dst_release(dst);
2013 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2014 daddr->sa.sa_family, dst_port,
2015 vxlan->flags);
2016 if (!dst_vxlan) {
2017 dev->stats.tx_errors++;
2018 kfree_skb(skb);
2019
2020 return -ENOENT;
2021 }
2022 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2023 return 1;
2024 }
2025
2026 return 0;
2027}
2028
2029static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2030 __be32 default_vni, struct vxlan_rdst *rdst,
2031 bool did_rsc)
2032{
2033 struct dst_cache *dst_cache;
2034 struct ip_tunnel_info *info;
2035 struct vxlan_dev *vxlan = netdev_priv(dev);
2036 const struct iphdr *old_iph = ip_hdr(skb);
2037 union vxlan_addr *dst;
2038 union vxlan_addr remote_ip, local_ip;
2039 union vxlan_addr *src;
2040 struct vxlan_metadata _md;
2041 struct vxlan_metadata *md = &_md;
2042 __be16 src_port = 0, dst_port;
2043 struct dst_entry *ndst = NULL;
2044 __be32 vni, label;
2045 __u8 tos, ttl;
2046 int err;
2047 u32 flags = vxlan->flags;
2048 bool udp_sum = false;
2049 bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2050
2051 info = skb_tunnel_info(skb);
2052
2053 if (rdst) {
2054 dst = &rdst->remote_ip;
2055 if (vxlan_addr_any(dst)) {
2056 if (did_rsc) {
2057 /* short-circuited back to local bridge */
2058 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2059 return;
2060 }
2061 goto drop;
2062 }
2063
2064 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2065 vni = (rdst->remote_vni) ? : default_vni;
2066 src = &vxlan->cfg.saddr;
2067 dst_cache = &rdst->dst_cache;
2068 md->gbp = skb->mark;
2069 ttl = vxlan->cfg.ttl;
2070 if (!ttl && vxlan_addr_multicast(dst))
2071 ttl = 1;
2072
2073 tos = vxlan->cfg.tos;
2074 if (tos == 1)
2075 tos = ip_tunnel_get_dsfield(old_iph, skb);
2076
2077 if (dst->sa.sa_family == AF_INET)
2078 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2079 else
2080 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2081 label = vxlan->cfg.label;
2082 } else {
2083 if (!info) {
2084 WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2085 dev->name);
2086 goto drop;
2087 }
2088 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2089 if (remote_ip.sa.sa_family == AF_INET) {
2090 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2091 local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2092 } else {
2093 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2094 local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2095 }
2096 dst = &remote_ip;
2097 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2098 vni = tunnel_id_to_key32(info->key.tun_id);
2099 src = &local_ip;
2100 dst_cache = &info->dst_cache;
2101 if (info->options_len)
2102 md = ip_tunnel_info_opts(info);
2103 ttl = info->key.ttl;
2104 tos = info->key.tos;
2105 label = info->key.label;
2106 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2107 }
2108 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2109 vxlan->cfg.port_max, true);
2110
2111 if (dst->sa.sa_family == AF_INET) {
2112 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2113 struct rtable *rt;
2114 __be16 df = 0;
2115
2116 rt = vxlan_get_route(vxlan, dev, sock4, skb,
2117 rdst ? rdst->remote_ifindex : 0, tos,
2118 dst->sin.sin_addr.s_addr,
2119 &src->sin.sin_addr.s_addr,
2120 dst_port, src_port,
2121 dst_cache, info);
2122 if (IS_ERR(rt)) {
2123 err = PTR_ERR(rt);
2124 goto tx_error;
2125 }
2126
2127 /* Bypass encapsulation if the destination is local */
2128 if (!info) {
2129 err = encap_bypass_if_local(skb, dev, vxlan, dst,
2130 dst_port, vni, &rt->dst,
2131 rt->rt_flags);
2132 if (err)
2133 return;
2134 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2135 df = htons(IP_DF);
2136 }
2137
2138 ndst = &rt->dst;
2139 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2140 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2141 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2142 vni, md, flags, udp_sum);
2143 if (err < 0)
2144 goto tx_error;
2145
2146 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, src->sin.sin_addr.s_addr,
2147 dst->sin.sin_addr.s_addr, tos, ttl, df,
2148 src_port, dst_port, xnet, !udp_sum);
2149#if IS_ENABLED(CONFIG_IPV6)
2150 } else {
2151 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2152
2153 ndst = vxlan6_get_route(vxlan, dev, sock6, skb,
2154 rdst ? rdst->remote_ifindex : 0, tos,
2155 label, &dst->sin6.sin6_addr,
2156 &src->sin6.sin6_addr,
2157 dst_port, src_port,
2158 dst_cache, info);
2159 if (IS_ERR(ndst)) {
2160 err = PTR_ERR(ndst);
2161 ndst = NULL;
2162 goto tx_error;
2163 }
2164
2165 if (!info) {
2166 u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2167
2168 err = encap_bypass_if_local(skb, dev, vxlan, dst,
2169 dst_port, vni, ndst,
2170 rt6i_flags);
2171 if (err)
2172 return;
2173 }
2174
2175 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2176 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2177 skb_scrub_packet(skb, xnet);
2178 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2179 vni, md, flags, udp_sum);
2180 if (err < 0)
2181 goto tx_error;
2182
2183 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2184 &src->sin6.sin6_addr,
2185 &dst->sin6.sin6_addr, tos, ttl,
2186 label, src_port, dst_port, !udp_sum);
2187#endif
2188 }
2189 return;
2190
2191drop:
2192 dev->stats.tx_dropped++;
2193 dev_kfree_skb(skb);
2194 return;
2195
2196tx_error:
2197 if (err == -ELOOP)
2198 dev->stats.collisions++;
2199 else if (err == -ENETUNREACH)
2200 dev->stats.tx_carrier_errors++;
2201 dst_release(ndst);
2202 dev->stats.tx_errors++;
2203 kfree_skb(skb);
2204}
2205
2206/* Transmit local packets over Vxlan
2207 *
2208 * Outer IP header inherits ECN and DF from inner header.
2209 * Outer UDP destination is the VXLAN assigned port.
2210 * source port is based on hash of flow
2211 */
2212static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2213{
2214 struct vxlan_dev *vxlan = netdev_priv(dev);
2215 const struct ip_tunnel_info *info;
2216 struct ethhdr *eth;
2217 bool did_rsc = false;
2218 struct vxlan_rdst *rdst, *fdst = NULL;
2219 struct vxlan_fdb *f;
2220 __be32 vni = 0;
2221
2222 info = skb_tunnel_info(skb);
2223
2224 skb_reset_mac_header(skb);
2225
2226 if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2227 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2228 info->mode & IP_TUNNEL_INFO_TX) {
2229 vni = tunnel_id_to_key32(info->key.tun_id);
2230 } else {
2231 if (info && info->mode & IP_TUNNEL_INFO_TX)
2232 vxlan_xmit_one(skb, dev, vni, NULL, false);
2233 else
2234 kfree_skb(skb);
2235 return NETDEV_TX_OK;
2236 }
2237 }
2238
2239 if (vxlan->flags & VXLAN_F_PROXY) {
2240 eth = eth_hdr(skb);
2241 if (ntohs(eth->h_proto) == ETH_P_ARP)
2242 return arp_reduce(dev, skb, vni);
2243#if IS_ENABLED(CONFIG_IPV6)
2244 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2245 pskb_may_pull(skb, sizeof(struct ipv6hdr)
2246 + sizeof(struct nd_msg)) &&
2247 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2248 struct nd_msg *msg;
2249
2250 msg = (struct nd_msg *)skb_transport_header(skb);
2251 if (msg->icmph.icmp6_code == 0 &&
2252 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2253 return neigh_reduce(dev, skb, vni);
2254 }
2255#endif
2256 }
2257
2258 eth = eth_hdr(skb);
2259 f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2260 did_rsc = false;
2261
2262 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2263 (ntohs(eth->h_proto) == ETH_P_IP ||
2264 ntohs(eth->h_proto) == ETH_P_IPV6)) {
2265 did_rsc = route_shortcircuit(dev, skb);
2266 if (did_rsc)
2267 f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2268 }
2269
2270 if (f == NULL) {
2271 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2272 if (f == NULL) {
2273 if ((vxlan->flags & VXLAN_F_L2MISS) &&
2274 !is_multicast_ether_addr(eth->h_dest))
2275 vxlan_fdb_miss(vxlan, eth->h_dest);
2276
2277 dev->stats.tx_dropped++;
2278 kfree_skb(skb);
2279 return NETDEV_TX_OK;
2280 }
2281 }
2282
2283 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2284 struct sk_buff *skb1;
2285
2286 if (!fdst) {
2287 fdst = rdst;
2288 continue;
2289 }
2290 skb1 = skb_clone(skb, GFP_ATOMIC);
2291 if (skb1)
2292 vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2293 }
2294
2295 if (fdst)
2296 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2297 else
2298 kfree_skb(skb);
2299 return NETDEV_TX_OK;
2300}
2301
2302/* Walk the forwarding table and purge stale entries */
2303static void vxlan_cleanup(unsigned long arg)
2304{
2305 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2306 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2307 unsigned int h;
2308
2309 if (!netif_running(vxlan->dev))
2310 return;
2311
2312 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2313 struct hlist_node *p, *n;
2314
2315 spin_lock_bh(&vxlan->hash_lock);
2316 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2317 struct vxlan_fdb *f
2318 = container_of(p, struct vxlan_fdb, hlist);
2319 unsigned long timeout;
2320
2321 if (f->state & (NUD_PERMANENT | NUD_NOARP))
2322 continue;
2323
2324 timeout = f->used + vxlan->cfg.age_interval * HZ;
2325 if (time_before_eq(timeout, jiffies)) {
2326 netdev_dbg(vxlan->dev,
2327 "garbage collect %pM\n",
2328 f->eth_addr);
2329 f->state = NUD_STALE;
2330 vxlan_fdb_destroy(vxlan, f);
2331 } else if (time_before(timeout, next_timer))
2332 next_timer = timeout;
2333 }
2334 spin_unlock_bh(&vxlan->hash_lock);
2335 }
2336
2337 mod_timer(&vxlan->age_timer, next_timer);
2338}
2339
2340static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2341{
2342 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2343 __be32 vni = vxlan->default_dst.remote_vni;
2344
2345 spin_lock(&vn->sock_lock);
2346 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2347 spin_unlock(&vn->sock_lock);
2348}
2349
2350/* Setup stats when device is created */
2351static int vxlan_init(struct net_device *dev)
2352{
2353 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2354 if (!dev->tstats)
2355 return -ENOMEM;
2356
2357 return 0;
2358}
2359
2360static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2361{
2362 struct vxlan_fdb *f;
2363
2364 spin_lock_bh(&vxlan->hash_lock);
2365 f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2366 if (f)
2367 vxlan_fdb_destroy(vxlan, f);
2368 spin_unlock_bh(&vxlan->hash_lock);
2369}
2370
2371static void vxlan_uninit(struct net_device *dev)
2372{
2373 struct vxlan_dev *vxlan = netdev_priv(dev);
2374
2375 vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2376
2377 free_percpu(dev->tstats);
2378}
2379
2380/* Start ageing timer and join group when device is brought up */
2381static int vxlan_open(struct net_device *dev)
2382{
2383 struct vxlan_dev *vxlan = netdev_priv(dev);
2384 int ret;
2385
2386 ret = vxlan_sock_add(vxlan);
2387 if (ret < 0)
2388 return ret;
2389
2390 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2391 ret = vxlan_igmp_join(vxlan);
2392 if (ret == -EADDRINUSE)
2393 ret = 0;
2394 if (ret) {
2395 vxlan_sock_release(vxlan);
2396 return ret;
2397 }
2398 }
2399
2400 if (vxlan->cfg.age_interval)
2401 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2402
2403 return ret;
2404}
2405
2406/* Purge the forwarding table */
2407static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2408{
2409 unsigned int h;
2410
2411 spin_lock_bh(&vxlan->hash_lock);
2412 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2413 struct hlist_node *p, *n;
2414 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2415 struct vxlan_fdb *f
2416 = container_of(p, struct vxlan_fdb, hlist);
2417 if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2418 continue;
2419 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2420 if (!is_zero_ether_addr(f->eth_addr))
2421 vxlan_fdb_destroy(vxlan, f);
2422 }
2423 }
2424 spin_unlock_bh(&vxlan->hash_lock);
2425}
2426
2427/* Cleanup timer and forwarding table on shutdown */
2428static int vxlan_stop(struct net_device *dev)
2429{
2430 struct vxlan_dev *vxlan = netdev_priv(dev);
2431 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2432 int ret = 0;
2433
2434 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2435 !vxlan_group_used(vn, vxlan))
2436 ret = vxlan_igmp_leave(vxlan);
2437
2438 del_timer_sync(&vxlan->age_timer);
2439
2440 vxlan_flush(vxlan, false);
2441 vxlan_sock_release(vxlan);
2442
2443 return ret;
2444}
2445
2446/* Stub, nothing needs to be done. */
2447static void vxlan_set_multicast_list(struct net_device *dev)
2448{
2449}
2450
2451static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2452{
2453 struct vxlan_dev *vxlan = netdev_priv(dev);
2454 struct vxlan_rdst *dst = &vxlan->default_dst;
2455 struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2456 dst->remote_ifindex);
2457 bool use_ipv6 = false;
2458
2459 if (dst->remote_ip.sa.sa_family == AF_INET6)
2460 use_ipv6 = true;
2461
2462 /* This check is different than dev->max_mtu, because it looks at
2463 * the lowerdev->mtu, rather than the static dev->max_mtu
2464 */
2465 if (lowerdev) {
2466 int max_mtu = lowerdev->mtu -
2467 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2468 if (new_mtu > max_mtu)
2469 return -EINVAL;
2470 }
2471
2472 dev->mtu = new_mtu;
2473 return 0;
2474}
2475
2476static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2477{
2478 struct vxlan_dev *vxlan = netdev_priv(dev);
2479 struct ip_tunnel_info *info = skb_tunnel_info(skb);
2480 __be16 sport, dport;
2481
2482 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2483 vxlan->cfg.port_max, true);
2484 dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2485
2486 if (ip_tunnel_info_af(info) == AF_INET) {
2487 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2488 struct rtable *rt;
2489
2490 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2491 info->key.u.ipv4.dst,
2492 &info->key.u.ipv4.src, dport, sport,
2493 &info->dst_cache, info);
2494 if (IS_ERR(rt))
2495 return PTR_ERR(rt);
2496 ip_rt_put(rt);
2497 } else {
2498#if IS_ENABLED(CONFIG_IPV6)
2499 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2500 struct dst_entry *ndst;
2501
2502 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2503 info->key.label, &info->key.u.ipv6.dst,
2504 &info->key.u.ipv6.src, dport, sport,
2505 &info->dst_cache, info);
2506 if (IS_ERR(ndst))
2507 return PTR_ERR(ndst);
2508 dst_release(ndst);
2509#else /* !CONFIG_IPV6 */
2510 return -EPFNOSUPPORT;
2511#endif
2512 }
2513 info->key.tp_src = sport;
2514 info->key.tp_dst = dport;
2515 return 0;
2516}
2517
2518static const struct net_device_ops vxlan_netdev_ether_ops = {
2519 .ndo_init = vxlan_init,
2520 .ndo_uninit = vxlan_uninit,
2521 .ndo_open = vxlan_open,
2522 .ndo_stop = vxlan_stop,
2523 .ndo_start_xmit = vxlan_xmit,
2524 .ndo_get_stats64 = ip_tunnel_get_stats64,
2525 .ndo_set_rx_mode = vxlan_set_multicast_list,
2526 .ndo_change_mtu = vxlan_change_mtu,
2527 .ndo_validate_addr = eth_validate_addr,
2528 .ndo_set_mac_address = eth_mac_addr,
2529 .ndo_fdb_add = vxlan_fdb_add,
2530 .ndo_fdb_del = vxlan_fdb_delete,
2531 .ndo_fdb_dump = vxlan_fdb_dump,
2532 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2533};
2534
2535static const struct net_device_ops vxlan_netdev_raw_ops = {
2536 .ndo_init = vxlan_init,
2537 .ndo_uninit = vxlan_uninit,
2538 .ndo_open = vxlan_open,
2539 .ndo_stop = vxlan_stop,
2540 .ndo_start_xmit = vxlan_xmit,
2541 .ndo_get_stats64 = ip_tunnel_get_stats64,
2542 .ndo_change_mtu = vxlan_change_mtu,
2543 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst,
2544};
2545
2546/* Info for udev, that this is a virtual tunnel endpoint */
2547static struct device_type vxlan_type = {
2548 .name = "vxlan",
2549};
2550
2551/* Calls the ndo_udp_tunnel_add of the caller in order to
2552 * supply the listening VXLAN udp ports. Callers are expected
2553 * to implement the ndo_udp_tunnel_add.
2554 */
2555static void vxlan_push_rx_ports(struct net_device *dev)
2556{
2557 struct vxlan_sock *vs;
2558 struct net *net = dev_net(dev);
2559 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2560 unsigned int i;
2561
2562 spin_lock(&vn->sock_lock);
2563 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2564 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2565 udp_tunnel_push_rx_port(dev, vs->sock,
2566 (vs->flags & VXLAN_F_GPE) ?
2567 UDP_TUNNEL_TYPE_VXLAN_GPE :
2568 UDP_TUNNEL_TYPE_VXLAN);
2569 }
2570 spin_unlock(&vn->sock_lock);
2571}
2572
2573/* Initialize the device structure. */
2574static void vxlan_setup(struct net_device *dev)
2575{
2576 struct vxlan_dev *vxlan = netdev_priv(dev);
2577 unsigned int h;
2578
2579 eth_hw_addr_random(dev);
2580 ether_setup(dev);
2581
2582 dev->destructor = free_netdev;
2583 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2584
2585 dev->features |= NETIF_F_LLTX;
2586 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2587 dev->features |= NETIF_F_RXCSUM;
2588 dev->features |= NETIF_F_GSO_SOFTWARE;
2589
2590 dev->vlan_features = dev->features;
2591 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2592 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2593 netif_keep_dst(dev);
2594 dev->priv_flags |= IFF_NO_QUEUE;
2595
2596 INIT_LIST_HEAD(&vxlan->next);
2597 spin_lock_init(&vxlan->hash_lock);
2598
2599 init_timer_deferrable(&vxlan->age_timer);
2600 vxlan->age_timer.function = vxlan_cleanup;
2601 vxlan->age_timer.data = (unsigned long) vxlan;
2602
2603 vxlan->cfg.dst_port = htons(vxlan_port);
2604
2605 vxlan->dev = dev;
2606
2607 gro_cells_init(&vxlan->gro_cells, dev);
2608
2609 for (h = 0; h < FDB_HASH_SIZE; ++h)
2610 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2611}
2612
2613static void vxlan_ether_setup(struct net_device *dev)
2614{
2615 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2616 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2617 dev->netdev_ops = &vxlan_netdev_ether_ops;
2618}
2619
2620static void vxlan_raw_setup(struct net_device *dev)
2621{
2622 dev->header_ops = NULL;
2623 dev->type = ARPHRD_NONE;
2624 dev->hard_header_len = 0;
2625 dev->addr_len = 0;
2626 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2627 dev->netdev_ops = &vxlan_netdev_raw_ops;
2628}
2629
2630static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2631 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2632 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2633 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2634 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2635 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2636 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2637 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2638 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2639 [IFLA_VXLAN_LABEL] = { .type = NLA_U32 },
2640 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2641 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2642 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2643 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2644 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2645 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2646 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2647 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2648 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 },
2649 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2650 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 },
2651 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
2652 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
2653 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2654 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2655 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, },
2656 [IFLA_VXLAN_GPE] = { .type = NLA_FLAG, },
2657 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG },
2658};
2659
2660static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2661{
2662 if (tb[IFLA_ADDRESS]) {
2663 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2664 pr_debug("invalid link address (not ethernet)\n");
2665 return -EINVAL;
2666 }
2667
2668 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2669 pr_debug("invalid all zero ethernet address\n");
2670 return -EADDRNOTAVAIL;
2671 }
2672 }
2673
2674 if (!data)
2675 return -EINVAL;
2676
2677 if (data[IFLA_VXLAN_ID]) {
2678 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2679 if (id >= VXLAN_VID_MASK)
2680 return -ERANGE;
2681 }
2682
2683 if (data[IFLA_VXLAN_PORT_RANGE]) {
2684 const struct ifla_vxlan_port_range *p
2685 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2686
2687 if (ntohs(p->high) < ntohs(p->low)) {
2688 pr_debug("port range %u .. %u not valid\n",
2689 ntohs(p->low), ntohs(p->high));
2690 return -EINVAL;
2691 }
2692 }
2693
2694 return 0;
2695}
2696
2697static void vxlan_get_drvinfo(struct net_device *netdev,
2698 struct ethtool_drvinfo *drvinfo)
2699{
2700 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2701 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2702}
2703
2704static const struct ethtool_ops vxlan_ethtool_ops = {
2705 .get_drvinfo = vxlan_get_drvinfo,
2706 .get_link = ethtool_op_get_link,
2707};
2708
2709static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2710 __be16 port, u32 flags)
2711{
2712 struct socket *sock;
2713 struct udp_port_cfg udp_conf;
2714 int err;
2715
2716 memset(&udp_conf, 0, sizeof(udp_conf));
2717
2718 if (ipv6) {
2719 udp_conf.family = AF_INET6;
2720 udp_conf.use_udp6_rx_checksums =
2721 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2722 udp_conf.ipv6_v6only = 1;
2723 } else {
2724 udp_conf.family = AF_INET;
2725 }
2726
2727 udp_conf.local_udp_port = port;
2728
2729 /* Open UDP socket */
2730 err = udp_sock_create(net, &udp_conf, &sock);
2731 if (err < 0)
2732 return ERR_PTR(err);
2733
2734 return sock;
2735}
2736
2737/* Create new listen socket if needed */
2738static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2739 __be16 port, u32 flags)
2740{
2741 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2742 struct vxlan_sock *vs;
2743 struct socket *sock;
2744 unsigned int h;
2745 struct udp_tunnel_sock_cfg tunnel_cfg;
2746
2747 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2748 if (!vs)
2749 return ERR_PTR(-ENOMEM);
2750
2751 for (h = 0; h < VNI_HASH_SIZE; ++h)
2752 INIT_HLIST_HEAD(&vs->vni_list[h]);
2753
2754 sock = vxlan_create_sock(net, ipv6, port, flags);
2755 if (IS_ERR(sock)) {
2756 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2757 PTR_ERR(sock));
2758 kfree(vs);
2759 return ERR_CAST(sock);
2760 }
2761
2762 vs->sock = sock;
2763 atomic_set(&vs->refcnt, 1);
2764 vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2765
2766 spin_lock(&vn->sock_lock);
2767 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2768 udp_tunnel_notify_add_rx_port(sock,
2769 (vs->flags & VXLAN_F_GPE) ?
2770 UDP_TUNNEL_TYPE_VXLAN_GPE :
2771 UDP_TUNNEL_TYPE_VXLAN);
2772 spin_unlock(&vn->sock_lock);
2773
2774 /* Mark socket as an encapsulation socket. */
2775 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2776 tunnel_cfg.sk_user_data = vs;
2777 tunnel_cfg.encap_type = 1;
2778 tunnel_cfg.encap_rcv = vxlan_rcv;
2779 tunnel_cfg.encap_destroy = NULL;
2780 tunnel_cfg.gro_receive = vxlan_gro_receive;
2781 tunnel_cfg.gro_complete = vxlan_gro_complete;
2782
2783 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2784
2785 return vs;
2786}
2787
2788static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2789{
2790 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2791 struct vxlan_sock *vs = NULL;
2792
2793 if (!vxlan->cfg.no_share) {
2794 spin_lock(&vn->sock_lock);
2795 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2796 vxlan->cfg.dst_port, vxlan->flags);
2797 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2798 spin_unlock(&vn->sock_lock);
2799 return -EBUSY;
2800 }
2801 spin_unlock(&vn->sock_lock);
2802 }
2803 if (!vs)
2804 vs = vxlan_socket_create(vxlan->net, ipv6,
2805 vxlan->cfg.dst_port, vxlan->flags);
2806 if (IS_ERR(vs))
2807 return PTR_ERR(vs);
2808#if IS_ENABLED(CONFIG_IPV6)
2809 if (ipv6)
2810 rcu_assign_pointer(vxlan->vn6_sock, vs);
2811 else
2812#endif
2813 rcu_assign_pointer(vxlan->vn4_sock, vs);
2814 vxlan_vs_add_dev(vs, vxlan);
2815 return 0;
2816}
2817
2818static int vxlan_sock_add(struct vxlan_dev *vxlan)
2819{
2820 bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2821 bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2822 int ret = 0;
2823
2824 RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2825#if IS_ENABLED(CONFIG_IPV6)
2826 RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2827 if (ipv6 || metadata)
2828 ret = __vxlan_sock_add(vxlan, true);
2829#endif
2830 if (!ret && (!ipv6 || metadata))
2831 ret = __vxlan_sock_add(vxlan, false);
2832 if (ret < 0)
2833 vxlan_sock_release(vxlan);
2834 return ret;
2835}
2836
2837static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2838 struct vxlan_config *conf)
2839{
2840 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2841 struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2842 struct vxlan_rdst *dst = &vxlan->default_dst;
2843 unsigned short needed_headroom = ETH_HLEN;
2844 int err;
2845 bool use_ipv6 = false;
2846 __be16 default_port = vxlan->cfg.dst_port;
2847 struct net_device *lowerdev = NULL;
2848
2849 if (conf->flags & VXLAN_F_GPE) {
2850 /* For now, allow GPE only together with COLLECT_METADATA.
2851 * This can be relaxed later; in such case, the other side
2852 * of the PtP link will have to be provided.
2853 */
2854 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2855 !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2856 pr_info("unsupported combination of extensions\n");
2857 return -EINVAL;
2858 }
2859
2860 vxlan_raw_setup(dev);
2861 } else {
2862 vxlan_ether_setup(dev);
2863 }
2864
2865 /* MTU range: 68 - 65535 */
2866 dev->min_mtu = ETH_MIN_MTU;
2867 dev->max_mtu = ETH_MAX_MTU;
2868
2869 vxlan->net = src_net;
2870
2871 dst->remote_vni = conf->vni;
2872
2873 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2874
2875 /* Unless IPv6 is explicitly requested, assume IPv4 */
2876 if (!dst->remote_ip.sa.sa_family)
2877 dst->remote_ip.sa.sa_family = AF_INET;
2878
2879 if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2880 vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2881 if (!IS_ENABLED(CONFIG_IPV6))
2882 return -EPFNOSUPPORT;
2883 use_ipv6 = true;
2884 vxlan->flags |= VXLAN_F_IPV6;
2885 }
2886
2887 if (conf->label && !use_ipv6) {
2888 pr_info("label only supported in use with IPv6\n");
2889 return -EINVAL;
2890 }
2891
2892 if (conf->remote_ifindex) {
2893 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2894 dst->remote_ifindex = conf->remote_ifindex;
2895
2896 if (!lowerdev) {
2897 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2898 return -ENODEV;
2899 }
2900
2901#if IS_ENABLED(CONFIG_IPV6)
2902 if (use_ipv6) {
2903 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2904 if (idev && idev->cnf.disable_ipv6) {
2905 pr_info("IPv6 is disabled via sysctl\n");
2906 return -EPERM;
2907 }
2908 }
2909#endif
2910
2911 if (!conf->mtu)
2912 dev->mtu = lowerdev->mtu -
2913 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2914
2915 needed_headroom = lowerdev->hard_header_len;
2916 } else if (vxlan_addr_multicast(&dst->remote_ip)) {
2917 pr_info("multicast destination requires interface to be specified\n");
2918 return -EINVAL;
2919 }
2920
2921 if (conf->mtu) {
2922 int max_mtu = ETH_MAX_MTU;
2923
2924 if (lowerdev)
2925 max_mtu = lowerdev->mtu;
2926
2927 max_mtu -= (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2928
2929 if (conf->mtu < dev->min_mtu || conf->mtu > dev->max_mtu)
2930 return -EINVAL;
2931
2932 dev->mtu = conf->mtu;
2933
2934 if (conf->mtu > max_mtu)
2935 dev->mtu = max_mtu;
2936 }
2937
2938 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2939 needed_headroom += VXLAN6_HEADROOM;
2940 else
2941 needed_headroom += VXLAN_HEADROOM;
2942 dev->needed_headroom = needed_headroom;
2943
2944 memcpy(&vxlan->cfg, conf, sizeof(*conf));
2945 if (!vxlan->cfg.dst_port) {
2946 if (conf->flags & VXLAN_F_GPE)
2947 vxlan->cfg.dst_port = htons(4790); /* IANA VXLAN-GPE port */
2948 else
2949 vxlan->cfg.dst_port = default_port;
2950 }
2951 vxlan->flags |= conf->flags;
2952
2953 if (!vxlan->cfg.age_interval)
2954 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2955
2956 list_for_each_entry(tmp, &vn->vxlan_list, next) {
2957 if (tmp->cfg.vni == conf->vni &&
2958 (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2959 tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2960 tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2961 (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2962 (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2963 pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2964 return -EEXIST;
2965 }
2966 }
2967
2968 dev->ethtool_ops = &vxlan_ethtool_ops;
2969
2970 /* create an fdb entry for a valid default destination */
2971 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2972 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2973 &vxlan->default_dst.remote_ip,
2974 NUD_REACHABLE|NUD_PERMANENT,
2975 NLM_F_EXCL|NLM_F_CREATE,
2976 vxlan->cfg.dst_port,
2977 vxlan->default_dst.remote_vni,
2978 vxlan->default_dst.remote_vni,
2979 vxlan->default_dst.remote_ifindex,
2980 NTF_SELF);
2981 if (err)
2982 return err;
2983 }
2984
2985 err = register_netdevice(dev);
2986 if (err) {
2987 vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2988 return err;
2989 }
2990
2991 list_add(&vxlan->next, &vn->vxlan_list);
2992
2993 return 0;
2994}
2995
2996static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2997 struct nlattr *tb[], struct nlattr *data[])
2998{
2999 struct vxlan_config conf;
3000
3001 memset(&conf, 0, sizeof(conf));
3002
3003 if (data[IFLA_VXLAN_ID])
3004 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3005
3006 if (data[IFLA_VXLAN_GROUP]) {
3007 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3008 } else if (data[IFLA_VXLAN_GROUP6]) {
3009 if (!IS_ENABLED(CONFIG_IPV6))
3010 return -EPFNOSUPPORT;
3011
3012 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3013 conf.remote_ip.sa.sa_family = AF_INET6;
3014 }
3015
3016 if (data[IFLA_VXLAN_LOCAL]) {
3017 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3018 conf.saddr.sa.sa_family = AF_INET;
3019 } else if (data[IFLA_VXLAN_LOCAL6]) {
3020 if (!IS_ENABLED(CONFIG_IPV6))
3021 return -EPFNOSUPPORT;
3022
3023 /* TODO: respect scope id */
3024 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3025 conf.saddr.sa.sa_family = AF_INET6;
3026 }
3027
3028 if (data[IFLA_VXLAN_LINK])
3029 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3030
3031 if (data[IFLA_VXLAN_TOS])
3032 conf.tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
3033
3034 if (data[IFLA_VXLAN_TTL])
3035 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3036
3037 if (data[IFLA_VXLAN_LABEL])
3038 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3039 IPV6_FLOWLABEL_MASK;
3040
3041 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
3042 conf.flags |= VXLAN_F_LEARN;
3043
3044 if (data[IFLA_VXLAN_AGEING])
3045 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3046
3047 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
3048 conf.flags |= VXLAN_F_PROXY;
3049
3050 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
3051 conf.flags |= VXLAN_F_RSC;
3052
3053 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
3054 conf.flags |= VXLAN_F_L2MISS;
3055
3056 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
3057 conf.flags |= VXLAN_F_L3MISS;
3058
3059 if (data[IFLA_VXLAN_LIMIT])
3060 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3061
3062 if (data[IFLA_VXLAN_COLLECT_METADATA] &&
3063 nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3064 conf.flags |= VXLAN_F_COLLECT_METADATA;
3065
3066 if (data[IFLA_VXLAN_PORT_RANGE]) {
3067 const struct ifla_vxlan_port_range *p
3068 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3069 conf.port_min = ntohs(p->low);
3070 conf.port_max = ntohs(p->high);
3071 }
3072
3073 if (data[IFLA_VXLAN_PORT])
3074 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3075
3076 if (data[IFLA_VXLAN_UDP_CSUM] &&
3077 !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3078 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3079
3080 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3081 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3082 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3083
3084 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3085 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3086 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3087
3088 if (data[IFLA_VXLAN_REMCSUM_TX] &&
3089 nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3090 conf.flags |= VXLAN_F_REMCSUM_TX;
3091
3092 if (data[IFLA_VXLAN_REMCSUM_RX] &&
3093 nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3094 conf.flags |= VXLAN_F_REMCSUM_RX;
3095
3096 if (data[IFLA_VXLAN_GBP])
3097 conf.flags |= VXLAN_F_GBP;
3098
3099 if (data[IFLA_VXLAN_GPE])
3100 conf.flags |= VXLAN_F_GPE;
3101
3102 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3103 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3104
3105 if (tb[IFLA_MTU])
3106 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3107
3108 return vxlan_dev_configure(src_net, dev, &conf);
3109}
3110
3111static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3112{
3113 struct vxlan_dev *vxlan = netdev_priv(dev);
3114 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3115
3116 vxlan_flush(vxlan, true);
3117
3118 spin_lock(&vn->sock_lock);
3119 if (!hlist_unhashed(&vxlan->hlist))
3120 hlist_del_rcu(&vxlan->hlist);
3121 spin_unlock(&vn->sock_lock);
3122
3123 gro_cells_destroy(&vxlan->gro_cells);
3124 list_del(&vxlan->next);
3125 unregister_netdevice_queue(dev, head);
3126}
3127
3128static size_t vxlan_get_size(const struct net_device *dev)
3129{
3130
3131 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
3132 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3133 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3134 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3135 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
3136 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
3137 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3138 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
3139 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
3140 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
3141 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
3142 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
3143 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */
3144 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3145 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3146 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3147 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3148 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3149 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3150 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3151 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3152 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3153 0;
3154}
3155
3156static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3157{
3158 const struct vxlan_dev *vxlan = netdev_priv(dev);
3159 const struct vxlan_rdst *dst = &vxlan->default_dst;
3160 struct ifla_vxlan_port_range ports = {
3161 .low = htons(vxlan->cfg.port_min),
3162 .high = htons(vxlan->cfg.port_max),
3163 };
3164
3165 if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3166 goto nla_put_failure;
3167
3168 if (!vxlan_addr_any(&dst->remote_ip)) {
3169 if (dst->remote_ip.sa.sa_family == AF_INET) {
3170 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3171 dst->remote_ip.sin.sin_addr.s_addr))
3172 goto nla_put_failure;
3173#if IS_ENABLED(CONFIG_IPV6)
3174 } else {
3175 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3176 &dst->remote_ip.sin6.sin6_addr))
3177 goto nla_put_failure;
3178#endif
3179 }
3180 }
3181
3182 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3183 goto nla_put_failure;
3184
3185 if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3186 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3187 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3188 vxlan->cfg.saddr.sin.sin_addr.s_addr))
3189 goto nla_put_failure;
3190#if IS_ENABLED(CONFIG_IPV6)
3191 } else {
3192 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3193 &vxlan->cfg.saddr.sin6.sin6_addr))
3194 goto nla_put_failure;
3195#endif
3196 }
3197 }
3198
3199 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3200 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3201 nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3202 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3203 !!(vxlan->flags & VXLAN_F_LEARN)) ||
3204 nla_put_u8(skb, IFLA_VXLAN_PROXY,
3205 !!(vxlan->flags & VXLAN_F_PROXY)) ||
3206 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3207 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3208 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3209 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3210 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3211 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3212 !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3213 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3214 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3215 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3216 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3217 !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3218 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3219 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3220 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3221 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3222 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3223 !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3224 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3225 !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3226 goto nla_put_failure;
3227
3228 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3229 goto nla_put_failure;
3230
3231 if (vxlan->flags & VXLAN_F_GBP &&
3232 nla_put_flag(skb, IFLA_VXLAN_GBP))
3233 goto nla_put_failure;
3234
3235 if (vxlan->flags & VXLAN_F_GPE &&
3236 nla_put_flag(skb, IFLA_VXLAN_GPE))
3237 goto nla_put_failure;
3238
3239 if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3240 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3241 goto nla_put_failure;
3242
3243 return 0;
3244
3245nla_put_failure:
3246 return -EMSGSIZE;
3247}
3248
3249static struct net *vxlan_get_link_net(const struct net_device *dev)
3250{
3251 struct vxlan_dev *vxlan = netdev_priv(dev);
3252
3253 return vxlan->net;
3254}
3255
3256static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3257 .kind = "vxlan",
3258 .maxtype = IFLA_VXLAN_MAX,
3259 .policy = vxlan_policy,
3260 .priv_size = sizeof(struct vxlan_dev),
3261 .setup = vxlan_setup,
3262 .validate = vxlan_validate,
3263 .newlink = vxlan_newlink,
3264 .dellink = vxlan_dellink,
3265 .get_size = vxlan_get_size,
3266 .fill_info = vxlan_fill_info,
3267 .get_link_net = vxlan_get_link_net,
3268};
3269
3270struct net_device *vxlan_dev_create(struct net *net, const char *name,
3271 u8 name_assign_type,
3272 struct vxlan_config *conf)
3273{
3274 struct nlattr *tb[IFLA_MAX + 1];
3275 struct net_device *dev;
3276 int err;
3277
3278 memset(&tb, 0, sizeof(tb));
3279
3280 dev = rtnl_create_link(net, name, name_assign_type,
3281 &vxlan_link_ops, tb);
3282 if (IS_ERR(dev))
3283 return dev;
3284
3285 err = vxlan_dev_configure(net, dev, conf);
3286 if (err < 0) {
3287 free_netdev(dev);
3288 return ERR_PTR(err);
3289 }
3290
3291 err = rtnl_configure_link(dev, NULL);
3292 if (err < 0) {
3293 LIST_HEAD(list_kill);
3294
3295 vxlan_dellink(dev, &list_kill);
3296 unregister_netdevice_many(&list_kill);
3297 return ERR_PTR(err);
3298 }
3299
3300 return dev;
3301}
3302EXPORT_SYMBOL_GPL(vxlan_dev_create);
3303
3304static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3305 struct net_device *dev)
3306{
3307 struct vxlan_dev *vxlan, *next;
3308 LIST_HEAD(list_kill);
3309
3310 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3311 struct vxlan_rdst *dst = &vxlan->default_dst;
3312
3313 /* In case we created vxlan device with carrier
3314 * and we loose the carrier due to module unload
3315 * we also need to remove vxlan device. In other
3316 * cases, it's not necessary and remote_ifindex
3317 * is 0 here, so no matches.
3318 */
3319 if (dst->remote_ifindex == dev->ifindex)
3320 vxlan_dellink(vxlan->dev, &list_kill);
3321 }
3322
3323 unregister_netdevice_many(&list_kill);
3324}
3325
3326static int vxlan_netdevice_event(struct notifier_block *unused,
3327 unsigned long event, void *ptr)
3328{
3329 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3330 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3331
3332 if (event == NETDEV_UNREGISTER)
3333 vxlan_handle_lowerdev_unregister(vn, dev);
3334 else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3335 vxlan_push_rx_ports(dev);
3336
3337 return NOTIFY_DONE;
3338}
3339
3340static struct notifier_block vxlan_notifier_block __read_mostly = {
3341 .notifier_call = vxlan_netdevice_event,
3342};
3343
3344static __net_init int vxlan_init_net(struct net *net)
3345{
3346 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3347 unsigned int h;
3348
3349 INIT_LIST_HEAD(&vn->vxlan_list);
3350 spin_lock_init(&vn->sock_lock);
3351
3352 for (h = 0; h < PORT_HASH_SIZE; ++h)
3353 INIT_HLIST_HEAD(&vn->sock_list[h]);
3354
3355 return 0;
3356}
3357
3358static void __net_exit vxlan_exit_net(struct net *net)
3359{
3360 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3361 struct vxlan_dev *vxlan, *next;
3362 struct net_device *dev, *aux;
3363 LIST_HEAD(list);
3364
3365 rtnl_lock();
3366 for_each_netdev_safe(net, dev, aux)
3367 if (dev->rtnl_link_ops == &vxlan_link_ops)
3368 unregister_netdevice_queue(dev, &list);
3369
3370 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3371 /* If vxlan->dev is in the same netns, it has already been added
3372 * to the list by the previous loop.
3373 */
3374 if (!net_eq(dev_net(vxlan->dev), net)) {
3375 gro_cells_destroy(&vxlan->gro_cells);
3376 unregister_netdevice_queue(vxlan->dev, &list);
3377 }
3378 }
3379
3380 unregister_netdevice_many(&list);
3381 rtnl_unlock();
3382}
3383
3384static struct pernet_operations vxlan_net_ops = {
3385 .init = vxlan_init_net,
3386 .exit = vxlan_exit_net,
3387 .id = &vxlan_net_id,
3388 .size = sizeof(struct vxlan_net),
3389};
3390
3391static int __init vxlan_init_module(void)
3392{
3393 int rc;
3394
3395 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3396
3397 rc = register_pernet_subsys(&vxlan_net_ops);
3398 if (rc)
3399 goto out1;
3400
3401 rc = register_netdevice_notifier(&vxlan_notifier_block);
3402 if (rc)
3403 goto out2;
3404
3405 rc = rtnl_link_register(&vxlan_link_ops);
3406 if (rc)
3407 goto out3;
3408
3409 return 0;
3410out3:
3411 unregister_netdevice_notifier(&vxlan_notifier_block);
3412out2:
3413 unregister_pernet_subsys(&vxlan_net_ops);
3414out1:
3415 return rc;
3416}
3417late_initcall(vxlan_init_module);
3418
3419static void __exit vxlan_cleanup_module(void)
3420{
3421 rtnl_link_unregister(&vxlan_link_ops);
3422 unregister_netdevice_notifier(&vxlan_notifier_block);
3423 unregister_pernet_subsys(&vxlan_net_ops);
3424 /* rcu_barrier() is called by netns */
3425}
3426module_exit(vxlan_cleanup_module);
3427
3428MODULE_LICENSE("GPL");
3429MODULE_VERSION(VXLAN_VERSION);
3430MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3431MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3432MODULE_ALIAS_RTNL_LINK("vxlan");
Ubuntu Artful kernel mirror