Merge tag 'tty-5.17-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty
[linux-2.6-microblaze.git] / drivers / net / vxlan.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * VXLAN: Virtual eXtensible Local Area Network
4  *
5  * Copyright (c) 2012-2013 Vyatta Inc.
6  */
7
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10 #include <linux/kernel.h>
11 #include <linux/module.h>
12 #include <linux/errno.h>
13 #include <linux/slab.h>
14 #include <linux/udp.h>
15 #include <linux/igmp.h>
16 #include <linux/if_ether.h>
17 #include <linux/ethtool.h>
18 #include <net/arp.h>
19 #include <net/ndisc.h>
20 #include <net/gro.h>
21 #include <net/ipv6_stubs.h>
22 #include <net/ip.h>
23 #include <net/icmp.h>
24 #include <net/rtnetlink.h>
25 #include <net/inet_ecn.h>
26 #include <net/net_namespace.h>
27 #include <net/netns/generic.h>
28 #include <net/tun_proto.h>
29 #include <net/vxlan.h>
30 #include <net/nexthop.h>
31
32 #if IS_ENABLED(CONFIG_IPV6)
33 #include <net/ip6_tunnel.h>
34 #include <net/ip6_checksum.h>
35 #endif
36
37 #define VXLAN_VERSION   "0.1"
38
39 #define PORT_HASH_BITS  8
40 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
41 #define FDB_AGE_DEFAULT 300 /* 5 min */
42 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
43
44 /* UDP port for VXLAN traffic.
45  * The IANA assigned port is 4789, but the Linux default is 8472
46  * for compatibility with early adopters.
47  */
48 static unsigned short vxlan_port __read_mostly = 8472;
49 module_param_named(udp_port, vxlan_port, ushort, 0444);
50 MODULE_PARM_DESC(udp_port, "Destination UDP port");
51
52 static bool log_ecn_error = true;
53 module_param(log_ecn_error, bool, 0644);
54 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
55
56 static unsigned int vxlan_net_id;
57 static struct rtnl_link_ops vxlan_link_ops;
58
59 static const u8 all_zeros_mac[ETH_ALEN + 2];
60
61 static int vxlan_sock_add(struct vxlan_dev *vxlan);
62
63 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
64
65 /* per-network namespace private data for this module */
66 struct vxlan_net {
67         struct list_head  vxlan_list;
68         struct hlist_head sock_list[PORT_HASH_SIZE];
69         spinlock_t        sock_lock;
70         struct notifier_block nexthop_notifier_block;
71 };
72
73 /* Forwarding table entry */
74 struct vxlan_fdb {
75         struct hlist_node hlist;        /* linked list of entries */
76         struct rcu_head   rcu;
77         unsigned long     updated;      /* jiffies */
78         unsigned long     used;
79         struct list_head  remotes;
80         u8                eth_addr[ETH_ALEN];
81         u16               state;        /* see ndm_state */
82         __be32            vni;
83         u16               flags;        /* see ndm_flags and below */
84         struct list_head  nh_list;
85         struct nexthop __rcu *nh;
86         struct vxlan_dev  __rcu *vdev;
87 };
88
89 #define NTF_VXLAN_ADDED_BY_USER 0x100
90
91 /* salt for hash table */
92 static u32 vxlan_salt __read_mostly;
93
94 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
95 {
96         return vs->flags & VXLAN_F_COLLECT_METADATA ||
97                ip_tunnel_collect_metadata();
98 }
99
100 #if IS_ENABLED(CONFIG_IPV6)
101 static inline
102 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
103 {
104         if (a->sa.sa_family != b->sa.sa_family)
105                 return false;
106         if (a->sa.sa_family == AF_INET6)
107                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
108         else
109                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
110 }
111
112 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
113 {
114         if (nla_len(nla) >= sizeof(struct in6_addr)) {
115                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
116                 ip->sa.sa_family = AF_INET6;
117                 return 0;
118         } else if (nla_len(nla) >= sizeof(__be32)) {
119                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
120                 ip->sa.sa_family = AF_INET;
121                 return 0;
122         } else {
123                 return -EAFNOSUPPORT;
124         }
125 }
126
127 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
128                               const union vxlan_addr *ip)
129 {
130         if (ip->sa.sa_family == AF_INET6)
131                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
132         else
133                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
134 }
135
136 #else /* !CONFIG_IPV6 */
137
138 static inline
139 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
140 {
141         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
142 }
143
144 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
145 {
146         if (nla_len(nla) >= sizeof(struct in6_addr)) {
147                 return -EAFNOSUPPORT;
148         } else if (nla_len(nla) >= sizeof(__be32)) {
149                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
150                 ip->sa.sa_family = AF_INET;
151                 return 0;
152         } else {
153                 return -EAFNOSUPPORT;
154         }
155 }
156
157 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
158                               const union vxlan_addr *ip)
159 {
160         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
161 }
162 #endif
163
164 /* Virtual Network hash table head */
165 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
166 {
167         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
168 }
169
170 /* Socket hash table head */
171 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
172 {
173         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
174
175         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
176 }
177
178 /* First remote destination for a forwarding entry.
179  * Guaranteed to be non-NULL because remotes are never deleted.
180  */
181 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
182 {
183         if (rcu_access_pointer(fdb->nh))
184                 return NULL;
185         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
186 }
187
188 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
189 {
190         if (rcu_access_pointer(fdb->nh))
191                 return NULL;
192         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
193 }
194
195 /* Find VXLAN socket based on network namespace, address family, UDP port,
196  * enabled unshareable flags and socket device binding (see l3mdev with
197  * non-default VRF).
198  */
199 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
200                                           __be16 port, u32 flags, int ifindex)
201 {
202         struct vxlan_sock *vs;
203
204         flags &= VXLAN_F_RCV_FLAGS;
205
206         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
207                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
208                     vxlan_get_sk_family(vs) == family &&
209                     vs->flags == flags &&
210                     vs->sock->sk->sk_bound_dev_if == ifindex)
211                         return vs;
212         }
213         return NULL;
214 }
215
216 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex,
217                                            __be32 vni)
218 {
219         struct vxlan_dev_node *node;
220
221         /* For flow based devices, map all packets to VNI 0 */
222         if (vs->flags & VXLAN_F_COLLECT_METADATA)
223                 vni = 0;
224
225         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
226                 if (node->vxlan->default_dst.remote_vni != vni)
227                         continue;
228
229                 if (IS_ENABLED(CONFIG_IPV6)) {
230                         const struct vxlan_config *cfg = &node->vxlan->cfg;
231
232                         if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
233                             cfg->remote_ifindex != ifindex)
234                                 continue;
235                 }
236
237                 return node->vxlan;
238         }
239
240         return NULL;
241 }
242
243 /* Look up VNI in a per net namespace table */
244 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
245                                         __be32 vni, sa_family_t family,
246                                         __be16 port, u32 flags)
247 {
248         struct vxlan_sock *vs;
249
250         vs = vxlan_find_sock(net, family, port, flags, ifindex);
251         if (!vs)
252                 return NULL;
253
254         return vxlan_vs_find_vni(vs, ifindex, vni);
255 }
256
257 /* Fill in neighbour message in skbuff. */
258 static 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         bool send_ip, send_eth;
266         struct nlmsghdr *nlh;
267         struct nexthop *nh;
268         struct ndmsg *ndm;
269         int nh_family;
270         u32 nh_id;
271
272         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
273         if (nlh == NULL)
274                 return -EMSGSIZE;
275
276         ndm = nlmsg_data(nlh);
277         memset(ndm, 0, sizeof(*ndm));
278
279         send_eth = send_ip = true;
280
281         rcu_read_lock();
282         nh = rcu_dereference(fdb->nh);
283         if (nh) {
284                 nh_family = nexthop_get_family(nh);
285                 nh_id = nh->id;
286         }
287         rcu_read_unlock();
288
289         if (type == RTM_GETNEIGH) {
290                 if (rdst) {
291                         send_ip = !vxlan_addr_any(&rdst->remote_ip);
292                         ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
293                 } else if (nh) {
294                         ndm->ndm_family = nh_family;
295                 }
296                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
297         } else
298                 ndm->ndm_family = AF_BRIDGE;
299         ndm->ndm_state = fdb->state;
300         ndm->ndm_ifindex = vxlan->dev->ifindex;
301         ndm->ndm_flags = fdb->flags;
302         if (rdst && rdst->offloaded)
303                 ndm->ndm_flags |= NTF_OFFLOADED;
304         ndm->ndm_type = RTN_UNICAST;
305
306         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
307             nla_put_s32(skb, NDA_LINK_NETNSID,
308                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
309                 goto nla_put_failure;
310
311         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
312                 goto nla_put_failure;
313         if (nh) {
314                 if (nla_put_u32(skb, NDA_NH_ID, nh_id))
315                         goto nla_put_failure;
316         } else if (rdst) {
317                 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST,
318                                                   &rdst->remote_ip))
319                         goto nla_put_failure;
320
321                 if (rdst->remote_port &&
322                     rdst->remote_port != vxlan->cfg.dst_port &&
323                     nla_put_be16(skb, NDA_PORT, rdst->remote_port))
324                         goto nla_put_failure;
325                 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
326                     nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
327                         goto nla_put_failure;
328                 if (rdst->remote_ifindex &&
329                     nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
330                         goto nla_put_failure;
331         }
332
333         if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
334             nla_put_u32(skb, NDA_SRC_VNI,
335                         be32_to_cpu(fdb->vni)))
336                 goto nla_put_failure;
337
338         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
339         ci.ndm_confirmed = 0;
340         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
341         ci.ndm_refcnt    = 0;
342
343         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
344                 goto nla_put_failure;
345
346         nlmsg_end(skb, nlh);
347         return 0;
348
349 nla_put_failure:
350         nlmsg_cancel(skb, nlh);
351         return -EMSGSIZE;
352 }
353
354 static inline size_t vxlan_nlmsg_size(void)
355 {
356         return NLMSG_ALIGN(sizeof(struct ndmsg))
357                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
358                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
359                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
360                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
361                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
362                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
363                 + nla_total_size(sizeof(struct nda_cacheinfo));
364 }
365
366 static void __vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
367                                struct vxlan_rdst *rd, int type)
368 {
369         struct net *net = dev_net(vxlan->dev);
370         struct sk_buff *skb;
371         int err = -ENOBUFS;
372
373         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
374         if (skb == NULL)
375                 goto errout;
376
377         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
378         if (err < 0) {
379                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
380                 WARN_ON(err == -EMSGSIZE);
381                 kfree_skb(skb);
382                 goto errout;
383         }
384
385         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
386         return;
387 errout:
388         if (err < 0)
389                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
390 }
391
392 static void vxlan_fdb_switchdev_notifier_info(const struct vxlan_dev *vxlan,
393                             const struct vxlan_fdb *fdb,
394                             const struct vxlan_rdst *rd,
395                             struct netlink_ext_ack *extack,
396                             struct switchdev_notifier_vxlan_fdb_info *fdb_info)
397 {
398         fdb_info->info.dev = vxlan->dev;
399         fdb_info->info.extack = extack;
400         fdb_info->remote_ip = rd->remote_ip;
401         fdb_info->remote_port = rd->remote_port;
402         fdb_info->remote_vni = rd->remote_vni;
403         fdb_info->remote_ifindex = rd->remote_ifindex;
404         memcpy(fdb_info->eth_addr, fdb->eth_addr, ETH_ALEN);
405         fdb_info->vni = fdb->vni;
406         fdb_info->offloaded = rd->offloaded;
407         fdb_info->added_by_user = fdb->flags & NTF_VXLAN_ADDED_BY_USER;
408 }
409
410 static int vxlan_fdb_switchdev_call_notifiers(struct vxlan_dev *vxlan,
411                                               struct vxlan_fdb *fdb,
412                                               struct vxlan_rdst *rd,
413                                               bool adding,
414                                               struct netlink_ext_ack *extack)
415 {
416         struct switchdev_notifier_vxlan_fdb_info info;
417         enum switchdev_notifier_type notifier_type;
418         int ret;
419
420         if (WARN_ON(!rd))
421                 return 0;
422
423         notifier_type = adding ? SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE
424                                : SWITCHDEV_VXLAN_FDB_DEL_TO_DEVICE;
425         vxlan_fdb_switchdev_notifier_info(vxlan, fdb, rd, NULL, &info);
426         ret = call_switchdev_notifiers(notifier_type, vxlan->dev,
427                                        &info.info, extack);
428         return notifier_to_errno(ret);
429 }
430
431 static int vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
432                             struct vxlan_rdst *rd, int type, bool swdev_notify,
433                             struct netlink_ext_ack *extack)
434 {
435         int err;
436
437         if (swdev_notify && rd) {
438                 switch (type) {
439                 case RTM_NEWNEIGH:
440                         err = vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
441                                                                  true, extack);
442                         if (err)
443                                 return err;
444                         break;
445                 case RTM_DELNEIGH:
446                         vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
447                                                            false, extack);
448                         break;
449                 }
450         }
451
452         __vxlan_fdb_notify(vxlan, fdb, rd, type);
453         return 0;
454 }
455
456 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
457 {
458         struct vxlan_dev *vxlan = netdev_priv(dev);
459         struct vxlan_fdb f = {
460                 .state = NUD_STALE,
461         };
462         struct vxlan_rdst remote = {
463                 .remote_ip = *ipa, /* goes to NDA_DST */
464                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
465         };
466
467         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
468 }
469
470 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
471 {
472         struct vxlan_fdb f = {
473                 .state = NUD_STALE,
474         };
475         struct vxlan_rdst remote = { };
476
477         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
478
479         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
480 }
481
482 /* Hash Ethernet address */
483 static u32 eth_hash(const unsigned char *addr)
484 {
485         u64 value = get_unaligned((u64 *)addr);
486
487         /* only want 6 bytes */
488 #ifdef __BIG_ENDIAN
489         value >>= 16;
490 #else
491         value <<= 16;
492 #endif
493         return hash_64(value, FDB_HASH_BITS);
494 }
495
496 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
497 {
498         /* use 1 byte of OUI and 3 bytes of NIC */
499         u32 key = get_unaligned((u32 *)(addr + 2));
500
501         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
502 }
503
504 static u32 fdb_head_index(struct vxlan_dev *vxlan, const u8 *mac, __be32 vni)
505 {
506         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
507                 return eth_vni_hash(mac, vni);
508         else
509                 return eth_hash(mac);
510 }
511
512 /* Hash chain to use given mac address */
513 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
514                                                 const u8 *mac, __be32 vni)
515 {
516         return &vxlan->fdb_head[fdb_head_index(vxlan, mac, vni)];
517 }
518
519 /* Look up Ethernet address in forwarding table */
520 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
521                                           const u8 *mac, __be32 vni)
522 {
523         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
524         struct vxlan_fdb *f;
525
526         hlist_for_each_entry_rcu(f, head, hlist) {
527                 if (ether_addr_equal(mac, f->eth_addr)) {
528                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
529                                 if (vni == f->vni)
530                                         return f;
531                         } else {
532                                 return f;
533                         }
534                 }
535         }
536
537         return NULL;
538 }
539
540 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
541                                         const u8 *mac, __be32 vni)
542 {
543         struct vxlan_fdb *f;
544
545         f = __vxlan_find_mac(vxlan, mac, vni);
546         if (f && f->used != jiffies)
547                 f->used = jiffies;
548
549         return f;
550 }
551
552 /* caller should hold vxlan->hash_lock */
553 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
554                                               union vxlan_addr *ip, __be16 port,
555                                               __be32 vni, __u32 ifindex)
556 {
557         struct vxlan_rdst *rd;
558
559         list_for_each_entry(rd, &f->remotes, list) {
560                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
561                     rd->remote_port == port &&
562                     rd->remote_vni == vni &&
563                     rd->remote_ifindex == ifindex)
564                         return rd;
565         }
566
567         return NULL;
568 }
569
570 int vxlan_fdb_find_uc(struct net_device *dev, const u8 *mac, __be32 vni,
571                       struct switchdev_notifier_vxlan_fdb_info *fdb_info)
572 {
573         struct vxlan_dev *vxlan = netdev_priv(dev);
574         u8 eth_addr[ETH_ALEN + 2] = { 0 };
575         struct vxlan_rdst *rdst;
576         struct vxlan_fdb *f;
577         int rc = 0;
578
579         if (is_multicast_ether_addr(mac) ||
580             is_zero_ether_addr(mac))
581                 return -EINVAL;
582
583         ether_addr_copy(eth_addr, mac);
584
585         rcu_read_lock();
586
587         f = __vxlan_find_mac(vxlan, eth_addr, vni);
588         if (!f) {
589                 rc = -ENOENT;
590                 goto out;
591         }
592
593         rdst = first_remote_rcu(f);
594         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, NULL, fdb_info);
595
596 out:
597         rcu_read_unlock();
598         return rc;
599 }
600 EXPORT_SYMBOL_GPL(vxlan_fdb_find_uc);
601
602 static int vxlan_fdb_notify_one(struct notifier_block *nb,
603                                 const struct vxlan_dev *vxlan,
604                                 const struct vxlan_fdb *f,
605                                 const struct vxlan_rdst *rdst,
606                                 struct netlink_ext_ack *extack)
607 {
608         struct switchdev_notifier_vxlan_fdb_info fdb_info;
609         int rc;
610
611         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, extack, &fdb_info);
612         rc = nb->notifier_call(nb, SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE,
613                                &fdb_info);
614         return notifier_to_errno(rc);
615 }
616
617 int vxlan_fdb_replay(const struct net_device *dev, __be32 vni,
618                      struct notifier_block *nb,
619                      struct netlink_ext_ack *extack)
620 {
621         struct vxlan_dev *vxlan;
622         struct vxlan_rdst *rdst;
623         struct vxlan_fdb *f;
624         unsigned int h;
625         int rc = 0;
626
627         if (!netif_is_vxlan(dev))
628                 return -EINVAL;
629         vxlan = netdev_priv(dev);
630
631         for (h = 0; h < FDB_HASH_SIZE; ++h) {
632                 spin_lock_bh(&vxlan->hash_lock[h]);
633                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) {
634                         if (f->vni == vni) {
635                                 list_for_each_entry(rdst, &f->remotes, list) {
636                                         rc = vxlan_fdb_notify_one(nb, vxlan,
637                                                                   f, rdst,
638                                                                   extack);
639                                         if (rc)
640                                                 goto unlock;
641                                 }
642                         }
643                 }
644                 spin_unlock_bh(&vxlan->hash_lock[h]);
645         }
646         return 0;
647
648 unlock:
649         spin_unlock_bh(&vxlan->hash_lock[h]);
650         return rc;
651 }
652 EXPORT_SYMBOL_GPL(vxlan_fdb_replay);
653
654 void vxlan_fdb_clear_offload(const struct net_device *dev, __be32 vni)
655 {
656         struct vxlan_dev *vxlan;
657         struct vxlan_rdst *rdst;
658         struct vxlan_fdb *f;
659         unsigned int h;
660
661         if (!netif_is_vxlan(dev))
662                 return;
663         vxlan = netdev_priv(dev);
664
665         for (h = 0; h < FDB_HASH_SIZE; ++h) {
666                 spin_lock_bh(&vxlan->hash_lock[h]);
667                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist)
668                         if (f->vni == vni)
669                                 list_for_each_entry(rdst, &f->remotes, list)
670                                         rdst->offloaded = false;
671                 spin_unlock_bh(&vxlan->hash_lock[h]);
672         }
673
674 }
675 EXPORT_SYMBOL_GPL(vxlan_fdb_clear_offload);
676
677 /* Replace destination of unicast mac */
678 static int vxlan_fdb_replace(struct vxlan_fdb *f,
679                              union vxlan_addr *ip, __be16 port, __be32 vni,
680                              __u32 ifindex, struct vxlan_rdst *oldrd)
681 {
682         struct vxlan_rdst *rd;
683
684         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
685         if (rd)
686                 return 0;
687
688         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
689         if (!rd)
690                 return 0;
691
692         *oldrd = *rd;
693         dst_cache_reset(&rd->dst_cache);
694         rd->remote_ip = *ip;
695         rd->remote_port = port;
696         rd->remote_vni = vni;
697         rd->remote_ifindex = ifindex;
698         rd->offloaded = false;
699         return 1;
700 }
701
702 /* Add/update destinations for multicast */
703 static int vxlan_fdb_append(struct vxlan_fdb *f,
704                             union vxlan_addr *ip, __be16 port, __be32 vni,
705                             __u32 ifindex, struct vxlan_rdst **rdp)
706 {
707         struct vxlan_rdst *rd;
708
709         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
710         if (rd)
711                 return 0;
712
713         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
714         if (rd == NULL)
715                 return -ENOBUFS;
716
717         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
718                 kfree(rd);
719                 return -ENOBUFS;
720         }
721
722         rd->remote_ip = *ip;
723         rd->remote_port = port;
724         rd->offloaded = false;
725         rd->remote_vni = vni;
726         rd->remote_ifindex = ifindex;
727
728         list_add_tail_rcu(&rd->list, &f->remotes);
729
730         *rdp = rd;
731         return 1;
732 }
733
734 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
735                                           unsigned int off,
736                                           struct vxlanhdr *vh, size_t hdrlen,
737                                           __be32 vni_field,
738                                           struct gro_remcsum *grc,
739                                           bool nopartial)
740 {
741         size_t start, offset;
742
743         if (skb->remcsum_offload)
744                 return vh;
745
746         if (!NAPI_GRO_CB(skb)->csum_valid)
747                 return NULL;
748
749         start = vxlan_rco_start(vni_field);
750         offset = start + vxlan_rco_offset(vni_field);
751
752         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
753                                      start, offset, grc, nopartial);
754
755         skb->remcsum_offload = 1;
756
757         return vh;
758 }
759
760 static struct sk_buff *vxlan_gro_receive(struct sock *sk,
761                                          struct list_head *head,
762                                          struct sk_buff *skb)
763 {
764         struct sk_buff *pp = NULL;
765         struct sk_buff *p;
766         struct vxlanhdr *vh, *vh2;
767         unsigned int hlen, off_vx;
768         int flush = 1;
769         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
770         __be32 flags;
771         struct gro_remcsum grc;
772
773         skb_gro_remcsum_init(&grc);
774
775         off_vx = skb_gro_offset(skb);
776         hlen = off_vx + sizeof(*vh);
777         vh   = skb_gro_header_fast(skb, off_vx);
778         if (skb_gro_header_hard(skb, hlen)) {
779                 vh = skb_gro_header_slow(skb, hlen, off_vx);
780                 if (unlikely(!vh))
781                         goto out;
782         }
783
784         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
785
786         flags = vh->vx_flags;
787
788         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
789                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
790                                        vh->vx_vni, &grc,
791                                        !!(vs->flags &
792                                           VXLAN_F_REMCSUM_NOPARTIAL));
793
794                 if (!vh)
795                         goto out;
796         }
797
798         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
799
800         list_for_each_entry(p, head, list) {
801                 if (!NAPI_GRO_CB(p)->same_flow)
802                         continue;
803
804                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
805                 if (vh->vx_flags != vh2->vx_flags ||
806                     vh->vx_vni != vh2->vx_vni) {
807                         NAPI_GRO_CB(p)->same_flow = 0;
808                         continue;
809                 }
810         }
811
812         pp = call_gro_receive(eth_gro_receive, head, skb);
813         flush = 0;
814
815 out:
816         skb_gro_flush_final_remcsum(skb, pp, flush, &grc);
817
818         return pp;
819 }
820
821 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
822 {
823         /* Sets 'skb->inner_mac_header' since we are always called with
824          * 'skb->encapsulation' set.
825          */
826         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
827 }
828
829 static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan, const u8 *mac,
830                                          __u16 state, __be32 src_vni,
831                                          __u16 ndm_flags)
832 {
833         struct vxlan_fdb *f;
834
835         f = kmalloc(sizeof(*f), GFP_ATOMIC);
836         if (!f)
837                 return NULL;
838         f->state = state;
839         f->flags = ndm_flags;
840         f->updated = f->used = jiffies;
841         f->vni = src_vni;
842         f->nh = NULL;
843         RCU_INIT_POINTER(f->vdev, vxlan);
844         INIT_LIST_HEAD(&f->nh_list);
845         INIT_LIST_HEAD(&f->remotes);
846         memcpy(f->eth_addr, mac, ETH_ALEN);
847
848         return f;
849 }
850
851 static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac,
852                              __be32 src_vni, struct vxlan_fdb *f)
853 {
854         ++vxlan->addrcnt;
855         hlist_add_head_rcu(&f->hlist,
856                            vxlan_fdb_head(vxlan, mac, src_vni));
857 }
858
859 static int vxlan_fdb_nh_update(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
860                                u32 nhid, struct netlink_ext_ack *extack)
861 {
862         struct nexthop *old_nh = rtnl_dereference(fdb->nh);
863         struct nexthop *nh;
864         int err = -EINVAL;
865
866         if (old_nh && old_nh->id == nhid)
867                 return 0;
868
869         nh = nexthop_find_by_id(vxlan->net, nhid);
870         if (!nh) {
871                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
872                 goto err_inval;
873         }
874
875         if (nh) {
876                 if (!nexthop_get(nh)) {
877                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
878                         nh = NULL;
879                         goto err_inval;
880                 }
881                 if (!nexthop_is_fdb(nh)) {
882                         NL_SET_ERR_MSG(extack, "Nexthop is not a fdb nexthop");
883                         goto err_inval;
884                 }
885
886                 if (!nexthop_is_multipath(nh)) {
887                         NL_SET_ERR_MSG(extack, "Nexthop is not a multipath group");
888                         goto err_inval;
889                 }
890
891                 /* check nexthop group family */
892                 switch (vxlan->default_dst.remote_ip.sa.sa_family) {
893                 case AF_INET:
894                         if (!nexthop_has_v4(nh)) {
895                                 err = -EAFNOSUPPORT;
896                                 NL_SET_ERR_MSG(extack, "Nexthop group family not supported");
897                                 goto err_inval;
898                         }
899                         break;
900                 case AF_INET6:
901                         if (nexthop_has_v4(nh)) {
902                                 err = -EAFNOSUPPORT;
903                                 NL_SET_ERR_MSG(extack, "Nexthop group family not supported");
904                                 goto err_inval;
905                         }
906                 }
907         }
908
909         if (old_nh) {
910                 list_del_rcu(&fdb->nh_list);
911                 nexthop_put(old_nh);
912         }
913         rcu_assign_pointer(fdb->nh, nh);
914         list_add_tail_rcu(&fdb->nh_list, &nh->fdb_list);
915         return 1;
916
917 err_inval:
918         if (nh)
919                 nexthop_put(nh);
920         return err;
921 }
922
923 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
924                             const u8 *mac, union vxlan_addr *ip,
925                             __u16 state, __be16 port, __be32 src_vni,
926                             __be32 vni, __u32 ifindex, __u16 ndm_flags,
927                             u32 nhid, struct vxlan_fdb **fdb,
928                             struct netlink_ext_ack *extack)
929 {
930         struct vxlan_rdst *rd = NULL;
931         struct vxlan_fdb *f;
932         int rc;
933
934         if (vxlan->cfg.addrmax &&
935             vxlan->addrcnt >= vxlan->cfg.addrmax)
936                 return -ENOSPC;
937
938         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
939         f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags);
940         if (!f)
941                 return -ENOMEM;
942
943         if (nhid)
944                 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack);
945         else
946                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
947         if (rc < 0)
948                 goto errout;
949
950         *fdb = f;
951
952         return 0;
953
954 errout:
955         kfree(f);
956         return rc;
957 }
958
959 static void __vxlan_fdb_free(struct vxlan_fdb *f)
960 {
961         struct vxlan_rdst *rd, *nd;
962         struct nexthop *nh;
963
964         nh = rcu_dereference_raw(f->nh);
965         if (nh) {
966                 rcu_assign_pointer(f->nh, NULL);
967                 rcu_assign_pointer(f->vdev, NULL);
968                 nexthop_put(nh);
969         }
970
971         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
972                 dst_cache_destroy(&rd->dst_cache);
973                 kfree(rd);
974         }
975         kfree(f);
976 }
977
978 static void vxlan_fdb_free(struct rcu_head *head)
979 {
980         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
981
982         __vxlan_fdb_free(f);
983 }
984
985 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
986                               bool do_notify, bool swdev_notify)
987 {
988         struct vxlan_rdst *rd;
989
990         netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr);
991
992         --vxlan->addrcnt;
993         if (do_notify) {
994                 if (rcu_access_pointer(f->nh))
995                         vxlan_fdb_notify(vxlan, f, NULL, RTM_DELNEIGH,
996                                          swdev_notify, NULL);
997                 else
998                         list_for_each_entry(rd, &f->remotes, list)
999                                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH,
1000                                                  swdev_notify, NULL);
1001         }
1002
1003         hlist_del_rcu(&f->hlist);
1004         list_del_rcu(&f->nh_list);
1005         call_rcu(&f->rcu, vxlan_fdb_free);
1006 }
1007
1008 static void vxlan_dst_free(struct rcu_head *head)
1009 {
1010         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
1011
1012         dst_cache_destroy(&rd->dst_cache);
1013         kfree(rd);
1014 }
1015
1016 static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan,
1017                                      union vxlan_addr *ip,
1018                                      __u16 state, __u16 flags,
1019                                      __be16 port, __be32 vni,
1020                                      __u32 ifindex, __u16 ndm_flags,
1021                                      struct vxlan_fdb *f, u32 nhid,
1022                                      bool swdev_notify,
1023                                      struct netlink_ext_ack *extack)
1024 {
1025         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
1026         struct vxlan_rdst *rd = NULL;
1027         struct vxlan_rdst oldrd;
1028         int notify = 0;
1029         int rc = 0;
1030         int err;
1031
1032         if (nhid && !rcu_access_pointer(f->nh)) {
1033                 NL_SET_ERR_MSG(extack,
1034                                "Cannot replace an existing non nexthop fdb with a nexthop");
1035                 return -EOPNOTSUPP;
1036         }
1037
1038         if (nhid && (flags & NLM_F_APPEND)) {
1039                 NL_SET_ERR_MSG(extack,
1040                                "Cannot append to a nexthop fdb");
1041                 return -EOPNOTSUPP;
1042         }
1043
1044         /* Do not allow an externally learned entry to take over an entry added
1045          * by the user.
1046          */
1047         if (!(fdb_flags & NTF_EXT_LEARNED) ||
1048             !(f->flags & NTF_VXLAN_ADDED_BY_USER)) {
1049                 if (f->state != state) {
1050                         f->state = state;
1051                         f->updated = jiffies;
1052                         notify = 1;
1053                 }
1054                 if (f->flags != fdb_flags) {
1055                         f->flags = fdb_flags;
1056                         f->updated = jiffies;
1057                         notify = 1;
1058                 }
1059         }
1060
1061         if ((flags & NLM_F_REPLACE)) {
1062                 /* Only change unicasts */
1063                 if (!(is_multicast_ether_addr(f->eth_addr) ||
1064                       is_zero_ether_addr(f->eth_addr))) {
1065                         if (nhid) {
1066                                 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack);
1067                                 if (rc < 0)
1068                                         return rc;
1069                         } else {
1070                                 rc = vxlan_fdb_replace(f, ip, port, vni,
1071                                                        ifindex, &oldrd);
1072                         }
1073                         notify |= rc;
1074                 } else {
1075                         NL_SET_ERR_MSG(extack, "Cannot replace non-unicast fdb entries");
1076                         return -EOPNOTSUPP;
1077                 }
1078         }
1079         if ((flags & NLM_F_APPEND) &&
1080             (is_multicast_ether_addr(f->eth_addr) ||
1081              is_zero_ether_addr(f->eth_addr))) {
1082                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
1083
1084                 if (rc < 0)
1085                         return rc;
1086                 notify |= rc;
1087         }
1088
1089         if (ndm_flags & NTF_USE)
1090                 f->used = jiffies;
1091
1092         if (notify) {
1093                 if (rd == NULL)
1094                         rd = first_remote_rtnl(f);
1095
1096                 err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH,
1097                                        swdev_notify, extack);
1098                 if (err)
1099                         goto err_notify;
1100         }
1101
1102         return 0;
1103
1104 err_notify:
1105         if (nhid)
1106                 return err;
1107         if ((flags & NLM_F_REPLACE) && rc)
1108                 *rd = oldrd;
1109         else if ((flags & NLM_F_APPEND) && rc) {
1110                 list_del_rcu(&rd->list);
1111                 call_rcu(&rd->rcu, vxlan_dst_free);
1112         }
1113         return err;
1114 }
1115
1116 static int vxlan_fdb_update_create(struct vxlan_dev *vxlan,
1117                                    const u8 *mac, union vxlan_addr *ip,
1118                                    __u16 state, __u16 flags,
1119                                    __be16 port, __be32 src_vni, __be32 vni,
1120                                    __u32 ifindex, __u16 ndm_flags, u32 nhid,
1121                                    bool swdev_notify,
1122                                    struct netlink_ext_ack *extack)
1123 {
1124         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
1125         struct vxlan_fdb *f;
1126         int rc;
1127
1128         /* Disallow replace to add a multicast entry */
1129         if ((flags & NLM_F_REPLACE) &&
1130             (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
1131                 return -EOPNOTSUPP;
1132
1133         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
1134         rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
1135                               vni, ifindex, fdb_flags, nhid, &f, extack);
1136         if (rc < 0)
1137                 return rc;
1138
1139         vxlan_fdb_insert(vxlan, mac, src_vni, f);
1140         rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH,
1141                               swdev_notify, extack);
1142         if (rc)
1143                 goto err_notify;
1144
1145         return 0;
1146
1147 err_notify:
1148         vxlan_fdb_destroy(vxlan, f, false, false);
1149         return rc;
1150 }
1151
1152 /* Add new entry to forwarding table -- assumes lock held */
1153 static int vxlan_fdb_update(struct vxlan_dev *vxlan,
1154                             const u8 *mac, union vxlan_addr *ip,
1155                             __u16 state, __u16 flags,
1156                             __be16 port, __be32 src_vni, __be32 vni,
1157                             __u32 ifindex, __u16 ndm_flags, u32 nhid,
1158                             bool swdev_notify,
1159                             struct netlink_ext_ack *extack)
1160 {
1161         struct vxlan_fdb *f;
1162
1163         f = __vxlan_find_mac(vxlan, mac, src_vni);
1164         if (f) {
1165                 if (flags & NLM_F_EXCL) {
1166                         netdev_dbg(vxlan->dev,
1167                                    "lost race to create %pM\n", mac);
1168                         return -EEXIST;
1169                 }
1170
1171                 return vxlan_fdb_update_existing(vxlan, ip, state, flags, port,
1172                                                  vni, ifindex, ndm_flags, f,
1173                                                  nhid, swdev_notify, extack);
1174         } else {
1175                 if (!(flags & NLM_F_CREATE))
1176                         return -ENOENT;
1177
1178                 return vxlan_fdb_update_create(vxlan, mac, ip, state, flags,
1179                                                port, src_vni, vni, ifindex,
1180                                                ndm_flags, nhid, swdev_notify,
1181                                                extack);
1182         }
1183 }
1184
1185 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
1186                                   struct vxlan_rdst *rd, bool swdev_notify)
1187 {
1188         list_del_rcu(&rd->list);
1189         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL);
1190         call_rcu(&rd->rcu, vxlan_dst_free);
1191 }
1192
1193 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
1194                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
1195                            __be32 *vni, u32 *ifindex, u32 *nhid)
1196 {
1197         struct net *net = dev_net(vxlan->dev);
1198         int err;
1199
1200         if (tb[NDA_NH_ID] && (tb[NDA_DST] || tb[NDA_VNI] || tb[NDA_IFINDEX] ||
1201             tb[NDA_PORT]))
1202                 return -EINVAL;
1203
1204         if (tb[NDA_DST]) {
1205                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
1206                 if (err)
1207                         return err;
1208         } else {
1209                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
1210
1211                 if (remote->sa.sa_family == AF_INET) {
1212                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
1213                         ip->sa.sa_family = AF_INET;
1214 #if IS_ENABLED(CONFIG_IPV6)
1215                 } else {
1216                         ip->sin6.sin6_addr = in6addr_any;
1217                         ip->sa.sa_family = AF_INET6;
1218 #endif
1219                 }
1220         }
1221
1222         if (tb[NDA_PORT]) {
1223                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
1224                         return -EINVAL;
1225                 *port = nla_get_be16(tb[NDA_PORT]);
1226         } else {
1227                 *port = vxlan->cfg.dst_port;
1228         }
1229
1230         if (tb[NDA_VNI]) {
1231                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
1232                         return -EINVAL;
1233                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1234         } else {
1235                 *vni = vxlan->default_dst.remote_vni;
1236         }
1237
1238         if (tb[NDA_SRC_VNI]) {
1239                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
1240                         return -EINVAL;
1241                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
1242         } else {
1243                 *src_vni = vxlan->default_dst.remote_vni;
1244         }
1245
1246         if (tb[NDA_IFINDEX]) {
1247                 struct net_device *tdev;
1248
1249                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
1250                         return -EINVAL;
1251                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
1252                 tdev = __dev_get_by_index(net, *ifindex);
1253                 if (!tdev)
1254                         return -EADDRNOTAVAIL;
1255         } else {
1256                 *ifindex = 0;
1257         }
1258
1259         if (tb[NDA_NH_ID])
1260                 *nhid = nla_get_u32(tb[NDA_NH_ID]);
1261         else
1262                 *nhid = 0;
1263
1264         return 0;
1265 }
1266
1267 /* Add static entry (via netlink) */
1268 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1269                          struct net_device *dev,
1270                          const unsigned char *addr, u16 vid, u16 flags,
1271                          struct netlink_ext_ack *extack)
1272 {
1273         struct vxlan_dev *vxlan = netdev_priv(dev);
1274         /* struct net *net = dev_net(vxlan->dev); */
1275         union vxlan_addr ip;
1276         __be16 port;
1277         __be32 src_vni, vni;
1278         u32 ifindex, nhid;
1279         u32 hash_index;
1280         int err;
1281
1282         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
1283                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
1284                         ndm->ndm_state);
1285                 return -EINVAL;
1286         }
1287
1288         if (!tb || (!tb[NDA_DST] && !tb[NDA_NH_ID]))
1289                 return -EINVAL;
1290
1291         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex,
1292                               &nhid);
1293         if (err)
1294                 return err;
1295
1296         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
1297                 return -EAFNOSUPPORT;
1298
1299         hash_index = fdb_head_index(vxlan, addr, src_vni);
1300         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1301         err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
1302                                port, src_vni, vni, ifindex,
1303                                ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER,
1304                                nhid, true, extack);
1305         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1306
1307         return err;
1308 }
1309
1310 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
1311                               const unsigned char *addr, union vxlan_addr ip,
1312                               __be16 port, __be32 src_vni, __be32 vni,
1313                               u32 ifindex, bool swdev_notify)
1314 {
1315         struct vxlan_rdst *rd = NULL;
1316         struct vxlan_fdb *f;
1317         int err = -ENOENT;
1318
1319         f = vxlan_find_mac(vxlan, addr, src_vni);
1320         if (!f)
1321                 return err;
1322
1323         if (!vxlan_addr_any(&ip)) {
1324                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
1325                 if (!rd)
1326                         goto out;
1327         }
1328
1329         /* remove a destination if it's not the only one on the list,
1330          * otherwise destroy the fdb entry
1331          */
1332         if (rd && !list_is_singular(&f->remotes)) {
1333                 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify);
1334                 goto out;
1335         }
1336
1337         vxlan_fdb_destroy(vxlan, f, true, swdev_notify);
1338
1339 out:
1340         return 0;
1341 }
1342
1343 /* Delete entry (via netlink) */
1344 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
1345                             struct net_device *dev,
1346                             const unsigned char *addr, u16 vid)
1347 {
1348         struct vxlan_dev *vxlan = netdev_priv(dev);
1349         union vxlan_addr ip;
1350         __be32 src_vni, vni;
1351         u32 ifindex, nhid;
1352         u32 hash_index;
1353         __be16 port;
1354         int err;
1355
1356         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex,
1357                               &nhid);
1358         if (err)
1359                 return err;
1360
1361         hash_index = fdb_head_index(vxlan, addr, src_vni);
1362         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1363         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
1364                                  true);
1365         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1366
1367         return err;
1368 }
1369
1370 /* Dump forwarding table */
1371 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1372                           struct net_device *dev,
1373                           struct net_device *filter_dev, int *idx)
1374 {
1375         struct vxlan_dev *vxlan = netdev_priv(dev);
1376         unsigned int h;
1377         int err = 0;
1378
1379         for (h = 0; h < FDB_HASH_SIZE; ++h) {
1380                 struct vxlan_fdb *f;
1381
1382                 rcu_read_lock();
1383                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
1384                         struct vxlan_rdst *rd;
1385
1386                         if (rcu_access_pointer(f->nh)) {
1387                                 if (*idx < cb->args[2])
1388                                         goto skip_nh;
1389                                 err = vxlan_fdb_info(skb, vxlan, f,
1390                                                      NETLINK_CB(cb->skb).portid,
1391                                                      cb->nlh->nlmsg_seq,
1392                                                      RTM_NEWNEIGH,
1393                                                      NLM_F_MULTI, NULL);
1394                                 if (err < 0) {
1395                                         rcu_read_unlock();
1396                                         goto out;
1397                                 }
1398 skip_nh:
1399                                 *idx += 1;
1400                                 continue;
1401                         }
1402
1403                         list_for_each_entry_rcu(rd, &f->remotes, list) {
1404                                 if (*idx < cb->args[2])
1405                                         goto skip;
1406
1407                                 err = vxlan_fdb_info(skb, vxlan, f,
1408                                                      NETLINK_CB(cb->skb).portid,
1409                                                      cb->nlh->nlmsg_seq,
1410                                                      RTM_NEWNEIGH,
1411                                                      NLM_F_MULTI, rd);
1412                                 if (err < 0) {
1413                                         rcu_read_unlock();
1414                                         goto out;
1415                                 }
1416 skip:
1417                                 *idx += 1;
1418                         }
1419                 }
1420                 rcu_read_unlock();
1421         }
1422 out:
1423         return err;
1424 }
1425
1426 static int vxlan_fdb_get(struct sk_buff *skb,
1427                          struct nlattr *tb[],
1428                          struct net_device *dev,
1429                          const unsigned char *addr,
1430                          u16 vid, u32 portid, u32 seq,
1431                          struct netlink_ext_ack *extack)
1432 {
1433         struct vxlan_dev *vxlan = netdev_priv(dev);
1434         struct vxlan_fdb *f;
1435         __be32 vni;
1436         int err;
1437
1438         if (tb[NDA_VNI])
1439                 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1440         else
1441                 vni = vxlan->default_dst.remote_vni;
1442
1443         rcu_read_lock();
1444
1445         f = __vxlan_find_mac(vxlan, addr, vni);
1446         if (!f) {
1447                 NL_SET_ERR_MSG(extack, "Fdb entry not found");
1448                 err = -ENOENT;
1449                 goto errout;
1450         }
1451
1452         err = vxlan_fdb_info(skb, vxlan, f, portid, seq,
1453                              RTM_NEWNEIGH, 0, first_remote_rcu(f));
1454 errout:
1455         rcu_read_unlock();
1456         return err;
1457 }
1458
1459 /* Watch incoming packets to learn mapping between Ethernet address
1460  * and Tunnel endpoint.
1461  * Return true if packet is bogus and should be dropped.
1462  */
1463 static bool vxlan_snoop(struct net_device *dev,
1464                         union vxlan_addr *src_ip, const u8 *src_mac,
1465                         u32 src_ifindex, __be32 vni)
1466 {
1467         struct vxlan_dev *vxlan = netdev_priv(dev);
1468         struct vxlan_fdb *f;
1469         u32 ifindex = 0;
1470
1471 #if IS_ENABLED(CONFIG_IPV6)
1472         if (src_ip->sa.sa_family == AF_INET6 &&
1473             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1474                 ifindex = src_ifindex;
1475 #endif
1476
1477         f = vxlan_find_mac(vxlan, src_mac, vni);
1478         if (likely(f)) {
1479                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1480
1481                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1482                            rdst->remote_ifindex == ifindex))
1483                         return false;
1484
1485                 /* Don't migrate static entries, drop packets */
1486                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1487                         return true;
1488
1489                 /* Don't override an fdb with nexthop with a learnt entry */
1490                 if (rcu_access_pointer(f->nh))
1491                         return true;
1492
1493                 if (net_ratelimit())
1494                         netdev_info(dev,
1495                                     "%pM migrated from %pIS to %pIS\n",
1496                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1497
1498                 rdst->remote_ip = *src_ip;
1499                 f->updated = jiffies;
1500                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL);
1501         } else {
1502                 u32 hash_index = fdb_head_index(vxlan, src_mac, vni);
1503
1504                 /* learned new entry */
1505                 spin_lock(&vxlan->hash_lock[hash_index]);
1506
1507                 /* close off race between vxlan_flush and incoming packets */
1508                 if (netif_running(dev))
1509                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1510                                          NUD_REACHABLE,
1511                                          NLM_F_EXCL|NLM_F_CREATE,
1512                                          vxlan->cfg.dst_port,
1513                                          vni,
1514                                          vxlan->default_dst.remote_vni,
1515                                          ifindex, NTF_SELF, 0, true, NULL);
1516                 spin_unlock(&vxlan->hash_lock[hash_index]);
1517         }
1518
1519         return false;
1520 }
1521
1522 /* See if multicast group is already in use by other ID */
1523 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1524 {
1525         struct vxlan_dev *vxlan;
1526         struct vxlan_sock *sock4;
1527 #if IS_ENABLED(CONFIG_IPV6)
1528         struct vxlan_sock *sock6;
1529 #endif
1530         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1531
1532         sock4 = rtnl_dereference(dev->vn4_sock);
1533
1534         /* The vxlan_sock is only used by dev, leaving group has
1535          * no effect on other vxlan devices.
1536          */
1537         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1538                 return false;
1539 #if IS_ENABLED(CONFIG_IPV6)
1540         sock6 = rtnl_dereference(dev->vn6_sock);
1541         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1542                 return false;
1543 #endif
1544
1545         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1546                 if (!netif_running(vxlan->dev) || vxlan == dev)
1547                         continue;
1548
1549                 if (family == AF_INET &&
1550                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1551                         continue;
1552 #if IS_ENABLED(CONFIG_IPV6)
1553                 if (family == AF_INET6 &&
1554                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1555                         continue;
1556 #endif
1557
1558                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1559                                       &dev->default_dst.remote_ip))
1560                         continue;
1561
1562                 if (vxlan->default_dst.remote_ifindex !=
1563                     dev->default_dst.remote_ifindex)
1564                         continue;
1565
1566                 return true;
1567         }
1568
1569         return false;
1570 }
1571
1572 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1573 {
1574         struct vxlan_net *vn;
1575
1576         if (!vs)
1577                 return false;
1578         if (!refcount_dec_and_test(&vs->refcnt))
1579                 return false;
1580
1581         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1582         spin_lock(&vn->sock_lock);
1583         hlist_del_rcu(&vs->hlist);
1584         udp_tunnel_notify_del_rx_port(vs->sock,
1585                                       (vs->flags & VXLAN_F_GPE) ?
1586                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1587                                       UDP_TUNNEL_TYPE_VXLAN);
1588         spin_unlock(&vn->sock_lock);
1589
1590         return true;
1591 }
1592
1593 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1594 {
1595         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1596 #if IS_ENABLED(CONFIG_IPV6)
1597         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1598
1599         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1600 #endif
1601
1602         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1603         synchronize_net();
1604
1605         vxlan_vs_del_dev(vxlan);
1606
1607         if (__vxlan_sock_release_prep(sock4)) {
1608                 udp_tunnel_sock_release(sock4->sock);
1609                 kfree(sock4);
1610         }
1611
1612 #if IS_ENABLED(CONFIG_IPV6)
1613         if (__vxlan_sock_release_prep(sock6)) {
1614                 udp_tunnel_sock_release(sock6->sock);
1615                 kfree(sock6);
1616         }
1617 #endif
1618 }
1619
1620 /* Update multicast group membership when first VNI on
1621  * multicast address is brought up
1622  */
1623 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1624 {
1625         struct sock *sk;
1626         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1627         int ifindex = vxlan->default_dst.remote_ifindex;
1628         int ret = -EINVAL;
1629
1630         if (ip->sa.sa_family == AF_INET) {
1631                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1632                 struct ip_mreqn mreq = {
1633                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1634                         .imr_ifindex            = ifindex,
1635                 };
1636
1637                 sk = sock4->sock->sk;
1638                 lock_sock(sk);
1639                 ret = ip_mc_join_group(sk, &mreq);
1640                 release_sock(sk);
1641 #if IS_ENABLED(CONFIG_IPV6)
1642         } else {
1643                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1644
1645                 sk = sock6->sock->sk;
1646                 lock_sock(sk);
1647                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1648                                                    &ip->sin6.sin6_addr);
1649                 release_sock(sk);
1650 #endif
1651         }
1652
1653         return ret;
1654 }
1655
1656 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1657 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1658 {
1659         struct sock *sk;
1660         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1661         int ifindex = vxlan->default_dst.remote_ifindex;
1662         int ret = -EINVAL;
1663
1664         if (ip->sa.sa_family == AF_INET) {
1665                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1666                 struct ip_mreqn mreq = {
1667                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1668                         .imr_ifindex            = ifindex,
1669                 };
1670
1671                 sk = sock4->sock->sk;
1672                 lock_sock(sk);
1673                 ret = ip_mc_leave_group(sk, &mreq);
1674                 release_sock(sk);
1675 #if IS_ENABLED(CONFIG_IPV6)
1676         } else {
1677                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1678
1679                 sk = sock6->sock->sk;
1680                 lock_sock(sk);
1681                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1682                                                    &ip->sin6.sin6_addr);
1683                 release_sock(sk);
1684 #endif
1685         }
1686
1687         return ret;
1688 }
1689
1690 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1691                           struct sk_buff *skb, u32 vxflags)
1692 {
1693         size_t start, offset;
1694
1695         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1696                 goto out;
1697
1698         start = vxlan_rco_start(unparsed->vx_vni);
1699         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1700
1701         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1702                 return false;
1703
1704         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1705                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1706 out:
1707         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1708         unparsed->vx_vni &= VXLAN_VNI_MASK;
1709         return true;
1710 }
1711
1712 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1713                                 struct sk_buff *skb, u32 vxflags,
1714                                 struct vxlan_metadata *md)
1715 {
1716         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1717         struct metadata_dst *tun_dst;
1718
1719         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1720                 goto out;
1721
1722         md->gbp = ntohs(gbp->policy_id);
1723
1724         tun_dst = (struct metadata_dst *)skb_dst(skb);
1725         if (tun_dst) {
1726                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1727                 tun_dst->u.tun_info.options_len = sizeof(*md);
1728         }
1729         if (gbp->dont_learn)
1730                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1731
1732         if (gbp->policy_applied)
1733                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1734
1735         /* In flow-based mode, GBP is carried in dst_metadata */
1736         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1737                 skb->mark = md->gbp;
1738 out:
1739         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1740 }
1741
1742 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1743                                 __be16 *protocol,
1744                                 struct sk_buff *skb, u32 vxflags)
1745 {
1746         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1747
1748         /* Need to have Next Protocol set for interfaces in GPE mode. */
1749         if (!gpe->np_applied)
1750                 return false;
1751         /* "The initial version is 0. If a receiver does not support the
1752          * version indicated it MUST drop the packet.
1753          */
1754         if (gpe->version != 0)
1755                 return false;
1756         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1757          * processing MUST occur." However, we don't implement OAM
1758          * processing, thus drop the packet.
1759          */
1760         if (gpe->oam_flag)
1761                 return false;
1762
1763         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1764         if (!*protocol)
1765                 return false;
1766
1767         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1768         return true;
1769 }
1770
1771 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1772                           struct vxlan_sock *vs,
1773                           struct sk_buff *skb, __be32 vni)
1774 {
1775         union vxlan_addr saddr;
1776         u32 ifindex = skb->dev->ifindex;
1777
1778         skb_reset_mac_header(skb);
1779         skb->protocol = eth_type_trans(skb, vxlan->dev);
1780         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1781
1782         /* Ignore packet loops (and multicast echo) */
1783         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1784                 return false;
1785
1786         /* Get address from the outer IP header */
1787         if (vxlan_get_sk_family(vs) == AF_INET) {
1788                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1789                 saddr.sa.sa_family = AF_INET;
1790 #if IS_ENABLED(CONFIG_IPV6)
1791         } else {
1792                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1793                 saddr.sa.sa_family = AF_INET6;
1794 #endif
1795         }
1796
1797         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1798             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1799                 return false;
1800
1801         return true;
1802 }
1803
1804 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1805                                   struct sk_buff *skb)
1806 {
1807         int err = 0;
1808
1809         if (vxlan_get_sk_family(vs) == AF_INET)
1810                 err = IP_ECN_decapsulate(oiph, skb);
1811 #if IS_ENABLED(CONFIG_IPV6)
1812         else
1813                 err = IP6_ECN_decapsulate(oiph, skb);
1814 #endif
1815
1816         if (unlikely(err) && log_ecn_error) {
1817                 if (vxlan_get_sk_family(vs) == AF_INET)
1818                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1819                                              &((struct iphdr *)oiph)->saddr,
1820                                              ((struct iphdr *)oiph)->tos);
1821                 else
1822                         net_info_ratelimited("non-ECT from %pI6\n",
1823                                              &((struct ipv6hdr *)oiph)->saddr);
1824         }
1825         return err <= 1;
1826 }
1827
1828 /* Callback from net/ipv4/udp.c to receive packets */
1829 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1830 {
1831         struct vxlan_dev *vxlan;
1832         struct vxlan_sock *vs;
1833         struct vxlanhdr unparsed;
1834         struct vxlan_metadata _md;
1835         struct vxlan_metadata *md = &_md;
1836         __be16 protocol = htons(ETH_P_TEB);
1837         bool raw_proto = false;
1838         void *oiph;
1839         __be32 vni = 0;
1840
1841         /* Need UDP and VXLAN header to be present */
1842         if (!pskb_may_pull(skb, VXLAN_HLEN))
1843                 goto drop;
1844
1845         unparsed = *vxlan_hdr(skb);
1846         /* VNI flag always required to be set */
1847         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1848                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1849                            ntohl(vxlan_hdr(skb)->vx_flags),
1850                            ntohl(vxlan_hdr(skb)->vx_vni));
1851                 /* Return non vxlan pkt */
1852                 goto drop;
1853         }
1854         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1855         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1856
1857         vs = rcu_dereference_sk_user_data(sk);
1858         if (!vs)
1859                 goto drop;
1860
1861         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1862
1863         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1864         if (!vxlan)
1865                 goto drop;
1866
1867         /* For backwards compatibility, only allow reserved fields to be
1868          * used by VXLAN extensions if explicitly requested.
1869          */
1870         if (vs->flags & VXLAN_F_GPE) {
1871                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1872                         goto drop;
1873                 raw_proto = true;
1874         }
1875
1876         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1877                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1878                 goto drop;
1879
1880         if (vs->flags & VXLAN_F_REMCSUM_RX)
1881                 if (unlikely(!vxlan_remcsum(&unparsed, skb, vs->flags)))
1882                         goto drop;
1883
1884         if (vxlan_collect_metadata(vs)) {
1885                 struct metadata_dst *tun_dst;
1886
1887                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1888                                          key32_to_tunnel_id(vni), sizeof(*md));
1889
1890                 if (!tun_dst)
1891                         goto drop;
1892
1893                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1894
1895                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1896         } else {
1897                 memset(md, 0, sizeof(*md));
1898         }
1899
1900         if (vs->flags & VXLAN_F_GBP)
1901                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1902         /* Note that GBP and GPE can never be active together. This is
1903          * ensured in vxlan_dev_configure.
1904          */
1905
1906         if (unparsed.vx_flags || unparsed.vx_vni) {
1907                 /* If there are any unprocessed flags remaining treat
1908                  * this as a malformed packet. This behavior diverges from
1909                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1910                  * in reserved fields are to be ignored. The approach here
1911                  * maintains compatibility with previous stack code, and also
1912                  * is more robust and provides a little more security in
1913                  * adding extensions to VXLAN.
1914                  */
1915                 goto drop;
1916         }
1917
1918         if (!raw_proto) {
1919                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1920                         goto drop;
1921         } else {
1922                 skb_reset_mac_header(skb);
1923                 skb->dev = vxlan->dev;
1924                 skb->pkt_type = PACKET_HOST;
1925         }
1926
1927         oiph = skb_network_header(skb);
1928         skb_reset_network_header(skb);
1929
1930         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1931                 ++vxlan->dev->stats.rx_frame_errors;
1932                 ++vxlan->dev->stats.rx_errors;
1933                 goto drop;
1934         }
1935
1936         rcu_read_lock();
1937
1938         if (unlikely(!(vxlan->dev->flags & IFF_UP))) {
1939                 rcu_read_unlock();
1940                 atomic_long_inc(&vxlan->dev->rx_dropped);
1941                 goto drop;
1942         }
1943
1944         dev_sw_netstats_rx_add(vxlan->dev, skb->len);
1945         gro_cells_receive(&vxlan->gro_cells, skb);
1946
1947         rcu_read_unlock();
1948
1949         return 0;
1950
1951 drop:
1952         /* Consume bad packet */
1953         kfree_skb(skb);
1954         return 0;
1955 }
1956
1957 /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */
1958 static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb)
1959 {
1960         struct vxlan_dev *vxlan;
1961         struct vxlan_sock *vs;
1962         struct vxlanhdr *hdr;
1963         __be32 vni;
1964
1965         if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN))
1966                 return -EINVAL;
1967
1968         hdr = vxlan_hdr(skb);
1969
1970         if (!(hdr->vx_flags & VXLAN_HF_VNI))
1971                 return -EINVAL;
1972
1973         vs = rcu_dereference_sk_user_data(sk);
1974         if (!vs)
1975                 return -ENOENT;
1976
1977         vni = vxlan_vni(hdr->vx_vni);
1978         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1979         if (!vxlan)
1980                 return -ENOENT;
1981
1982         return 0;
1983 }
1984
1985 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1986 {
1987         struct vxlan_dev *vxlan = netdev_priv(dev);
1988         struct arphdr *parp;
1989         u8 *arpptr, *sha;
1990         __be32 sip, tip;
1991         struct neighbour *n;
1992
1993         if (dev->flags & IFF_NOARP)
1994                 goto out;
1995
1996         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1997                 dev->stats.tx_dropped++;
1998                 goto out;
1999         }
2000         parp = arp_hdr(skb);
2001
2002         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
2003              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
2004             parp->ar_pro != htons(ETH_P_IP) ||
2005             parp->ar_op != htons(ARPOP_REQUEST) ||
2006             parp->ar_hln != dev->addr_len ||
2007             parp->ar_pln != 4)
2008                 goto out;
2009         arpptr = (u8 *)parp + sizeof(struct arphdr);
2010         sha = arpptr;
2011         arpptr += dev->addr_len;        /* sha */
2012         memcpy(&sip, arpptr, sizeof(sip));
2013         arpptr += sizeof(sip);
2014         arpptr += dev->addr_len;        /* tha */
2015         memcpy(&tip, arpptr, sizeof(tip));
2016
2017         if (ipv4_is_loopback(tip) ||
2018             ipv4_is_multicast(tip))
2019                 goto out;
2020
2021         n = neigh_lookup(&arp_tbl, &tip, dev);
2022
2023         if (n) {
2024                 struct vxlan_fdb *f;
2025                 struct sk_buff  *reply;
2026
2027                 if (!(n->nud_state & NUD_CONNECTED)) {
2028                         neigh_release(n);
2029                         goto out;
2030                 }
2031
2032                 f = vxlan_find_mac(vxlan, n->ha, vni);
2033                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2034                         /* bridge-local neighbor */
2035                         neigh_release(n);
2036                         goto out;
2037                 }
2038
2039                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
2040                                 n->ha, sha);
2041
2042                 neigh_release(n);
2043
2044                 if (reply == NULL)
2045                         goto out;
2046
2047                 skb_reset_mac_header(reply);
2048                 __skb_pull(reply, skb_network_offset(reply));
2049                 reply->ip_summed = CHECKSUM_UNNECESSARY;
2050                 reply->pkt_type = PACKET_HOST;
2051
2052                 if (netif_rx_ni(reply) == NET_RX_DROP)
2053                         dev->stats.rx_dropped++;
2054         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2055                 union vxlan_addr ipa = {
2056                         .sin.sin_addr.s_addr = tip,
2057                         .sin.sin_family = AF_INET,
2058                 };
2059
2060                 vxlan_ip_miss(dev, &ipa);
2061         }
2062 out:
2063         consume_skb(skb);
2064         return NETDEV_TX_OK;
2065 }
2066
2067 #if IS_ENABLED(CONFIG_IPV6)
2068 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
2069         struct neighbour *n, bool isrouter)
2070 {
2071         struct net_device *dev = request->dev;
2072         struct sk_buff *reply;
2073         struct nd_msg *ns, *na;
2074         struct ipv6hdr *pip6;
2075         u8 *daddr;
2076         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
2077         int ns_olen;
2078         int i, len;
2079
2080         if (dev == NULL || !pskb_may_pull(request, request->len))
2081                 return NULL;
2082
2083         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
2084                 sizeof(*na) + na_olen + dev->needed_tailroom;
2085         reply = alloc_skb(len, GFP_ATOMIC);
2086         if (reply == NULL)
2087                 return NULL;
2088
2089         reply->protocol = htons(ETH_P_IPV6);
2090         reply->dev = dev;
2091         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
2092         skb_push(reply, sizeof(struct ethhdr));
2093         skb_reset_mac_header(reply);
2094
2095         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
2096
2097         daddr = eth_hdr(request)->h_source;
2098         ns_olen = request->len - skb_network_offset(request) -
2099                 sizeof(struct ipv6hdr) - sizeof(*ns);
2100         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
2101                 if (!ns->opt[i + 1]) {
2102                         kfree_skb(reply);
2103                         return NULL;
2104                 }
2105                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
2106                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
2107                         break;
2108                 }
2109         }
2110
2111         /* Ethernet header */
2112         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
2113         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
2114         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
2115         reply->protocol = htons(ETH_P_IPV6);
2116
2117         skb_pull(reply, sizeof(struct ethhdr));
2118         skb_reset_network_header(reply);
2119         skb_put(reply, sizeof(struct ipv6hdr));
2120
2121         /* IPv6 header */
2122
2123         pip6 = ipv6_hdr(reply);
2124         memset(pip6, 0, sizeof(struct ipv6hdr));
2125         pip6->version = 6;
2126         pip6->priority = ipv6_hdr(request)->priority;
2127         pip6->nexthdr = IPPROTO_ICMPV6;
2128         pip6->hop_limit = 255;
2129         pip6->daddr = ipv6_hdr(request)->saddr;
2130         pip6->saddr = *(struct in6_addr *)n->primary_key;
2131
2132         skb_pull(reply, sizeof(struct ipv6hdr));
2133         skb_reset_transport_header(reply);
2134
2135         /* Neighbor Advertisement */
2136         na = skb_put_zero(reply, sizeof(*na) + na_olen);
2137         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
2138         na->icmph.icmp6_router = isrouter;
2139         na->icmph.icmp6_override = 1;
2140         na->icmph.icmp6_solicited = 1;
2141         na->target = ns->target;
2142         ether_addr_copy(&na->opt[2], n->ha);
2143         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
2144         na->opt[1] = na_olen >> 3;
2145
2146         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
2147                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
2148                 csum_partial(na, sizeof(*na)+na_olen, 0));
2149
2150         pip6->payload_len = htons(sizeof(*na)+na_olen);
2151
2152         skb_push(reply, sizeof(struct ipv6hdr));
2153
2154         reply->ip_summed = CHECKSUM_UNNECESSARY;
2155
2156         return reply;
2157 }
2158
2159 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
2160 {
2161         struct vxlan_dev *vxlan = netdev_priv(dev);
2162         const struct in6_addr *daddr;
2163         const struct ipv6hdr *iphdr;
2164         struct inet6_dev *in6_dev;
2165         struct neighbour *n;
2166         struct nd_msg *msg;
2167
2168         rcu_read_lock();
2169         in6_dev = __in6_dev_get(dev);
2170         if (!in6_dev)
2171                 goto out;
2172
2173         iphdr = ipv6_hdr(skb);
2174         daddr = &iphdr->daddr;
2175         msg = (struct nd_msg *)(iphdr + 1);
2176
2177         if (ipv6_addr_loopback(daddr) ||
2178             ipv6_addr_is_multicast(&msg->target))
2179                 goto out;
2180
2181         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
2182
2183         if (n) {
2184                 struct vxlan_fdb *f;
2185                 struct sk_buff *reply;
2186
2187                 if (!(n->nud_state & NUD_CONNECTED)) {
2188                         neigh_release(n);
2189                         goto out;
2190                 }
2191
2192                 f = vxlan_find_mac(vxlan, n->ha, vni);
2193                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2194                         /* bridge-local neighbor */
2195                         neigh_release(n);
2196                         goto out;
2197                 }
2198
2199                 reply = vxlan_na_create(skb, n,
2200                                         !!(f ? f->flags & NTF_ROUTER : 0));
2201
2202                 neigh_release(n);
2203
2204                 if (reply == NULL)
2205                         goto out;
2206
2207                 if (netif_rx_ni(reply) == NET_RX_DROP)
2208                         dev->stats.rx_dropped++;
2209
2210         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2211                 union vxlan_addr ipa = {
2212                         .sin6.sin6_addr = msg->target,
2213                         .sin6.sin6_family = AF_INET6,
2214                 };
2215
2216                 vxlan_ip_miss(dev, &ipa);
2217         }
2218
2219 out:
2220         rcu_read_unlock();
2221         consume_skb(skb);
2222         return NETDEV_TX_OK;
2223 }
2224 #endif
2225
2226 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
2227 {
2228         struct vxlan_dev *vxlan = netdev_priv(dev);
2229         struct neighbour *n;
2230
2231         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
2232                 return false;
2233
2234         n = NULL;
2235         switch (ntohs(eth_hdr(skb)->h_proto)) {
2236         case ETH_P_IP:
2237         {
2238                 struct iphdr *pip;
2239
2240                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
2241                         return false;
2242                 pip = ip_hdr(skb);
2243                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
2244                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2245                         union vxlan_addr ipa = {
2246                                 .sin.sin_addr.s_addr = pip->daddr,
2247                                 .sin.sin_family = AF_INET,
2248                         };
2249
2250                         vxlan_ip_miss(dev, &ipa);
2251                         return false;
2252                 }
2253
2254                 break;
2255         }
2256 #if IS_ENABLED(CONFIG_IPV6)
2257         case ETH_P_IPV6:
2258         {
2259                 struct ipv6hdr *pip6;
2260
2261                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
2262                         return false;
2263                 pip6 = ipv6_hdr(skb);
2264                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
2265                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2266                         union vxlan_addr ipa = {
2267                                 .sin6.sin6_addr = pip6->daddr,
2268                                 .sin6.sin6_family = AF_INET6,
2269                         };
2270
2271                         vxlan_ip_miss(dev, &ipa);
2272                         return false;
2273                 }
2274
2275                 break;
2276         }
2277 #endif
2278         default:
2279                 return false;
2280         }
2281
2282         if (n) {
2283                 bool diff;
2284
2285                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
2286                 if (diff) {
2287                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
2288                                 dev->addr_len);
2289                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
2290                 }
2291                 neigh_release(n);
2292                 return diff;
2293         }
2294
2295         return false;
2296 }
2297
2298 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
2299                                 struct vxlan_metadata *md)
2300 {
2301         struct vxlanhdr_gbp *gbp;
2302
2303         if (!md->gbp)
2304                 return;
2305
2306         gbp = (struct vxlanhdr_gbp *)vxh;
2307         vxh->vx_flags |= VXLAN_HF_GBP;
2308
2309         if (md->gbp & VXLAN_GBP_DONT_LEARN)
2310                 gbp->dont_learn = 1;
2311
2312         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
2313                 gbp->policy_applied = 1;
2314
2315         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
2316 }
2317
2318 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
2319                                __be16 protocol)
2320 {
2321         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
2322
2323         gpe->np_applied = 1;
2324         gpe->next_protocol = tun_p_from_eth_p(protocol);
2325         if (!gpe->next_protocol)
2326                 return -EPFNOSUPPORT;
2327         return 0;
2328 }
2329
2330 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
2331                            int iphdr_len, __be32 vni,
2332                            struct vxlan_metadata *md, u32 vxflags,
2333                            bool udp_sum)
2334 {
2335         struct vxlanhdr *vxh;
2336         int min_headroom;
2337         int err;
2338         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
2339         __be16 inner_protocol = htons(ETH_P_TEB);
2340
2341         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
2342             skb->ip_summed == CHECKSUM_PARTIAL) {
2343                 int csum_start = skb_checksum_start_offset(skb);
2344
2345                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
2346                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
2347                     (skb->csum_offset == offsetof(struct udphdr, check) ||
2348                      skb->csum_offset == offsetof(struct tcphdr, check)))
2349                         type |= SKB_GSO_TUNNEL_REMCSUM;
2350         }
2351
2352         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
2353                         + VXLAN_HLEN + iphdr_len;
2354
2355         /* Need space for new headers (invalidates iph ptr) */
2356         err = skb_cow_head(skb, min_headroom);
2357         if (unlikely(err))
2358                 return err;
2359
2360         err = iptunnel_handle_offloads(skb, type);
2361         if (err)
2362                 return err;
2363
2364         vxh = __skb_push(skb, sizeof(*vxh));
2365         vxh->vx_flags = VXLAN_HF_VNI;
2366         vxh->vx_vni = vxlan_vni_field(vni);
2367
2368         if (type & SKB_GSO_TUNNEL_REMCSUM) {
2369                 unsigned int start;
2370
2371                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
2372                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
2373                 vxh->vx_flags |= VXLAN_HF_RCO;
2374
2375                 if (!skb_is_gso(skb)) {
2376                         skb->ip_summed = CHECKSUM_NONE;
2377                         skb->encapsulation = 0;
2378                 }
2379         }
2380
2381         if (vxflags & VXLAN_F_GBP)
2382                 vxlan_build_gbp_hdr(vxh, vxflags, md);
2383         if (vxflags & VXLAN_F_GPE) {
2384                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
2385                 if (err < 0)
2386                         return err;
2387                 inner_protocol = skb->protocol;
2388         }
2389
2390         skb_set_inner_protocol(skb, inner_protocol);
2391         return 0;
2392 }
2393
2394 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
2395                                       struct vxlan_sock *sock4,
2396                                       struct sk_buff *skb, int oif, u8 tos,
2397                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
2398                                       struct dst_cache *dst_cache,
2399                                       const struct ip_tunnel_info *info)
2400 {
2401         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2402         struct rtable *rt = NULL;
2403         struct flowi4 fl4;
2404
2405         if (!sock4)
2406                 return ERR_PTR(-EIO);
2407
2408         if (tos && !info)
2409                 use_cache = false;
2410         if (use_cache) {
2411                 rt = dst_cache_get_ip4(dst_cache, saddr);
2412                 if (rt)
2413                         return rt;
2414         }
2415
2416         memset(&fl4, 0, sizeof(fl4));
2417         fl4.flowi4_oif = oif;
2418         fl4.flowi4_tos = RT_TOS(tos);
2419         fl4.flowi4_mark = skb->mark;
2420         fl4.flowi4_proto = IPPROTO_UDP;
2421         fl4.daddr = daddr;
2422         fl4.saddr = *saddr;
2423         fl4.fl4_dport = dport;
2424         fl4.fl4_sport = sport;
2425
2426         rt = ip_route_output_key(vxlan->net, &fl4);
2427         if (!IS_ERR(rt)) {
2428                 if (rt->dst.dev == dev) {
2429                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
2430                         ip_rt_put(rt);
2431                         return ERR_PTR(-ELOOP);
2432                 }
2433
2434                 *saddr = fl4.saddr;
2435                 if (use_cache)
2436                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2437         } else {
2438                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
2439                 return ERR_PTR(-ENETUNREACH);
2440         }
2441         return rt;
2442 }
2443
2444 #if IS_ENABLED(CONFIG_IPV6)
2445 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
2446                                           struct net_device *dev,
2447                                           struct vxlan_sock *sock6,
2448                                           struct sk_buff *skb, int oif, u8 tos,
2449                                           __be32 label,
2450                                           const struct in6_addr *daddr,
2451                                           struct in6_addr *saddr,
2452                                           __be16 dport, __be16 sport,
2453                                           struct dst_cache *dst_cache,
2454                                           const struct ip_tunnel_info *info)
2455 {
2456         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2457         struct dst_entry *ndst;
2458         struct flowi6 fl6;
2459
2460         if (!sock6)
2461                 return ERR_PTR(-EIO);
2462
2463         if (tos && !info)
2464                 use_cache = false;
2465         if (use_cache) {
2466                 ndst = dst_cache_get_ip6(dst_cache, saddr);
2467                 if (ndst)
2468                         return ndst;
2469         }
2470
2471         memset(&fl6, 0, sizeof(fl6));
2472         fl6.flowi6_oif = oif;
2473         fl6.daddr = *daddr;
2474         fl6.saddr = *saddr;
2475         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
2476         fl6.flowi6_mark = skb->mark;
2477         fl6.flowi6_proto = IPPROTO_UDP;
2478         fl6.fl6_dport = dport;
2479         fl6.fl6_sport = sport;
2480
2481         ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk,
2482                                                &fl6, NULL);
2483         if (IS_ERR(ndst)) {
2484                 netdev_dbg(dev, "no route to %pI6\n", daddr);
2485                 return ERR_PTR(-ENETUNREACH);
2486         }
2487
2488         if (unlikely(ndst->dev == dev)) {
2489                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2490                 dst_release(ndst);
2491                 return ERR_PTR(-ELOOP);
2492         }
2493
2494         *saddr = fl6.saddr;
2495         if (use_cache)
2496                 dst_cache_set_ip6(dst_cache, ndst, saddr);
2497         return ndst;
2498 }
2499 #endif
2500
2501 /* Bypass encapsulation if the destination is local */
2502 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2503                                struct vxlan_dev *dst_vxlan, __be32 vni,
2504                                bool snoop)
2505 {
2506         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2507         union vxlan_addr loopback;
2508         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2509         struct net_device *dev;
2510         int len = skb->len;
2511
2512         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2513         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2514         skb->pkt_type = PACKET_HOST;
2515         skb->encapsulation = 0;
2516         skb->dev = dst_vxlan->dev;
2517         __skb_pull(skb, skb_network_offset(skb));
2518
2519         if (remote_ip->sa.sa_family == AF_INET) {
2520                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2521                 loopback.sa.sa_family =  AF_INET;
2522 #if IS_ENABLED(CONFIG_IPV6)
2523         } else {
2524                 loopback.sin6.sin6_addr = in6addr_loopback;
2525                 loopback.sa.sa_family =  AF_INET6;
2526 #endif
2527         }
2528
2529         rcu_read_lock();
2530         dev = skb->dev;
2531         if (unlikely(!(dev->flags & IFF_UP))) {
2532                 kfree_skb(skb);
2533                 goto drop;
2534         }
2535
2536         if ((dst_vxlan->cfg.flags & VXLAN_F_LEARN) && snoop)
2537                 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2538
2539         u64_stats_update_begin(&tx_stats->syncp);
2540         tx_stats->tx_packets++;
2541         tx_stats->tx_bytes += len;
2542         u64_stats_update_end(&tx_stats->syncp);
2543
2544         if (netif_rx(skb) == NET_RX_SUCCESS) {
2545                 u64_stats_update_begin(&rx_stats->syncp);
2546                 rx_stats->rx_packets++;
2547                 rx_stats->rx_bytes += len;
2548                 u64_stats_update_end(&rx_stats->syncp);
2549         } else {
2550 drop:
2551                 dev->stats.rx_dropped++;
2552         }
2553         rcu_read_unlock();
2554 }
2555
2556 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2557                                  struct vxlan_dev *vxlan,
2558                                  union vxlan_addr *daddr,
2559                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2560                                  struct dst_entry *dst,
2561                                  u32 rt_flags)
2562 {
2563 #if IS_ENABLED(CONFIG_IPV6)
2564         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2565          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2566          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2567          */
2568         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2569 #endif
2570         /* Bypass encapsulation if the destination is local */
2571         if (rt_flags & RTCF_LOCAL &&
2572             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2573                 struct vxlan_dev *dst_vxlan;
2574
2575                 dst_release(dst);
2576                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2577                                            daddr->sa.sa_family, dst_port,
2578                                            vxlan->cfg.flags);
2579                 if (!dst_vxlan) {
2580                         dev->stats.tx_errors++;
2581                         kfree_skb(skb);
2582
2583                         return -ENOENT;
2584                 }
2585                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni, true);
2586                 return 1;
2587         }
2588
2589         return 0;
2590 }
2591
2592 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2593                            __be32 default_vni, struct vxlan_rdst *rdst,
2594                            bool did_rsc)
2595 {
2596         struct dst_cache *dst_cache;
2597         struct ip_tunnel_info *info;
2598         struct vxlan_dev *vxlan = netdev_priv(dev);
2599         const struct iphdr *old_iph = ip_hdr(skb);
2600         union vxlan_addr *dst;
2601         union vxlan_addr remote_ip, local_ip;
2602         struct vxlan_metadata _md;
2603         struct vxlan_metadata *md = &_md;
2604         __be16 src_port = 0, dst_port;
2605         struct dst_entry *ndst = NULL;
2606         __be32 vni, label;
2607         __u8 tos, ttl;
2608         int ifindex;
2609         int err;
2610         u32 flags = vxlan->cfg.flags;
2611         bool udp_sum = false;
2612         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2613
2614         info = skb_tunnel_info(skb);
2615
2616         if (rdst) {
2617                 dst = &rdst->remote_ip;
2618                 if (vxlan_addr_any(dst)) {
2619                         if (did_rsc) {
2620                                 /* short-circuited back to local bridge */
2621                                 vxlan_encap_bypass(skb, vxlan, vxlan,
2622                                                    default_vni, true);
2623                                 return;
2624                         }
2625                         goto drop;
2626                 }
2627
2628                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2629                 vni = (rdst->remote_vni) ? : default_vni;
2630                 ifindex = rdst->remote_ifindex;
2631                 local_ip = vxlan->cfg.saddr;
2632                 dst_cache = &rdst->dst_cache;
2633                 md->gbp = skb->mark;
2634                 if (flags & VXLAN_F_TTL_INHERIT) {
2635                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2636                 } else {
2637                         ttl = vxlan->cfg.ttl;
2638                         if (!ttl && vxlan_addr_multicast(dst))
2639                                 ttl = 1;
2640                 }
2641
2642                 tos = vxlan->cfg.tos;
2643                 if (tos == 1)
2644                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2645
2646                 if (dst->sa.sa_family == AF_INET)
2647                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2648                 else
2649                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2650                 label = vxlan->cfg.label;
2651         } else {
2652                 if (!info) {
2653                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2654                                   dev->name);
2655                         goto drop;
2656                 }
2657                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2658                 if (remote_ip.sa.sa_family == AF_INET) {
2659                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2660                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2661                 } else {
2662                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2663                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2664                 }
2665                 dst = &remote_ip;
2666                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2667                 vni = tunnel_id_to_key32(info->key.tun_id);
2668                 ifindex = 0;
2669                 dst_cache = &info->dst_cache;
2670                 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) {
2671                         if (info->options_len < sizeof(*md))
2672                                 goto drop;
2673                         md = ip_tunnel_info_opts(info);
2674                 }
2675                 ttl = info->key.ttl;
2676                 tos = info->key.tos;
2677                 label = info->key.label;
2678                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2679         }
2680         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2681                                      vxlan->cfg.port_max, true);
2682
2683         rcu_read_lock();
2684         if (dst->sa.sa_family == AF_INET) {
2685                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2686                 struct rtable *rt;
2687                 __be16 df = 0;
2688
2689                 if (!ifindex)
2690                         ifindex = sock4->sock->sk->sk_bound_dev_if;
2691
2692                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2693                                      dst->sin.sin_addr.s_addr,
2694                                      &local_ip.sin.sin_addr.s_addr,
2695                                      dst_port, src_port,
2696                                      dst_cache, info);
2697                 if (IS_ERR(rt)) {
2698                         err = PTR_ERR(rt);
2699                         goto tx_error;
2700                 }
2701
2702                 if (!info) {
2703                         /* Bypass encapsulation if the destination is local */
2704                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2705                                                     dst_port, ifindex, vni,
2706                                                     &rt->dst, rt->rt_flags);
2707                         if (err)
2708                                 goto out_unlock;
2709
2710                         if (vxlan->cfg.df == VXLAN_DF_SET) {
2711                                 df = htons(IP_DF);
2712                         } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) {
2713                                 struct ethhdr *eth = eth_hdr(skb);
2714
2715                                 if (ntohs(eth->h_proto) == ETH_P_IPV6 ||
2716                                     (ntohs(eth->h_proto) == ETH_P_IP &&
2717                                      old_iph->frag_off & htons(IP_DF)))
2718                                         df = htons(IP_DF);
2719                         }
2720                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2721                         df = htons(IP_DF);
2722                 }
2723
2724                 ndst = &rt->dst;
2725                 err = skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM,
2726                                             netif_is_any_bridge_port(dev));
2727                 if (err < 0) {
2728                         goto tx_error;
2729                 } else if (err) {
2730                         if (info) {
2731                                 struct ip_tunnel_info *unclone;
2732                                 struct in_addr src, dst;
2733
2734                                 unclone = skb_tunnel_info_unclone(skb);
2735                                 if (unlikely(!unclone))
2736                                         goto tx_error;
2737
2738                                 src = remote_ip.sin.sin_addr;
2739                                 dst = local_ip.sin.sin_addr;
2740                                 unclone->key.u.ipv4.src = src.s_addr;
2741                                 unclone->key.u.ipv4.dst = dst.s_addr;
2742                         }
2743                         vxlan_encap_bypass(skb, vxlan, vxlan, vni, false);
2744                         dst_release(ndst);
2745                         goto out_unlock;
2746                 }
2747
2748                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2749                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2750                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2751                                       vni, md, flags, udp_sum);
2752                 if (err < 0)
2753                         goto tx_error;
2754
2755                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2756                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2757                                     src_port, dst_port, xnet, !udp_sum);
2758 #if IS_ENABLED(CONFIG_IPV6)
2759         } else {
2760                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2761
2762                 if (!ifindex)
2763                         ifindex = sock6->sock->sk->sk_bound_dev_if;
2764
2765                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2766                                         label, &dst->sin6.sin6_addr,
2767                                         &local_ip.sin6.sin6_addr,
2768                                         dst_port, src_port,
2769                                         dst_cache, info);
2770                 if (IS_ERR(ndst)) {
2771                         err = PTR_ERR(ndst);
2772                         ndst = NULL;
2773                         goto tx_error;
2774                 }
2775
2776                 if (!info) {
2777                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2778
2779                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2780                                                     dst_port, ifindex, vni,
2781                                                     ndst, rt6i_flags);
2782                         if (err)
2783                                 goto out_unlock;
2784                 }
2785
2786                 err = skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM,
2787                                             netif_is_any_bridge_port(dev));
2788                 if (err < 0) {
2789                         goto tx_error;
2790                 } else if (err) {
2791                         if (info) {
2792                                 struct ip_tunnel_info *unclone;
2793                                 struct in6_addr src, dst;
2794
2795                                 unclone = skb_tunnel_info_unclone(skb);
2796                                 if (unlikely(!unclone))
2797                                         goto tx_error;
2798
2799                                 src = remote_ip.sin6.sin6_addr;
2800                                 dst = local_ip.sin6.sin6_addr;
2801                                 unclone->key.u.ipv6.src = src;
2802                                 unclone->key.u.ipv6.dst = dst;
2803                         }
2804
2805                         vxlan_encap_bypass(skb, vxlan, vxlan, vni, false);
2806                         dst_release(ndst);
2807                         goto out_unlock;
2808                 }
2809
2810                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2811                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2812                 skb_scrub_packet(skb, xnet);
2813                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2814                                       vni, md, flags, udp_sum);
2815                 if (err < 0)
2816                         goto tx_error;
2817
2818                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2819                                      &local_ip.sin6.sin6_addr,
2820                                      &dst->sin6.sin6_addr, tos, ttl,
2821                                      label, src_port, dst_port, !udp_sum);
2822 #endif
2823         }
2824 out_unlock:
2825         rcu_read_unlock();
2826         return;
2827
2828 drop:
2829         dev->stats.tx_dropped++;
2830         dev_kfree_skb(skb);
2831         return;
2832
2833 tx_error:
2834         rcu_read_unlock();
2835         if (err == -ELOOP)
2836                 dev->stats.collisions++;
2837         else if (err == -ENETUNREACH)
2838                 dev->stats.tx_carrier_errors++;
2839         dst_release(ndst);
2840         dev->stats.tx_errors++;
2841         kfree_skb(skb);
2842 }
2843
2844 static void vxlan_xmit_nh(struct sk_buff *skb, struct net_device *dev,
2845                           struct vxlan_fdb *f, __be32 vni, bool did_rsc)
2846 {
2847         struct vxlan_rdst nh_rdst;
2848         struct nexthop *nh;
2849         bool do_xmit;
2850         u32 hash;
2851
2852         memset(&nh_rdst, 0, sizeof(struct vxlan_rdst));
2853         hash = skb_get_hash(skb);
2854
2855         rcu_read_lock();
2856         nh = rcu_dereference(f->nh);
2857         if (!nh) {
2858                 rcu_read_unlock();
2859                 goto drop;
2860         }
2861         do_xmit = vxlan_fdb_nh_path_select(nh, hash, &nh_rdst);
2862         rcu_read_unlock();
2863
2864         if (likely(do_xmit))
2865                 vxlan_xmit_one(skb, dev, vni, &nh_rdst, did_rsc);
2866         else
2867                 goto drop;
2868
2869         return;
2870
2871 drop:
2872         dev->stats.tx_dropped++;
2873         dev_kfree_skb(skb);
2874 }
2875
2876 /* Transmit local packets over Vxlan
2877  *
2878  * Outer IP header inherits ECN and DF from inner header.
2879  * Outer UDP destination is the VXLAN assigned port.
2880  *           source port is based on hash of flow
2881  */
2882 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2883 {
2884         struct vxlan_dev *vxlan = netdev_priv(dev);
2885         struct vxlan_rdst *rdst, *fdst = NULL;
2886         const struct ip_tunnel_info *info;
2887         bool did_rsc = false;
2888         struct vxlan_fdb *f;
2889         struct ethhdr *eth;
2890         __be32 vni = 0;
2891
2892         info = skb_tunnel_info(skb);
2893
2894         skb_reset_mac_header(skb);
2895
2896         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2897                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2898                     info->mode & IP_TUNNEL_INFO_TX) {
2899                         vni = tunnel_id_to_key32(info->key.tun_id);
2900                 } else {
2901                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2902                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2903                         else
2904                                 kfree_skb(skb);
2905                         return NETDEV_TX_OK;
2906                 }
2907         }
2908
2909         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2910                 eth = eth_hdr(skb);
2911                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2912                         return arp_reduce(dev, skb, vni);
2913 #if IS_ENABLED(CONFIG_IPV6)
2914                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2915                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2916                                             sizeof(struct nd_msg)) &&
2917                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2918                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2919
2920                         if (m->icmph.icmp6_code == 0 &&
2921                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2922                                 return neigh_reduce(dev, skb, vni);
2923                 }
2924 #endif
2925         }
2926
2927         eth = eth_hdr(skb);
2928         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2929         did_rsc = false;
2930
2931         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2932             (ntohs(eth->h_proto) == ETH_P_IP ||
2933              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2934                 did_rsc = route_shortcircuit(dev, skb);
2935                 if (did_rsc)
2936                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2937         }
2938
2939         if (f == NULL) {
2940                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2941                 if (f == NULL) {
2942                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2943                             !is_multicast_ether_addr(eth->h_dest))
2944                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2945
2946                         dev->stats.tx_dropped++;
2947                         kfree_skb(skb);
2948                         return NETDEV_TX_OK;
2949                 }
2950         }
2951
2952         if (rcu_access_pointer(f->nh)) {
2953                 vxlan_xmit_nh(skb, dev, f,
2954                               (vni ? : vxlan->default_dst.remote_vni), did_rsc);
2955         } else {
2956                 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2957                         struct sk_buff *skb1;
2958
2959                         if (!fdst) {
2960                                 fdst = rdst;
2961                                 continue;
2962                         }
2963                         skb1 = skb_clone(skb, GFP_ATOMIC);
2964                         if (skb1)
2965                                 vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2966                 }
2967                 if (fdst)
2968                         vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2969                 else
2970                         kfree_skb(skb);
2971         }
2972
2973         return NETDEV_TX_OK;
2974 }
2975
2976 /* Walk the forwarding table and purge stale entries */
2977 static void vxlan_cleanup(struct timer_list *t)
2978 {
2979         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2980         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2981         unsigned int h;
2982
2983         if (!netif_running(vxlan->dev))
2984                 return;
2985
2986         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2987                 struct hlist_node *p, *n;
2988
2989                 spin_lock(&vxlan->hash_lock[h]);
2990                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2991                         struct vxlan_fdb *f
2992                                 = container_of(p, struct vxlan_fdb, hlist);
2993                         unsigned long timeout;
2994
2995                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2996                                 continue;
2997
2998                         if (f->flags & NTF_EXT_LEARNED)
2999                                 continue;
3000
3001                         timeout = f->used + vxlan->cfg.age_interval * HZ;
3002                         if (time_before_eq(timeout, jiffies)) {
3003                                 netdev_dbg(vxlan->dev,
3004                                            "garbage collect %pM\n",
3005                                            f->eth_addr);
3006                                 f->state = NUD_STALE;
3007                                 vxlan_fdb_destroy(vxlan, f, true, true);
3008                         } else if (time_before(timeout, next_timer))
3009                                 next_timer = timeout;
3010                 }
3011                 spin_unlock(&vxlan->hash_lock[h]);
3012         }
3013
3014         mod_timer(&vxlan->age_timer, next_timer);
3015 }
3016
3017 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
3018 {
3019         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3020
3021         spin_lock(&vn->sock_lock);
3022         hlist_del_init_rcu(&vxlan->hlist4.hlist);
3023 #if IS_ENABLED(CONFIG_IPV6)
3024         hlist_del_init_rcu(&vxlan->hlist6.hlist);
3025 #endif
3026         spin_unlock(&vn->sock_lock);
3027 }
3028
3029 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
3030                              struct vxlan_dev_node *node)
3031 {
3032         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3033         __be32 vni = vxlan->default_dst.remote_vni;
3034
3035         node->vxlan = vxlan;
3036         spin_lock(&vn->sock_lock);
3037         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
3038         spin_unlock(&vn->sock_lock);
3039 }
3040
3041 /* Setup stats when device is created */
3042 static int vxlan_init(struct net_device *dev)
3043 {
3044         struct vxlan_dev *vxlan = netdev_priv(dev);
3045         int err;
3046
3047         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
3048         if (!dev->tstats)
3049                 return -ENOMEM;
3050
3051         err = gro_cells_init(&vxlan->gro_cells, dev);
3052         if (err) {
3053                 free_percpu(dev->tstats);
3054                 return err;
3055         }
3056
3057         return 0;
3058 }
3059
3060 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
3061 {
3062         struct vxlan_fdb *f;
3063         u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni);
3064
3065         spin_lock_bh(&vxlan->hash_lock[hash_index]);
3066         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
3067         if (f)
3068                 vxlan_fdb_destroy(vxlan, f, true, true);
3069         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
3070 }
3071
3072 static void vxlan_uninit(struct net_device *dev)
3073 {
3074         struct vxlan_dev *vxlan = netdev_priv(dev);
3075
3076         gro_cells_destroy(&vxlan->gro_cells);
3077
3078         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
3079
3080         free_percpu(dev->tstats);
3081 }
3082
3083 /* Start ageing timer and join group when device is brought up */
3084 static int vxlan_open(struct net_device *dev)
3085 {
3086         struct vxlan_dev *vxlan = netdev_priv(dev);
3087         int ret;
3088
3089         ret = vxlan_sock_add(vxlan);
3090         if (ret < 0)
3091                 return ret;
3092
3093         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
3094                 ret = vxlan_igmp_join(vxlan);
3095                 if (ret == -EADDRINUSE)
3096                         ret = 0;
3097                 if (ret) {
3098                         vxlan_sock_release(vxlan);
3099                         return ret;
3100                 }
3101         }
3102
3103         if (vxlan->cfg.age_interval)
3104                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
3105
3106         return ret;
3107 }
3108
3109 /* Purge the forwarding table */
3110 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
3111 {
3112         unsigned int h;
3113
3114         for (h = 0; h < FDB_HASH_SIZE; ++h) {
3115                 struct hlist_node *p, *n;
3116
3117                 spin_lock_bh(&vxlan->hash_lock[h]);
3118                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
3119                         struct vxlan_fdb *f
3120                                 = container_of(p, struct vxlan_fdb, hlist);
3121                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
3122                                 continue;
3123                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
3124                         if (is_zero_ether_addr(f->eth_addr) &&
3125                             f->vni == vxlan->cfg.vni)
3126                                 continue;
3127                         vxlan_fdb_destroy(vxlan, f, true, true);
3128                 }
3129                 spin_unlock_bh(&vxlan->hash_lock[h]);
3130         }
3131 }
3132
3133 /* Cleanup timer and forwarding table on shutdown */
3134 static int vxlan_stop(struct net_device *dev)
3135 {
3136         struct vxlan_dev *vxlan = netdev_priv(dev);
3137         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3138         int ret = 0;
3139
3140         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
3141             !vxlan_group_used(vn, vxlan))
3142                 ret = vxlan_igmp_leave(vxlan);
3143
3144         del_timer_sync(&vxlan->age_timer);
3145
3146         vxlan_flush(vxlan, false);
3147         vxlan_sock_release(vxlan);
3148
3149         return ret;
3150 }
3151
3152 /* Stub, nothing needs to be done. */
3153 static void vxlan_set_multicast_list(struct net_device *dev)
3154 {
3155 }
3156
3157 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
3158 {
3159         struct vxlan_dev *vxlan = netdev_priv(dev);
3160         struct vxlan_rdst *dst = &vxlan->default_dst;
3161         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3162                                                          dst->remote_ifindex);
3163         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
3164
3165         /* This check is different than dev->max_mtu, because it looks at
3166          * the lowerdev->mtu, rather than the static dev->max_mtu
3167          */
3168         if (lowerdev) {
3169                 int max_mtu = lowerdev->mtu -
3170                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
3171                 if (new_mtu > max_mtu)
3172                         return -EINVAL;
3173         }
3174
3175         dev->mtu = new_mtu;
3176         return 0;
3177 }
3178
3179 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
3180 {
3181         struct vxlan_dev *vxlan = netdev_priv(dev);
3182         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3183         __be16 sport, dport;
3184
3185         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
3186                                   vxlan->cfg.port_max, true);
3187         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
3188
3189         if (ip_tunnel_info_af(info) == AF_INET) {
3190                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
3191                 struct rtable *rt;
3192
3193                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
3194                                      info->key.u.ipv4.dst,
3195                                      &info->key.u.ipv4.src, dport, sport,
3196                                      &info->dst_cache, info);
3197                 if (IS_ERR(rt))
3198                         return PTR_ERR(rt);
3199                 ip_rt_put(rt);
3200         } else {
3201 #if IS_ENABLED(CONFIG_IPV6)
3202                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
3203                 struct dst_entry *ndst;
3204
3205                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
3206                                         info->key.label, &info->key.u.ipv6.dst,
3207                                         &info->key.u.ipv6.src, dport, sport,
3208                                         &info->dst_cache, info);
3209                 if (IS_ERR(ndst))
3210                         return PTR_ERR(ndst);
3211                 dst_release(ndst);
3212 #else /* !CONFIG_IPV6 */
3213                 return -EPFNOSUPPORT;
3214 #endif
3215         }
3216         info->key.tp_src = sport;
3217         info->key.tp_dst = dport;
3218         return 0;
3219 }
3220
3221 static const struct net_device_ops vxlan_netdev_ether_ops = {
3222         .ndo_init               = vxlan_init,
3223         .ndo_uninit             = vxlan_uninit,
3224         .ndo_open               = vxlan_open,
3225         .ndo_stop               = vxlan_stop,
3226         .ndo_start_xmit         = vxlan_xmit,
3227         .ndo_get_stats64        = dev_get_tstats64,
3228         .ndo_set_rx_mode        = vxlan_set_multicast_list,
3229         .ndo_change_mtu         = vxlan_change_mtu,
3230         .ndo_validate_addr      = eth_validate_addr,
3231         .ndo_set_mac_address    = eth_mac_addr,
3232         .ndo_fdb_add            = vxlan_fdb_add,
3233         .ndo_fdb_del            = vxlan_fdb_delete,
3234         .ndo_fdb_dump           = vxlan_fdb_dump,
3235         .ndo_fdb_get            = vxlan_fdb_get,
3236         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
3237 };
3238
3239 static const struct net_device_ops vxlan_netdev_raw_ops = {
3240         .ndo_init               = vxlan_init,
3241         .ndo_uninit             = vxlan_uninit,
3242         .ndo_open               = vxlan_open,
3243         .ndo_stop               = vxlan_stop,
3244         .ndo_start_xmit         = vxlan_xmit,
3245         .ndo_get_stats64        = dev_get_tstats64,
3246         .ndo_change_mtu         = vxlan_change_mtu,
3247         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
3248 };
3249
3250 /* Info for udev, that this is a virtual tunnel endpoint */
3251 static struct device_type vxlan_type = {
3252         .name = "vxlan",
3253 };
3254
3255 /* Calls the ndo_udp_tunnel_add of the caller in order to
3256  * supply the listening VXLAN udp ports. Callers are expected
3257  * to implement the ndo_udp_tunnel_add.
3258  */
3259 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
3260 {
3261         struct vxlan_sock *vs;
3262         struct net *net = dev_net(dev);
3263         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3264         unsigned int i;
3265
3266         spin_lock(&vn->sock_lock);
3267         for (i = 0; i < PORT_HASH_SIZE; ++i) {
3268                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
3269                         unsigned short type;
3270
3271                         if (vs->flags & VXLAN_F_GPE)
3272                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
3273                         else
3274                                 type = UDP_TUNNEL_TYPE_VXLAN;
3275
3276                         if (push)
3277                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
3278                         else
3279                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
3280                 }
3281         }
3282         spin_unlock(&vn->sock_lock);
3283 }
3284
3285 /* Initialize the device structure. */
3286 static void vxlan_setup(struct net_device *dev)
3287 {
3288         struct vxlan_dev *vxlan = netdev_priv(dev);
3289         unsigned int h;
3290
3291         eth_hw_addr_random(dev);
3292         ether_setup(dev);
3293
3294         dev->needs_free_netdev = true;
3295         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
3296
3297         dev->features   |= NETIF_F_LLTX;
3298         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
3299         dev->features   |= NETIF_F_RXCSUM;
3300         dev->features   |= NETIF_F_GSO_SOFTWARE;
3301
3302         dev->vlan_features = dev->features;
3303         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
3304         dev->hw_features |= NETIF_F_RXCSUM;
3305         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
3306         netif_keep_dst(dev);
3307         dev->priv_flags |= IFF_NO_QUEUE | IFF_CHANGE_PROTO_DOWN;
3308
3309         /* MTU range: 68 - 65535 */
3310         dev->min_mtu = ETH_MIN_MTU;
3311         dev->max_mtu = ETH_MAX_MTU;
3312
3313         INIT_LIST_HEAD(&vxlan->next);
3314
3315         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
3316
3317         vxlan->dev = dev;
3318
3319         for (h = 0; h < FDB_HASH_SIZE; ++h) {
3320                 spin_lock_init(&vxlan->hash_lock[h]);
3321                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
3322         }
3323 }
3324
3325 static void vxlan_ether_setup(struct net_device *dev)
3326 {
3327         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
3328         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3329         dev->netdev_ops = &vxlan_netdev_ether_ops;
3330 }
3331
3332 static void vxlan_raw_setup(struct net_device *dev)
3333 {
3334         dev->header_ops = NULL;
3335         dev->type = ARPHRD_NONE;
3336         dev->hard_header_len = 0;
3337         dev->addr_len = 0;
3338         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
3339         dev->netdev_ops = &vxlan_netdev_raw_ops;
3340 }
3341
3342 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
3343         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
3344         [IFLA_VXLAN_GROUP]      = { .len = sizeof_field(struct iphdr, daddr) },
3345         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
3346         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
3347         [IFLA_VXLAN_LOCAL]      = { .len = sizeof_field(struct iphdr, saddr) },
3348         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
3349         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
3350         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
3351         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
3352         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
3353         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
3354         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
3355         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
3356         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
3357         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
3358         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
3359         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
3360         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
3361         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
3362         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
3363         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
3364         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
3365         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
3366         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
3367         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
3368         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
3369         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
3370         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
3371         [IFLA_VXLAN_DF]         = { .type = NLA_U8 },
3372 };
3373
3374 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
3375                           struct netlink_ext_ack *extack)
3376 {
3377         if (tb[IFLA_ADDRESS]) {
3378                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
3379                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3380                                             "Provided link layer address is not Ethernet");
3381                         return -EINVAL;
3382                 }
3383
3384                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
3385                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3386                                             "Provided Ethernet address is not unicast");
3387                         return -EADDRNOTAVAIL;
3388                 }
3389         }
3390
3391         if (tb[IFLA_MTU]) {
3392                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
3393
3394                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
3395                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3396                                             "MTU must be between 68 and 65535");
3397                         return -EINVAL;
3398                 }
3399         }
3400
3401         if (!data) {
3402                 NL_SET_ERR_MSG(extack,
3403                                "Required attributes not provided to perform the operation");
3404                 return -EINVAL;
3405         }
3406
3407         if (data[IFLA_VXLAN_ID]) {
3408                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
3409
3410                 if (id >= VXLAN_N_VID) {
3411                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_ID],
3412                                             "VXLAN ID must be lower than 16777216");
3413                         return -ERANGE;
3414                 }
3415         }
3416
3417         if (data[IFLA_VXLAN_PORT_RANGE]) {
3418                 const struct ifla_vxlan_port_range *p
3419                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3420
3421                 if (ntohs(p->high) < ntohs(p->low)) {
3422                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_PORT_RANGE],
3423                                             "Invalid source port range");
3424                         return -EINVAL;
3425                 }
3426         }
3427
3428         if (data[IFLA_VXLAN_DF]) {
3429                 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]);
3430
3431                 if (df < 0 || df > VXLAN_DF_MAX) {
3432                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_DF],
3433                                             "Invalid DF attribute");
3434                         return -EINVAL;
3435                 }
3436         }
3437
3438         return 0;
3439 }
3440
3441 static void vxlan_get_drvinfo(struct net_device *netdev,
3442                               struct ethtool_drvinfo *drvinfo)
3443 {
3444         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
3445         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
3446 }
3447
3448 static int vxlan_get_link_ksettings(struct net_device *dev,
3449                                     struct ethtool_link_ksettings *cmd)
3450 {
3451         struct vxlan_dev *vxlan = netdev_priv(dev);
3452         struct vxlan_rdst *dst = &vxlan->default_dst;
3453         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3454                                                          dst->remote_ifindex);
3455
3456         if (!lowerdev) {
3457                 cmd->base.duplex = DUPLEX_UNKNOWN;
3458                 cmd->base.port = PORT_OTHER;
3459                 cmd->base.speed = SPEED_UNKNOWN;
3460
3461                 return 0;
3462         }
3463
3464         return __ethtool_get_link_ksettings(lowerdev, cmd);
3465 }
3466
3467 static const struct ethtool_ops vxlan_ethtool_ops = {
3468         .get_drvinfo            = vxlan_get_drvinfo,
3469         .get_link               = ethtool_op_get_link,
3470         .get_link_ksettings     = vxlan_get_link_ksettings,
3471 };
3472
3473 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
3474                                         __be16 port, u32 flags, int ifindex)
3475 {
3476         struct socket *sock;
3477         struct udp_port_cfg udp_conf;
3478         int err;
3479
3480         memset(&udp_conf, 0, sizeof(udp_conf));
3481
3482         if (ipv6) {
3483                 udp_conf.family = AF_INET6;
3484                 udp_conf.use_udp6_rx_checksums =
3485                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
3486                 udp_conf.ipv6_v6only = 1;
3487         } else {
3488                 udp_conf.family = AF_INET;
3489         }
3490
3491         udp_conf.local_udp_port = port;
3492         udp_conf.bind_ifindex = ifindex;
3493
3494         /* Open UDP socket */
3495         err = udp_sock_create(net, &udp_conf, &sock);
3496         if (err < 0)
3497                 return ERR_PTR(err);
3498
3499         udp_allow_gso(sock->sk);
3500         return sock;
3501 }
3502
3503 /* Create new listen socket if needed */
3504 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
3505                                               __be16 port, u32 flags,
3506                                               int ifindex)
3507 {
3508         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3509         struct vxlan_sock *vs;
3510         struct socket *sock;
3511         unsigned int h;
3512         struct udp_tunnel_sock_cfg tunnel_cfg;
3513
3514         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
3515         if (!vs)
3516                 return ERR_PTR(-ENOMEM);
3517
3518         for (h = 0; h < VNI_HASH_SIZE; ++h)
3519                 INIT_HLIST_HEAD(&vs->vni_list[h]);
3520
3521         sock = vxlan_create_sock(net, ipv6, port, flags, ifindex);
3522         if (IS_ERR(sock)) {
3523                 kfree(vs);
3524                 return ERR_CAST(sock);
3525         }
3526
3527         vs->sock = sock;
3528         refcount_set(&vs->refcnt, 1);
3529         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
3530
3531         spin_lock(&vn->sock_lock);
3532         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
3533         udp_tunnel_notify_add_rx_port(sock,
3534                                       (vs->flags & VXLAN_F_GPE) ?
3535                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
3536                                       UDP_TUNNEL_TYPE_VXLAN);
3537         spin_unlock(&vn->sock_lock);
3538
3539         /* Mark socket as an encapsulation socket. */
3540         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
3541         tunnel_cfg.sk_user_data = vs;
3542         tunnel_cfg.encap_type = 1;
3543         tunnel_cfg.encap_rcv = vxlan_rcv;
3544         tunnel_cfg.encap_err_lookup = vxlan_err_lookup;
3545         tunnel_cfg.encap_destroy = NULL;
3546         tunnel_cfg.gro_receive = vxlan_gro_receive;
3547         tunnel_cfg.gro_complete = vxlan_gro_complete;
3548
3549         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
3550
3551         return vs;
3552 }
3553
3554 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
3555 {
3556         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3557         struct vxlan_sock *vs = NULL;
3558         struct vxlan_dev_node *node;
3559         int l3mdev_index = 0;
3560
3561         if (vxlan->cfg.remote_ifindex)
3562                 l3mdev_index = l3mdev_master_upper_ifindex_by_index(
3563                         vxlan->net, vxlan->cfg.remote_ifindex);
3564
3565         if (!vxlan->cfg.no_share) {
3566                 spin_lock(&vn->sock_lock);
3567                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
3568                                      vxlan->cfg.dst_port, vxlan->cfg.flags,
3569                                      l3mdev_index);
3570                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
3571                         spin_unlock(&vn->sock_lock);
3572                         return -EBUSY;
3573                 }
3574                 spin_unlock(&vn->sock_lock);
3575         }
3576         if (!vs)
3577                 vs = vxlan_socket_create(vxlan->net, ipv6,
3578                                          vxlan->cfg.dst_port, vxlan->cfg.flags,
3579                                          l3mdev_index);
3580         if (IS_ERR(vs))
3581                 return PTR_ERR(vs);
3582 #if IS_ENABLED(CONFIG_IPV6)
3583         if (ipv6) {
3584                 rcu_assign_pointer(vxlan->vn6_sock, vs);
3585                 node = &vxlan->hlist6;
3586         } else
3587 #endif
3588         {
3589                 rcu_assign_pointer(vxlan->vn4_sock, vs);
3590                 node = &vxlan->hlist4;
3591         }
3592         vxlan_vs_add_dev(vs, vxlan, node);
3593         return 0;
3594 }
3595
3596 static int vxlan_sock_add(struct vxlan_dev *vxlan)
3597 {
3598         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
3599         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
3600         bool ipv4 = !ipv6 || metadata;
3601         int ret = 0;
3602
3603         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
3604 #if IS_ENABLED(CONFIG_IPV6)
3605         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
3606         if (ipv6) {
3607                 ret = __vxlan_sock_add(vxlan, true);
3608                 if (ret < 0 && ret != -EAFNOSUPPORT)
3609                         ipv4 = false;
3610         }
3611 #endif
3612         if (ipv4)
3613                 ret = __vxlan_sock_add(vxlan, false);
3614         if (ret < 0)
3615                 vxlan_sock_release(vxlan);
3616         return ret;
3617 }
3618
3619 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
3620                                  struct net_device **lower,
3621                                  struct vxlan_dev *old,
3622                                  struct netlink_ext_ack *extack)
3623 {
3624         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
3625         struct vxlan_dev *tmp;
3626         bool use_ipv6 = false;
3627
3628         if (conf->flags & VXLAN_F_GPE) {
3629                 /* For now, allow GPE only together with
3630                  * COLLECT_METADATA. This can be relaxed later; in such
3631                  * case, the other side of the PtP link will have to be
3632                  * provided.
3633                  */
3634                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3635                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3636                         NL_SET_ERR_MSG(extack,
3637                                        "VXLAN GPE does not support this combination of attributes");
3638                         return -EINVAL;
3639                 }
3640         }
3641
3642         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3643                 /* Unless IPv6 is explicitly requested, assume IPv4 */
3644                 conf->remote_ip.sa.sa_family = AF_INET;
3645                 conf->saddr.sa.sa_family = AF_INET;
3646         } else if (!conf->remote_ip.sa.sa_family) {
3647                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3648         } else if (!conf->saddr.sa.sa_family) {
3649                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3650         }
3651
3652         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3653                 NL_SET_ERR_MSG(extack,
3654                                "Local and remote address must be from the same family");
3655                 return -EINVAL;
3656         }
3657
3658         if (vxlan_addr_multicast(&conf->saddr)) {
3659                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3660                 return -EINVAL;
3661         }
3662
3663         if (conf->saddr.sa.sa_family == AF_INET6) {
3664                 if (!IS_ENABLED(CONFIG_IPV6)) {
3665                         NL_SET_ERR_MSG(extack,
3666                                        "IPv6 support not enabled in the kernel");
3667                         return -EPFNOSUPPORT;
3668                 }
3669                 use_ipv6 = true;
3670                 conf->flags |= VXLAN_F_IPV6;
3671
3672                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3673                         int local_type =
3674                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3675                         int remote_type =
3676                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3677
3678                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3679                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3680                                     (remote_type != IPV6_ADDR_ANY)) {
3681                                         NL_SET_ERR_MSG(extack,
3682                                                        "Invalid combination of local and remote address scopes");
3683                                         return -EINVAL;
3684                                 }
3685
3686                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3687                         } else {
3688                                 if (remote_type ==
3689                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3690                                         NL_SET_ERR_MSG(extack,
3691                                                        "Invalid combination of local and remote address scopes");
3692                                         return -EINVAL;
3693                                 }
3694
3695                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3696                         }
3697                 }
3698         }
3699
3700         if (conf->label && !use_ipv6) {
3701                 NL_SET_ERR_MSG(extack,
3702                                "Label attribute only applies to IPv6 VXLAN devices");
3703                 return -EINVAL;
3704         }
3705
3706         if (conf->remote_ifindex) {
3707                 struct net_device *lowerdev;
3708
3709                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3710                 if (!lowerdev) {
3711                         NL_SET_ERR_MSG(extack,
3712                                        "Invalid local interface, device not found");
3713                         return -ENODEV;
3714                 }
3715
3716 #if IS_ENABLED(CONFIG_IPV6)
3717                 if (use_ipv6) {
3718                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3719
3720                         if (idev && idev->cnf.disable_ipv6) {
3721                                 NL_SET_ERR_MSG(extack,
3722                                                "IPv6 support disabled by administrator");
3723                                 return -EPERM;
3724                         }
3725                 }
3726 #endif
3727
3728                 *lower = lowerdev;
3729         } else {
3730                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3731                         NL_SET_ERR_MSG(extack,
3732                                        "Local interface required for multicast remote destination");
3733
3734                         return -EINVAL;
3735                 }
3736
3737 #if IS_ENABLED(CONFIG_IPV6)
3738                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3739                         NL_SET_ERR_MSG(extack,
3740                                        "Local interface required for link-local local/remote addresses");
3741                         return -EINVAL;
3742                 }
3743 #endif
3744
3745                 *lower = NULL;
3746         }
3747
3748         if (!conf->dst_port) {
3749                 if (conf->flags & VXLAN_F_GPE)
3750                         conf->dst_port = htons(IANA_VXLAN_GPE_UDP_PORT);
3751                 else
3752                         conf->dst_port = htons(vxlan_port);
3753         }
3754
3755         if (!conf->age_interval)
3756                 conf->age_interval = FDB_AGE_DEFAULT;
3757
3758         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3759                 if (tmp == old)
3760                         continue;
3761
3762                 if (tmp->cfg.vni != conf->vni)
3763                         continue;
3764                 if (tmp->cfg.dst_port != conf->dst_port)
3765                         continue;
3766                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3767                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3768                         continue;
3769
3770                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3771                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3772                         continue;
3773
3774                 NL_SET_ERR_MSG(extack,
3775                                "A VXLAN device with the specified VNI already exists");
3776                 return -EEXIST;
3777         }
3778
3779         return 0;
3780 }
3781
3782 static void vxlan_config_apply(struct net_device *dev,
3783                                struct vxlan_config *conf,
3784                                struct net_device *lowerdev,
3785                                struct net *src_net,
3786                                bool changelink)
3787 {
3788         struct vxlan_dev *vxlan = netdev_priv(dev);
3789         struct vxlan_rdst *dst = &vxlan->default_dst;
3790         unsigned short needed_headroom = ETH_HLEN;
3791         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3792         int max_mtu = ETH_MAX_MTU;
3793
3794         if (!changelink) {
3795                 if (conf->flags & VXLAN_F_GPE)
3796                         vxlan_raw_setup(dev);
3797                 else
3798                         vxlan_ether_setup(dev);
3799
3800                 if (conf->mtu)
3801                         dev->mtu = conf->mtu;
3802
3803                 vxlan->net = src_net;
3804         }
3805
3806         dst->remote_vni = conf->vni;
3807
3808         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3809
3810         if (lowerdev) {
3811                 dst->remote_ifindex = conf->remote_ifindex;
3812
3813                 netif_set_gso_max_size(dev, lowerdev->gso_max_size);
3814                 netif_set_gso_max_segs(dev, lowerdev->gso_max_segs);
3815
3816                 needed_headroom = lowerdev->hard_header_len;
3817                 needed_headroom += lowerdev->needed_headroom;
3818
3819                 dev->needed_tailroom = lowerdev->needed_tailroom;
3820
3821                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3822                                            VXLAN_HEADROOM);
3823                 if (max_mtu < ETH_MIN_MTU)
3824                         max_mtu = ETH_MIN_MTU;
3825
3826                 if (!changelink && !conf->mtu)
3827                         dev->mtu = max_mtu;
3828         }
3829
3830         if (dev->mtu > max_mtu)
3831                 dev->mtu = max_mtu;
3832
3833         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3834                 needed_headroom += VXLAN6_HEADROOM;
3835         else
3836                 needed_headroom += VXLAN_HEADROOM;
3837         dev->needed_headroom = needed_headroom;
3838
3839         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3840 }
3841
3842 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3843                                struct vxlan_config *conf, bool changelink,
3844                                struct netlink_ext_ack *extack)
3845 {
3846         struct vxlan_dev *vxlan = netdev_priv(dev);
3847         struct net_device *lowerdev;
3848         int ret;
3849
3850         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3851         if (ret)
3852                 return ret;
3853
3854         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3855
3856         return 0;
3857 }
3858
3859 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3860                               struct vxlan_config *conf,
3861                               struct netlink_ext_ack *extack)
3862 {
3863         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3864         struct vxlan_dev *vxlan = netdev_priv(dev);
3865         struct net_device *remote_dev = NULL;
3866         struct vxlan_fdb *f = NULL;
3867         bool unregister = false;
3868         struct vxlan_rdst *dst;
3869         int err;
3870
3871         dst = &vxlan->default_dst;
3872         err = vxlan_dev_configure(net, dev, conf, false, extack);
3873         if (err)
3874                 return err;
3875
3876         dev->ethtool_ops = &vxlan_ethtool_ops;
3877
3878         /* create an fdb entry for a valid default destination */
3879         if (!vxlan_addr_any(&dst->remote_ip)) {
3880                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3881                                        &dst->remote_ip,
3882                                        NUD_REACHABLE | NUD_PERMANENT,
3883                                        vxlan->cfg.dst_port,
3884                                        dst->remote_vni,
3885                                        dst->remote_vni,
3886                                        dst->remote_ifindex,
3887                                        NTF_SELF, 0, &f, extack);
3888                 if (err)
3889                         return err;
3890         }
3891
3892         err = register_netdevice(dev);
3893         if (err)
3894                 goto errout;
3895         unregister = true;
3896
3897         if (dst->remote_ifindex) {
3898                 remote_dev = __dev_get_by_index(net, dst->remote_ifindex);
3899                 if (!remote_dev) {
3900                         err = -ENODEV;
3901                         goto errout;
3902                 }
3903
3904                 err = netdev_upper_dev_link(remote_dev, dev, extack);
3905                 if (err)
3906                         goto errout;
3907         }
3908
3909         err = rtnl_configure_link(dev, NULL);
3910         if (err < 0)
3911                 goto unlink;
3912
3913         if (f) {
3914                 vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f);
3915
3916                 /* notify default fdb entry */
3917                 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f),
3918                                        RTM_NEWNEIGH, true, extack);
3919                 if (err) {
3920                         vxlan_fdb_destroy(vxlan, f, false, false);
3921                         if (remote_dev)
3922                                 netdev_upper_dev_unlink(remote_dev, dev);
3923                         goto unregister;
3924                 }
3925         }
3926
3927         list_add(&vxlan->next, &vn->vxlan_list);
3928         if (remote_dev)
3929                 dst->remote_dev = remote_dev;
3930         return 0;
3931 unlink:
3932         if (remote_dev)
3933                 netdev_upper_dev_unlink(remote_dev, dev);
3934 errout:
3935         /* unregister_netdevice() destroys the default FDB entry with deletion
3936          * notification. But the addition notification was not sent yet, so
3937          * destroy the entry by hand here.
3938          */
3939         if (f)
3940                 __vxlan_fdb_free(f);
3941 unregister:
3942         if (unregister)
3943                 unregister_netdevice(dev);
3944         return err;
3945 }
3946
3947 /* Set/clear flags based on attribute */
3948 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[],
3949                           int attrtype, unsigned long mask, bool changelink,
3950                           bool changelink_supported,
3951                           struct netlink_ext_ack *extack)
3952 {
3953         unsigned long flags;
3954
3955         if (!tb[attrtype])
3956                 return 0;
3957
3958         if (changelink && !changelink_supported) {
3959                 vxlan_flag_attr_error(attrtype, extack);
3960                 return -EOPNOTSUPP;
3961         }
3962
3963         if (vxlan_policy[attrtype].type == NLA_FLAG)
3964                 flags = conf->flags | mask;
3965         else if (nla_get_u8(tb[attrtype]))
3966                 flags = conf->flags | mask;
3967         else
3968                 flags = conf->flags & ~mask;
3969
3970         conf->flags = flags;
3971
3972         return 0;
3973 }
3974
3975 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3976                          struct net_device *dev, struct vxlan_config *conf,
3977                          bool changelink, struct netlink_ext_ack *extack)
3978 {
3979         struct vxlan_dev *vxlan = netdev_priv(dev);
3980         int err = 0;
3981
3982         memset(conf, 0, sizeof(*conf));
3983
3984         /* if changelink operation, start with old existing cfg */
3985         if (changelink)
3986                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3987
3988         if (data[IFLA_VXLAN_ID]) {
3989                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3990
3991                 if (changelink && (vni != conf->vni)) {
3992                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI");
3993                         return -EOPNOTSUPP;
3994                 }
3995                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3996         }
3997
3998         if (data[IFLA_VXLAN_GROUP]) {
3999                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) {
4000                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group");
4001                         return -EOPNOTSUPP;
4002                 }
4003
4004                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
4005                 conf->remote_ip.sa.sa_family = AF_INET;
4006         } else if (data[IFLA_VXLAN_GROUP6]) {
4007                 if (!IS_ENABLED(CONFIG_IPV6)) {
4008                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel");
4009                         return -EPFNOSUPPORT;
4010                 }
4011
4012                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) {
4013                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group");
4014                         return -EOPNOTSUPP;
4015                 }
4016
4017                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
4018                 conf->remote_ip.sa.sa_family = AF_INET6;
4019         }
4020
4021         if (data[IFLA_VXLAN_LOCAL]) {
4022                 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) {
4023                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old");
4024                         return -EOPNOTSUPP;
4025                 }
4026
4027                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
4028                 conf->saddr.sa.sa_family = AF_INET;
4029         } else if (data[IFLA_VXLAN_LOCAL6]) {
4030                 if (!IS_ENABLED(CONFIG_IPV6)) {
4031                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel");
4032                         return -EPFNOSUPPORT;
4033                 }
4034
4035                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) {
4036                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old");
4037                         return -EOPNOTSUPP;
4038                 }
4039
4040                 /* TODO: respect scope id */
4041                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
4042                 conf->saddr.sa.sa_family = AF_INET6;
4043         }
4044
4045         if (data[IFLA_VXLAN_LINK])
4046                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
4047
4048         if (data[IFLA_VXLAN_TOS])
4049                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
4050
4051         if (data[IFLA_VXLAN_TTL])
4052                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
4053
4054         if (data[IFLA_VXLAN_TTL_INHERIT]) {
4055                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT,
4056                                     VXLAN_F_TTL_INHERIT, changelink, false,
4057                                     extack);
4058                 if (err)
4059                         return err;
4060
4061         }
4062
4063         if (data[IFLA_VXLAN_LABEL])
4064                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
4065                              IPV6_FLOWLABEL_MASK;
4066
4067         if (data[IFLA_VXLAN_LEARNING]) {
4068                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING,
4069                                     VXLAN_F_LEARN, changelink, true,
4070                                     extack);
4071                 if (err)
4072                         return err;
4073         } else if (!changelink) {
4074                 /* default to learn on a new device */
4075                 conf->flags |= VXLAN_F_LEARN;
4076         }
4077
4078         if (data[IFLA_VXLAN_AGEING])
4079                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
4080
4081         if (data[IFLA_VXLAN_PROXY]) {
4082                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY,
4083                                     VXLAN_F_PROXY, changelink, false,
4084                                     extack);
4085                 if (err)
4086                         return err;
4087         }
4088
4089         if (data[IFLA_VXLAN_RSC]) {
4090                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC,
4091                                     VXLAN_F_RSC, changelink, false,
4092                                     extack);
4093                 if (err)
4094                         return err;
4095         }
4096
4097         if (data[IFLA_VXLAN_L2MISS]) {
4098                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS,
4099                                     VXLAN_F_L2MISS, changelink, false,
4100                                     extack);
4101                 if (err)
4102                         return err;
4103         }
4104
4105         if (data[IFLA_VXLAN_L3MISS]) {
4106                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS,
4107                                     VXLAN_F_L3MISS, changelink, false,
4108                                     extack);
4109                 if (err)
4110                         return err;
4111         }
4112
4113         if (data[IFLA_VXLAN_LIMIT]) {
4114                 if (changelink) {
4115                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT],
4116                                             "Cannot change limit");
4117                         return -EOPNOTSUPP;
4118                 }
4119                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
4120         }
4121
4122         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
4123                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA,
4124                                     VXLAN_F_COLLECT_METADATA, changelink, false,
4125                                     extack);
4126                 if (err)
4127                         return err;
4128         }
4129
4130         if (data[IFLA_VXLAN_PORT_RANGE]) {
4131                 if (!changelink) {
4132                         const struct ifla_vxlan_port_range *p
4133                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
4134                         conf->port_min = ntohs(p->low);
4135                         conf->port_max = ntohs(p->high);
4136                 } else {
4137                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
4138                                             "Cannot change port range");
4139                         return -EOPNOTSUPP;
4140                 }
4141         }
4142
4143         if (data[IFLA_VXLAN_PORT]) {
4144                 if (changelink) {
4145                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT],
4146                                             "Cannot change port");
4147                         return -EOPNOTSUPP;
4148                 }
4149                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
4150         }
4151
4152         if (data[IFLA_VXLAN_UDP_CSUM]) {
4153                 if (changelink) {
4154                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM],
4155                                             "Cannot change UDP_CSUM flag");
4156                         return -EOPNOTSUPP;
4157                 }
4158                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
4159                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
4160         }
4161
4162         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
4163                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4164                                     VXLAN_F_UDP_ZERO_CSUM6_TX, changelink,
4165                                     false, extack);
4166                 if (err)
4167                         return err;
4168         }
4169
4170         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
4171                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4172                                     VXLAN_F_UDP_ZERO_CSUM6_RX, changelink,
4173                                     false, extack);
4174                 if (err)
4175                         return err;
4176         }
4177
4178         if (data[IFLA_VXLAN_REMCSUM_TX]) {
4179                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX,
4180                                     VXLAN_F_REMCSUM_TX, changelink, false,
4181                                     extack);
4182                 if (err)
4183                         return err;
4184         }
4185
4186         if (data[IFLA_VXLAN_REMCSUM_RX]) {
4187                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX,
4188                                     VXLAN_F_REMCSUM_RX, changelink, false,
4189                                     extack);
4190                 if (err)
4191                         return err;
4192         }
4193
4194         if (data[IFLA_VXLAN_GBP]) {
4195                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP,
4196                                     VXLAN_F_GBP, changelink, false, extack);
4197                 if (err)
4198                         return err;
4199         }
4200
4201         if (data[IFLA_VXLAN_GPE]) {
4202                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE,
4203                                     VXLAN_F_GPE, changelink, false,
4204                                     extack);
4205                 if (err)
4206                         return err;
4207         }
4208
4209         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
4210                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL,
4211                                     VXLAN_F_REMCSUM_NOPARTIAL, changelink,
4212                                     false, extack);
4213                 if (err)
4214                         return err;
4215         }
4216
4217         if (tb[IFLA_MTU]) {
4218                 if (changelink) {
4219                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
4220                                             "Cannot change mtu");
4221                         return -EOPNOTSUPP;
4222                 }
4223                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
4224         }
4225
4226         if (data[IFLA_VXLAN_DF])
4227                 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]);
4228
4229         return 0;
4230 }
4231
4232 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
4233                          struct nlattr *tb[], struct nlattr *data[],
4234                          struct netlink_ext_ack *extack)
4235 {
4236         struct vxlan_config conf;
4237         int err;
4238
4239         err = vxlan_nl2conf(tb, data, dev, &conf, false, extack);
4240         if (err)
4241                 return err;
4242
4243         return __vxlan_dev_create(src_net, dev, &conf, extack);
4244 }
4245
4246 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
4247                             struct nlattr *data[],
4248                             struct netlink_ext_ack *extack)
4249 {
4250         struct vxlan_dev *vxlan = netdev_priv(dev);
4251         struct net_device *lowerdev;
4252         struct vxlan_config conf;
4253         struct vxlan_rdst *dst;
4254         int err;
4255
4256         dst = &vxlan->default_dst;
4257         err = vxlan_nl2conf(tb, data, dev, &conf, true, extack);
4258         if (err)
4259                 return err;
4260
4261         err = vxlan_config_validate(vxlan->net, &conf, &lowerdev,
4262                                     vxlan, extack);
4263         if (err)
4264                 return err;
4265
4266         if (dst->remote_dev == lowerdev)
4267                 lowerdev = NULL;
4268
4269         err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev,
4270                                              extack);
4271         if (err)
4272                 return err;
4273
4274         /* handle default dst entry */
4275         if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) {
4276                 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni);
4277
4278                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4279                 if (!vxlan_addr_any(&conf.remote_ip)) {
4280                         err = vxlan_fdb_update(vxlan, all_zeros_mac,
4281                                                &conf.remote_ip,
4282                                                NUD_REACHABLE | NUD_PERMANENT,
4283                                                NLM_F_APPEND | NLM_F_CREATE,
4284                                                vxlan->cfg.dst_port,
4285                                                conf.vni, conf.vni,
4286                                                conf.remote_ifindex,
4287                                                NTF_SELF, 0, true, extack);
4288                         if (err) {
4289                                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4290                                 netdev_adjacent_change_abort(dst->remote_dev,
4291                                                              lowerdev, dev);
4292                                 return err;
4293                         }
4294                 }
4295                 if (!vxlan_addr_any(&dst->remote_ip))
4296                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
4297                                            dst->remote_ip,
4298                                            vxlan->cfg.dst_port,
4299                                            dst->remote_vni,
4300                                            dst->remote_vni,
4301                                            dst->remote_ifindex,
4302                                            true);
4303                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4304         }
4305
4306         if (conf.age_interval != vxlan->cfg.age_interval)
4307                 mod_timer(&vxlan->age_timer, jiffies);
4308
4309         netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev);
4310         if (lowerdev && lowerdev != dst->remote_dev)
4311                 dst->remote_dev = lowerdev;
4312         vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true);
4313         return 0;
4314 }
4315
4316 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
4317 {
4318         struct vxlan_dev *vxlan = netdev_priv(dev);
4319
4320         vxlan_flush(vxlan, true);
4321
4322         list_del(&vxlan->next);
4323         unregister_netdevice_queue(dev, head);
4324         if (vxlan->default_dst.remote_dev)
4325                 netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev);
4326 }
4327
4328 static size_t vxlan_get_size(const struct net_device *dev)
4329 {
4330
4331         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
4332                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
4333                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
4334                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
4335                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
4336                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL_INHERIT */
4337                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
4338                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_DF */
4339                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
4340                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
4341                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
4342                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
4343                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
4344                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
4345                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
4346                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
4347                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
4348                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
4349                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
4350                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
4351                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
4352                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
4353                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
4354                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
4355                 0;
4356 }
4357
4358 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
4359 {
4360         const struct vxlan_dev *vxlan = netdev_priv(dev);
4361         const struct vxlan_rdst *dst = &vxlan->default_dst;
4362         struct ifla_vxlan_port_range ports = {
4363                 .low =  htons(vxlan->cfg.port_min),
4364                 .high = htons(vxlan->cfg.port_max),
4365         };
4366
4367         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
4368                 goto nla_put_failure;
4369
4370         if (!vxlan_addr_any(&dst->remote_ip)) {
4371                 if (dst->remote_ip.sa.sa_family == AF_INET) {
4372                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
4373                                             dst->remote_ip.sin.sin_addr.s_addr))
4374                                 goto nla_put_failure;
4375 #if IS_ENABLED(CONFIG_IPV6)
4376                 } else {
4377                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
4378                                              &dst->remote_ip.sin6.sin6_addr))
4379                                 goto nla_put_failure;
4380 #endif
4381                 }
4382         }
4383
4384         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
4385                 goto nla_put_failure;
4386
4387         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
4388                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
4389                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
4390                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
4391                                 goto nla_put_failure;
4392 #if IS_ENABLED(CONFIG_IPV6)
4393                 } else {
4394                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
4395                                              &vxlan->cfg.saddr.sin6.sin6_addr))
4396                                 goto nla_put_failure;
4397 #endif
4398                 }
4399         }
4400
4401         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
4402             nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
4403                        !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
4404             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
4405             nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) ||
4406             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
4407             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
4408                        !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
4409             nla_put_u8(skb, IFLA_VXLAN_PROXY,
4410                        !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
4411             nla_put_u8(skb, IFLA_VXLAN_RSC,
4412                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
4413             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
4414                        !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
4415             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
4416                        !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
4417             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
4418                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
4419             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
4420             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
4421             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
4422             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
4423                        !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
4424             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4425                        !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
4426             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4427                        !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
4428             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
4429                        !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
4430             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
4431                        !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
4432                 goto nla_put_failure;
4433
4434         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
4435                 goto nla_put_failure;
4436
4437         if (vxlan->cfg.flags & VXLAN_F_GBP &&
4438             nla_put_flag(skb, IFLA_VXLAN_GBP))
4439                 goto nla_put_failure;
4440
4441         if (vxlan->cfg.flags & VXLAN_F_GPE &&
4442             nla_put_flag(skb, IFLA_VXLAN_GPE))
4443                 goto nla_put_failure;
4444
4445         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
4446             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
4447                 goto nla_put_failure;
4448
4449         return 0;
4450
4451 nla_put_failure:
4452         return -EMSGSIZE;
4453 }
4454
4455 static struct net *vxlan_get_link_net(const struct net_device *dev)
4456 {
4457         struct vxlan_dev *vxlan = netdev_priv(dev);
4458
4459         return vxlan->net;
4460 }
4461
4462 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
4463         .kind           = "vxlan",
4464         .maxtype        = IFLA_VXLAN_MAX,
4465         .policy         = vxlan_policy,
4466         .priv_size      = sizeof(struct vxlan_dev),
4467         .setup          = vxlan_setup,
4468         .validate       = vxlan_validate,
4469         .newlink        = vxlan_newlink,
4470         .changelink     = vxlan_changelink,
4471         .dellink        = vxlan_dellink,
4472         .get_size       = vxlan_get_size,
4473         .fill_info      = vxlan_fill_info,
4474         .get_link_net   = vxlan_get_link_net,
4475 };
4476
4477 struct net_device *vxlan_dev_create(struct net *net, const char *name,
4478                                     u8 name_assign_type,
4479                                     struct vxlan_config *conf)
4480 {
4481         struct nlattr *tb[IFLA_MAX + 1];
4482         struct net_device *dev;
4483         int err;
4484
4485         memset(&tb, 0, sizeof(tb));
4486
4487         dev = rtnl_create_link(net, name, name_assign_type,
4488                                &vxlan_link_ops, tb, NULL);
4489         if (IS_ERR(dev))
4490                 return dev;
4491
4492         err = __vxlan_dev_create(net, dev, conf, NULL);
4493         if (err < 0) {
4494                 free_netdev(dev);
4495                 return ERR_PTR(err);
4496         }
4497
4498         err = rtnl_configure_link(dev, NULL);
4499         if (err < 0) {
4500                 LIST_HEAD(list_kill);
4501
4502                 vxlan_dellink(dev, &list_kill);
4503                 unregister_netdevice_many(&list_kill);
4504                 return ERR_PTR(err);
4505         }
4506
4507         return dev;
4508 }
4509 EXPORT_SYMBOL_GPL(vxlan_dev_create);
4510
4511 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
4512                                              struct net_device *dev)
4513 {
4514         struct vxlan_dev *vxlan, *next;
4515         LIST_HEAD(list_kill);
4516
4517         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4518                 struct vxlan_rdst *dst = &vxlan->default_dst;
4519
4520                 /* In case we created vxlan device with carrier
4521                  * and we loose the carrier due to module unload
4522                  * we also need to remove vxlan device. In other
4523                  * cases, it's not necessary and remote_ifindex
4524                  * is 0 here, so no matches.
4525                  */
4526                 if (dst->remote_ifindex == dev->ifindex)
4527                         vxlan_dellink(vxlan->dev, &list_kill);
4528         }
4529
4530         unregister_netdevice_many(&list_kill);
4531 }
4532
4533 static int vxlan_netdevice_event(struct notifier_block *unused,
4534                                  unsigned long event, void *ptr)
4535 {
4536         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4537         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
4538
4539         if (event == NETDEV_UNREGISTER)
4540                 vxlan_handle_lowerdev_unregister(vn, dev);
4541         else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
4542                 vxlan_offload_rx_ports(dev, true);
4543         else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
4544                 vxlan_offload_rx_ports(dev, false);
4545
4546         return NOTIFY_DONE;
4547 }
4548
4549 static struct notifier_block vxlan_notifier_block __read_mostly = {
4550         .notifier_call = vxlan_netdevice_event,
4551 };
4552
4553 static void
4554 vxlan_fdb_offloaded_set(struct net_device *dev,
4555                         struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4556 {
4557         struct vxlan_dev *vxlan = netdev_priv(dev);
4558         struct vxlan_rdst *rdst;
4559         struct vxlan_fdb *f;
4560         u32 hash_index;
4561
4562         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4563
4564         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4565
4566         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4567         if (!f)
4568                 goto out;
4569
4570         rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip,
4571                                    fdb_info->remote_port,
4572                                    fdb_info->remote_vni,
4573                                    fdb_info->remote_ifindex);
4574         if (!rdst)
4575                 goto out;
4576
4577         rdst->offloaded = fdb_info->offloaded;
4578
4579 out:
4580         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4581 }
4582
4583 static int
4584 vxlan_fdb_external_learn_add(struct net_device *dev,
4585                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4586 {
4587         struct vxlan_dev *vxlan = netdev_priv(dev);
4588         struct netlink_ext_ack *extack;
4589         u32 hash_index;
4590         int err;
4591
4592         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4593         extack = switchdev_notifier_info_to_extack(&fdb_info->info);
4594
4595         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4596         err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip,
4597                                NUD_REACHABLE,
4598                                NLM_F_CREATE | NLM_F_REPLACE,
4599                                fdb_info->remote_port,
4600                                fdb_info->vni,
4601                                fdb_info->remote_vni,
4602                                fdb_info->remote_ifindex,
4603                                NTF_USE | NTF_SELF | NTF_EXT_LEARNED,
4604                                0, false, extack);
4605         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4606
4607         return err;
4608 }
4609
4610 static int
4611 vxlan_fdb_external_learn_del(struct net_device *dev,
4612                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4613 {
4614         struct vxlan_dev *vxlan = netdev_priv(dev);
4615         struct vxlan_fdb *f;
4616         u32 hash_index;
4617         int err = 0;
4618
4619         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4620         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4621
4622         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4623         if (!f)
4624                 err = -ENOENT;
4625         else if (f->flags & NTF_EXT_LEARNED)
4626                 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr,
4627                                          fdb_info->remote_ip,
4628                                          fdb_info->remote_port,
4629                                          fdb_info->vni,
4630                                          fdb_info->remote_vni,
4631                                          fdb_info->remote_ifindex,
4632                                          false);
4633
4634         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4635
4636         return err;
4637 }
4638
4639 static int vxlan_switchdev_event(struct notifier_block *unused,
4640                                  unsigned long event, void *ptr)
4641 {
4642         struct net_device *dev = switchdev_notifier_info_to_dev(ptr);
4643         struct switchdev_notifier_vxlan_fdb_info *fdb_info;
4644         int err = 0;
4645
4646         switch (event) {
4647         case SWITCHDEV_VXLAN_FDB_OFFLOADED:
4648                 vxlan_fdb_offloaded_set(dev, ptr);
4649                 break;
4650         case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE:
4651                 fdb_info = ptr;
4652                 err = vxlan_fdb_external_learn_add(dev, fdb_info);
4653                 if (err) {
4654                         err = notifier_from_errno(err);
4655                         break;
4656                 }
4657                 fdb_info->offloaded = true;
4658                 vxlan_fdb_offloaded_set(dev, fdb_info);
4659                 break;
4660         case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE:
4661                 fdb_info = ptr;
4662                 err = vxlan_fdb_external_learn_del(dev, fdb_info);
4663                 if (err) {
4664                         err = notifier_from_errno(err);
4665                         break;
4666                 }
4667                 fdb_info->offloaded = false;
4668                 vxlan_fdb_offloaded_set(dev, fdb_info);
4669                 break;
4670         }
4671
4672         return err;
4673 }
4674
4675 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = {
4676         .notifier_call = vxlan_switchdev_event,
4677 };
4678
4679 static void vxlan_fdb_nh_flush(struct nexthop *nh)
4680 {
4681         struct vxlan_fdb *fdb;
4682         struct vxlan_dev *vxlan;
4683         u32 hash_index;
4684
4685         rcu_read_lock();
4686         list_for_each_entry_rcu(fdb, &nh->fdb_list, nh_list) {
4687                 vxlan = rcu_dereference(fdb->vdev);
4688                 WARN_ON(!vxlan);
4689                 hash_index = fdb_head_index(vxlan, fdb->eth_addr,
4690                                             vxlan->default_dst.remote_vni);
4691                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4692                 if (!hlist_unhashed(&fdb->hlist))
4693                         vxlan_fdb_destroy(vxlan, fdb, false, false);
4694                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4695         }
4696         rcu_read_unlock();
4697 }
4698
4699 static int vxlan_nexthop_event(struct notifier_block *nb,
4700                                unsigned long event, void *ptr)
4701 {
4702         struct nh_notifier_info *info = ptr;
4703         struct nexthop *nh;
4704
4705         if (event != NEXTHOP_EVENT_DEL)
4706                 return NOTIFY_DONE;
4707
4708         nh = nexthop_find_by_id(info->net, info->id);
4709         if (!nh)
4710                 return NOTIFY_DONE;
4711
4712         vxlan_fdb_nh_flush(nh);
4713
4714         return NOTIFY_DONE;
4715 }
4716
4717 static __net_init int vxlan_init_net(struct net *net)
4718 {
4719         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4720         unsigned int h;
4721
4722         INIT_LIST_HEAD(&vn->vxlan_list);
4723         spin_lock_init(&vn->sock_lock);
4724         vn->nexthop_notifier_block.notifier_call = vxlan_nexthop_event;
4725
4726         for (h = 0; h < PORT_HASH_SIZE; ++h)
4727                 INIT_HLIST_HEAD(&vn->sock_list[h]);
4728
4729         return register_nexthop_notifier(net, &vn->nexthop_notifier_block,
4730                                          NULL);
4731 }
4732
4733 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
4734 {
4735         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4736         struct vxlan_dev *vxlan, *next;
4737         struct net_device *dev, *aux;
4738
4739         for_each_netdev_safe(net, dev, aux)
4740                 if (dev->rtnl_link_ops == &vxlan_link_ops)
4741                         unregister_netdevice_queue(dev, head);
4742
4743         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4744                 /* If vxlan->dev is in the same netns, it has already been added
4745                  * to the list by the previous loop.
4746                  */
4747                 if (!net_eq(dev_net(vxlan->dev), net))
4748                         unregister_netdevice_queue(vxlan->dev, head);
4749         }
4750
4751 }
4752
4753 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
4754 {
4755         struct net *net;
4756         LIST_HEAD(list);
4757         unsigned int h;
4758
4759         list_for_each_entry(net, net_list, exit_list) {
4760                 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4761
4762                 unregister_nexthop_notifier(net, &vn->nexthop_notifier_block);
4763         }
4764         rtnl_lock();
4765         list_for_each_entry(net, net_list, exit_list)
4766                 vxlan_destroy_tunnels(net, &list);
4767
4768         unregister_netdevice_many(&list);
4769         rtnl_unlock();
4770
4771         list_for_each_entry(net, net_list, exit_list) {
4772                 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4773
4774                 for (h = 0; h < PORT_HASH_SIZE; ++h)
4775                         WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
4776         }
4777 }
4778
4779 static struct pernet_operations vxlan_net_ops = {
4780         .init = vxlan_init_net,
4781         .exit_batch = vxlan_exit_batch_net,
4782         .id   = &vxlan_net_id,
4783         .size = sizeof(struct vxlan_net),
4784 };
4785
4786 static int __init vxlan_init_module(void)
4787 {
4788         int rc;
4789
4790         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
4791
4792         rc = register_pernet_subsys(&vxlan_net_ops);
4793         if (rc)
4794                 goto out1;
4795
4796         rc = register_netdevice_notifier(&vxlan_notifier_block);
4797         if (rc)
4798                 goto out2;
4799
4800         rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block);
4801         if (rc)
4802                 goto out3;
4803
4804         rc = rtnl_link_register(&vxlan_link_ops);
4805         if (rc)
4806                 goto out4;
4807
4808         return 0;
4809 out4:
4810         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4811 out3:
4812         unregister_netdevice_notifier(&vxlan_notifier_block);
4813 out2:
4814         unregister_pernet_subsys(&vxlan_net_ops);
4815 out1:
4816         return rc;
4817 }
4818 late_initcall(vxlan_init_module);
4819
4820 static void __exit vxlan_cleanup_module(void)
4821 {
4822         rtnl_link_unregister(&vxlan_link_ops);
4823         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4824         unregister_netdevice_notifier(&vxlan_notifier_block);
4825         unregister_pernet_subsys(&vxlan_net_ops);
4826         /* rcu_barrier() is called by netns */
4827 }
4828 module_exit(vxlan_cleanup_module);
4829
4830 MODULE_LICENSE("GPL");
4831 MODULE_VERSION(VXLAN_VERSION);
4832 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
4833 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
4834 MODULE_ALIAS_RTNL_LINK("vxlan");