]>
Commit | Line | Data |
---|---|---|
d2912cb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
d342894c | 2 | /* |
eb5ce439 | 3 | * VXLAN: Virtual eXtensible Local Area Network |
d342894c | 4 | * |
3b8df3c6 | 5 | * Copyright (c) 2012-2013 Vyatta Inc. |
d342894c | 6 | */ |
7 | ||
8 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
9 | ||
10 | #include <linux/kernel.h> | |
d342894c | 11 | #include <linux/module.h> |
12 | #include <linux/errno.h> | |
13 | #include <linux/slab.h> | |
d342894c | 14 | #include <linux/udp.h> |
15 | #include <linux/igmp.h> | |
d342894c | 16 | #include <linux/if_ether.h> |
1b13c97f | 17 | #include <linux/ethtool.h> |
e4f67add DS |
18 | #include <net/arp.h> |
19 | #include <net/ndisc.h> | |
3616d08b | 20 | #include <net/ipv6_stubs.h> |
d342894c | 21 | #include <net/ip.h> |
22 | #include <net/icmp.h> | |
d342894c | 23 | #include <net/rtnetlink.h> |
d342894c | 24 | #include <net/inet_ecn.h> |
25 | #include <net/net_namespace.h> | |
26 | #include <net/netns/generic.h> | |
fa20e0e3 | 27 | #include <net/tun_proto.h> |
012a5729 | 28 | #include <net/vxlan.h> |
40d29af0 | 29 | |
e4c7ed41 | 30 | #if IS_ENABLED(CONFIG_IPV6) |
e4c7ed41 | 31 | #include <net/ip6_tunnel.h> |
660d98ca | 32 | #include <net/ip6_checksum.h> |
e4c7ed41 | 33 | #endif |
d342894c | 34 | |
35 | #define VXLAN_VERSION "0.1" | |
36 | ||
553675fb | 37 | #define PORT_HASH_BITS 8 |
38 | #define PORT_HASH_SIZE (1<<PORT_HASH_BITS) | |
d342894c | 39 | #define FDB_AGE_DEFAULT 300 /* 5 min */ |
40 | #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */ | |
41 | ||
23c578bf | 42 | /* UDP port for VXLAN traffic. |
43 | * The IANA assigned port is 4789, but the Linux default is 8472 | |
234f5b73 | 44 | * for compatibility with early adopters. |
23c578bf | 45 | */ |
9daaa397 SH |
46 | static unsigned short vxlan_port __read_mostly = 8472; |
47 | module_param_named(udp_port, vxlan_port, ushort, 0444); | |
d342894c | 48 | MODULE_PARM_DESC(udp_port, "Destination UDP port"); |
49 | ||
50 | static bool log_ecn_error = true; | |
51 | module_param(log_ecn_error, bool, 0644); | |
52 | MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN"); | |
53 | ||
c7d03a00 | 54 | static unsigned int vxlan_net_id; |
0dfbdf41 | 55 | static struct rtnl_link_ops vxlan_link_ops; |
553675fb | 56 | |
7256eac1 | 57 | static const u8 all_zeros_mac[ETH_ALEN + 2]; |
afbd8bae | 58 | |
205f356d | 59 | static int vxlan_sock_add(struct vxlan_dev *vxlan); |
614732ea | 60 | |
a53cb29b MB |
61 | static void vxlan_vs_del_dev(struct vxlan_dev *vxlan); |
62 | ||
553675fb | 63 | /* per-network namespace private data for this module */ |
64 | struct vxlan_net { | |
65 | struct list_head vxlan_list; | |
66 | struct hlist_head sock_list[PORT_HASH_SIZE]; | |
1c51a915 | 67 | spinlock_t sock_lock; |
553675fb | 68 | }; |
69 | ||
d342894c | 70 | /* Forwarding table entry */ |
71 | struct vxlan_fdb { | |
72 | struct hlist_node hlist; /* linked list of entries */ | |
73 | struct rcu_head rcu; | |
74 | unsigned long updated; /* jiffies */ | |
75 | unsigned long used; | |
3e61aa8f | 76 | struct list_head remotes; |
7177a3b0 | 77 | u8 eth_addr[ETH_ALEN]; |
d342894c | 78 | u16 state; /* see ndm_state */ |
3ad7a4b1 | 79 | __be32 vni; |
45598c1c | 80 | u16 flags; /* see ndm_flags and below */ |
d342894c | 81 | }; |
82 | ||
45598c1c PM |
83 | #define NTF_VXLAN_ADDED_BY_USER 0x100 |
84 | ||
d342894c | 85 | /* salt for hash table */ |
86 | static u32 vxlan_salt __read_mostly; | |
87 | ||
ee122c79 TG |
88 | static inline bool vxlan_collect_metadata(struct vxlan_sock *vs) |
89 | { | |
e7030878 TG |
90 | return vs->flags & VXLAN_F_COLLECT_METADATA || |
91 | ip_tunnel_collect_metadata(); | |
ee122c79 TG |
92 | } |
93 | ||
e4c7ed41 CW |
94 | #if IS_ENABLED(CONFIG_IPV6) |
95 | static inline | |
96 | bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b) | |
97 | { | |
f0ef3126 JB |
98 | if (a->sa.sa_family != b->sa.sa_family) |
99 | return false; | |
100 | if (a->sa.sa_family == AF_INET6) | |
101 | return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr); | |
102 | else | |
103 | return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr; | |
e4c7ed41 CW |
104 | } |
105 | ||
e4c7ed41 CW |
106 | static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla) |
107 | { | |
f0ef3126 | 108 | if (nla_len(nla) >= sizeof(struct in6_addr)) { |
67b61f6c | 109 | ip->sin6.sin6_addr = nla_get_in6_addr(nla); |
f0ef3126 JB |
110 | ip->sa.sa_family = AF_INET6; |
111 | return 0; | |
112 | } else if (nla_len(nla) >= sizeof(__be32)) { | |
67b61f6c | 113 | ip->sin.sin_addr.s_addr = nla_get_in_addr(nla); |
f0ef3126 JB |
114 | ip->sa.sa_family = AF_INET; |
115 | return 0; | |
116 | } else { | |
117 | return -EAFNOSUPPORT; | |
118 | } | |
e4c7ed41 CW |
119 | } |
120 | ||
121 | static int vxlan_nla_put_addr(struct sk_buff *skb, int attr, | |
f0ef3126 | 122 | const union vxlan_addr *ip) |
e4c7ed41 | 123 | { |
f0ef3126 | 124 | if (ip->sa.sa_family == AF_INET6) |
930345ea | 125 | return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr); |
f0ef3126 | 126 | else |
930345ea | 127 | return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr); |
e4c7ed41 CW |
128 | } |
129 | ||
130 | #else /* !CONFIG_IPV6 */ | |
131 | ||
132 | static inline | |
133 | bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b) | |
134 | { | |
f0ef3126 | 135 | return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr; |
e4c7ed41 CW |
136 | } |
137 | ||
e4c7ed41 CW |
138 | static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla) |
139 | { | |
f0ef3126 JB |
140 | if (nla_len(nla) >= sizeof(struct in6_addr)) { |
141 | return -EAFNOSUPPORT; | |
142 | } else if (nla_len(nla) >= sizeof(__be32)) { | |
67b61f6c | 143 | ip->sin.sin_addr.s_addr = nla_get_in_addr(nla); |
f0ef3126 JB |
144 | ip->sa.sa_family = AF_INET; |
145 | return 0; | |
146 | } else { | |
147 | return -EAFNOSUPPORT; | |
148 | } | |
e4c7ed41 CW |
149 | } |
150 | ||
151 | static int vxlan_nla_put_addr(struct sk_buff *skb, int attr, | |
f0ef3126 | 152 | const union vxlan_addr *ip) |
e4c7ed41 | 153 | { |
930345ea | 154 | return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr); |
e4c7ed41 CW |
155 | } |
156 | #endif | |
157 | ||
553675fb | 158 | /* Virtual Network hash table head */ |
54bfd872 | 159 | static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni) |
553675fb | 160 | { |
54bfd872 | 161 | return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)]; |
553675fb | 162 | } |
163 | ||
164 | /* Socket hash table head */ | |
165 | static inline struct hlist_head *vs_head(struct net *net, __be16 port) | |
d342894c | 166 | { |
167 | struct vxlan_net *vn = net_generic(net, vxlan_net_id); | |
168 | ||
553675fb | 169 | return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)]; |
170 | } | |
171 | ||
3e61aa8f SH |
172 | /* First remote destination for a forwarding entry. |
173 | * Guaranteed to be non-NULL because remotes are never deleted. | |
174 | */ | |
5ca5461c | 175 | static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb) |
3e61aa8f | 176 | { |
5ca5461c | 177 | return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list); |
178 | } | |
179 | ||
180 | static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb) | |
181 | { | |
182 | return list_first_entry(&fdb->remotes, struct vxlan_rdst, list); | |
3e61aa8f SH |
183 | } |
184 | ||
ac5132d1 TG |
185 | /* Find VXLAN socket based on network namespace, address family and UDP port |
186 | * and enabled unshareable flags. | |
187 | */ | |
188 | static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family, | |
aab8cc36 | 189 | __be16 port, u32 flags, int ifindex) |
553675fb | 190 | { |
191 | struct vxlan_sock *vs; | |
af33c1ad TH |
192 | |
193 | flags &= VXLAN_F_RCV_FLAGS; | |
553675fb | 194 | |
195 | hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) { | |
19ca9fc1 | 196 | if (inet_sk(vs->sock->sk)->inet_sport == port && |
705cc62f | 197 | vxlan_get_sk_family(vs) == family && |
aab8cc36 AB |
198 | vs->flags == flags && |
199 | vs->sock->sk->sk_bound_dev_if == ifindex) | |
553675fb | 200 | return vs; |
201 | } | |
202 | return NULL; | |
d342894c | 203 | } |
204 | ||
49f810f0 MS |
205 | static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex, |
206 | __be32 vni) | |
d342894c | 207 | { |
69e76661 | 208 | struct vxlan_dev_node *node; |
d342894c | 209 | |
b9167b2e JB |
210 | /* For flow based devices, map all packets to VNI 0 */ |
211 | if (vs->flags & VXLAN_F_COLLECT_METADATA) | |
212 | vni = 0; | |
213 | ||
69e76661 JB |
214 | hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) { |
215 | if (node->vxlan->default_dst.remote_vni != vni) | |
49f810f0 MS |
216 | continue; |
217 | ||
218 | if (IS_ENABLED(CONFIG_IPV6)) { | |
69e76661 | 219 | const struct vxlan_config *cfg = &node->vxlan->cfg; |
49f810f0 MS |
220 | |
221 | if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) && | |
222 | cfg->remote_ifindex != ifindex) | |
223 | continue; | |
224 | } | |
225 | ||
69e76661 | 226 | return node->vxlan; |
d342894c | 227 | } |
228 | ||
229 | return NULL; | |
230 | } | |
231 | ||
5cfccc5a | 232 | /* Look up VNI in a per net namespace table */ |
49f810f0 MS |
233 | static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex, |
234 | __be32 vni, sa_family_t family, | |
235 | __be16 port, u32 flags) | |
5cfccc5a PS |
236 | { |
237 | struct vxlan_sock *vs; | |
238 | ||
aab8cc36 | 239 | vs = vxlan_find_sock(net, family, port, flags, ifindex); |
5cfccc5a PS |
240 | if (!vs) |
241 | return NULL; | |
242 | ||
49f810f0 | 243 | return vxlan_vs_find_vni(vs, ifindex, vni); |
5cfccc5a PS |
244 | } |
245 | ||
d342894c | 246 | /* Fill in neighbour message in skbuff. */ |
247 | static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan, | |
234f5b73 SH |
248 | const struct vxlan_fdb *fdb, |
249 | u32 portid, u32 seq, int type, unsigned int flags, | |
250 | const struct vxlan_rdst *rdst) | |
d342894c | 251 | { |
252 | unsigned long now = jiffies; | |
253 | struct nda_cacheinfo ci; | |
254 | struct nlmsghdr *nlh; | |
255 | struct ndmsg *ndm; | |
e4f67add | 256 | bool send_ip, send_eth; |
d342894c | 257 | |
258 | nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags); | |
259 | if (nlh == NULL) | |
260 | return -EMSGSIZE; | |
261 | ||
262 | ndm = nlmsg_data(nlh); | |
263 | memset(ndm, 0, sizeof(*ndm)); | |
e4f67add DS |
264 | |
265 | send_eth = send_ip = true; | |
266 | ||
267 | if (type == RTM_GETNEIGH) { | |
e4c7ed41 | 268 | send_ip = !vxlan_addr_any(&rdst->remote_ip); |
e4f67add | 269 | send_eth = !is_zero_ether_addr(fdb->eth_addr); |
8f48ba71 | 270 | ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET; |
e4f67add DS |
271 | } else |
272 | ndm->ndm_family = AF_BRIDGE; | |
d342894c | 273 | ndm->ndm_state = fdb->state; |
274 | ndm->ndm_ifindex = vxlan->dev->ifindex; | |
ae884082 | 275 | ndm->ndm_flags = fdb->flags; |
0efe1173 PM |
276 | if (rdst->offloaded) |
277 | ndm->ndm_flags |= NTF_OFFLOADED; | |
545469f7 | 278 | ndm->ndm_type = RTN_UNICAST; |
d342894c | 279 | |
193523bf | 280 | if (!net_eq(dev_net(vxlan->dev), vxlan->net) && |
4967082b | 281 | nla_put_s32(skb, NDA_LINK_NETNSID, |
38f507f1 | 282 | peernet2id(dev_net(vxlan->dev), vxlan->net))) |
193523bf ND |
283 | goto nla_put_failure; |
284 | ||
e4f67add | 285 | if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr)) |
d342894c | 286 | goto nla_put_failure; |
287 | ||
e4c7ed41 | 288 | if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip)) |
6681712d DS |
289 | goto nla_put_failure; |
290 | ||
0dfbdf41 | 291 | if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port && |
6681712d DS |
292 | nla_put_be16(skb, NDA_PORT, rdst->remote_port)) |
293 | goto nla_put_failure; | |
c7995c43 | 294 | if (rdst->remote_vni != vxlan->default_dst.remote_vni && |
54bfd872 | 295 | nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni))) |
6681712d | 296 | goto nla_put_failure; |
dc5321d7 | 297 | if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni && |
3ad7a4b1 RP |
298 | nla_put_u32(skb, NDA_SRC_VNI, |
299 | be32_to_cpu(fdb->vni))) | |
300 | goto nla_put_failure; | |
6681712d DS |
301 | if (rdst->remote_ifindex && |
302 | nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex)) | |
d342894c | 303 | goto nla_put_failure; |
304 | ||
305 | ci.ndm_used = jiffies_to_clock_t(now - fdb->used); | |
306 | ci.ndm_confirmed = 0; | |
307 | ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated); | |
308 | ci.ndm_refcnt = 0; | |
309 | ||
310 | if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci)) | |
311 | goto nla_put_failure; | |
312 | ||
053c095a JB |
313 | nlmsg_end(skb, nlh); |
314 | return 0; | |
d342894c | 315 | |
316 | nla_put_failure: | |
317 | nlmsg_cancel(skb, nlh); | |
318 | return -EMSGSIZE; | |
319 | } | |
320 | ||
321 | static inline size_t vxlan_nlmsg_size(void) | |
322 | { | |
323 | return NLMSG_ALIGN(sizeof(struct ndmsg)) | |
324 | + nla_total_size(ETH_ALEN) /* NDA_LLADDR */ | |
e4c7ed41 | 325 | + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */ |
73cf3317 | 326 | + nla_total_size(sizeof(__be16)) /* NDA_PORT */ |
6681712d DS |
327 | + nla_total_size(sizeof(__be32)) /* NDA_VNI */ |
328 | + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */ | |
4967082b | 329 | + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */ |
d342894c | 330 | + nla_total_size(sizeof(struct nda_cacheinfo)); |
331 | } | |
332 | ||
9a997353 PM |
333 | static void __vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb, |
334 | struct vxlan_rdst *rd, int type) | |
d342894c | 335 | { |
336 | struct net *net = dev_net(vxlan->dev); | |
337 | struct sk_buff *skb; | |
338 | int err = -ENOBUFS; | |
339 | ||
340 | skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC); | |
341 | if (skb == NULL) | |
342 | goto errout; | |
343 | ||
9e4b93f9 | 344 | err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd); |
d342894c | 345 | if (err < 0) { |
346 | /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */ | |
347 | WARN_ON(err == -EMSGSIZE); | |
348 | kfree_skb(skb); | |
349 | goto errout; | |
350 | } | |
351 | ||
352 | rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC); | |
353 | return; | |
354 | errout: | |
355 | if (err < 0) | |
356 | rtnl_set_sk_err(net, RTNLGRP_NEIGH, err); | |
357 | } | |
358 | ||
ff23b91c PM |
359 | static void vxlan_fdb_switchdev_notifier_info(const struct vxlan_dev *vxlan, |
360 | const struct vxlan_fdb *fdb, | |
361 | const struct vxlan_rdst *rd, | |
4c59b7d1 | 362 | struct netlink_ext_ack *extack, |
ff23b91c PM |
363 | struct switchdev_notifier_vxlan_fdb_info *fdb_info) |
364 | { | |
365 | fdb_info->info.dev = vxlan->dev; | |
4c59b7d1 | 366 | fdb_info->info.extack = extack; |
ff23b91c PM |
367 | fdb_info->remote_ip = rd->remote_ip; |
368 | fdb_info->remote_port = rd->remote_port; | |
369 | fdb_info->remote_vni = rd->remote_vni; | |
370 | fdb_info->remote_ifindex = rd->remote_ifindex; | |
371 | memcpy(fdb_info->eth_addr, fdb->eth_addr, ETH_ALEN); | |
372 | fdb_info->vni = fdb->vni; | |
373 | fdb_info->offloaded = rd->offloaded; | |
374 | fdb_info->added_by_user = fdb->flags & NTF_VXLAN_ADDED_BY_USER; | |
375 | } | |
376 | ||
61f46fe8 PM |
377 | static int vxlan_fdb_switchdev_call_notifiers(struct vxlan_dev *vxlan, |
378 | struct vxlan_fdb *fdb, | |
379 | struct vxlan_rdst *rd, | |
4c59b7d1 PM |
380 | bool adding, |
381 | struct netlink_ext_ack *extack) | |
9a997353 PM |
382 | { |
383 | struct switchdev_notifier_vxlan_fdb_info info; | |
384 | enum switchdev_notifier_type notifier_type; | |
61f46fe8 | 385 | int ret; |
9a997353 PM |
386 | |
387 | if (WARN_ON(!rd)) | |
61f46fe8 | 388 | return 0; |
9a997353 PM |
389 | |
390 | notifier_type = adding ? SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE | |
391 | : SWITCHDEV_VXLAN_FDB_DEL_TO_DEVICE; | |
4c59b7d1 | 392 | vxlan_fdb_switchdev_notifier_info(vxlan, fdb, rd, NULL, &info); |
61f46fe8 | 393 | ret = call_switchdev_notifiers(notifier_type, vxlan->dev, |
6685987c | 394 | &info.info, extack); |
61f46fe8 | 395 | return notifier_to_errno(ret); |
9a997353 PM |
396 | } |
397 | ||
61f46fe8 | 398 | static int vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb, |
4c59b7d1 PM |
399 | struct vxlan_rdst *rd, int type, bool swdev_notify, |
400 | struct netlink_ext_ack *extack) | |
9a997353 | 401 | { |
61f46fe8 PM |
402 | int err; |
403 | ||
0e6160f3 PM |
404 | if (swdev_notify) { |
405 | switch (type) { | |
406 | case RTM_NEWNEIGH: | |
61f46fe8 | 407 | err = vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd, |
4c59b7d1 | 408 | true, extack); |
61f46fe8 PM |
409 | if (err) |
410 | return err; | |
0e6160f3 PM |
411 | break; |
412 | case RTM_DELNEIGH: | |
413 | vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd, | |
4c59b7d1 | 414 | false, extack); |
0e6160f3 PM |
415 | break; |
416 | } | |
9a997353 PM |
417 | } |
418 | ||
419 | __vxlan_fdb_notify(vxlan, fdb, rd, type); | |
61f46fe8 | 420 | return 0; |
9a997353 PM |
421 | } |
422 | ||
e4c7ed41 | 423 | static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa) |
e4f67add DS |
424 | { |
425 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
bb3fd687 SH |
426 | struct vxlan_fdb f = { |
427 | .state = NUD_STALE, | |
428 | }; | |
429 | struct vxlan_rdst remote = { | |
e4c7ed41 | 430 | .remote_ip = *ipa, /* goes to NDA_DST */ |
54bfd872 | 431 | .remote_vni = cpu_to_be32(VXLAN_N_VID), |
bb3fd687 | 432 | }; |
3e61aa8f | 433 | |
4c59b7d1 | 434 | vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL); |
e4f67add DS |
435 | } |
436 | ||
437 | static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN]) | |
438 | { | |
bb3fd687 SH |
439 | struct vxlan_fdb f = { |
440 | .state = NUD_STALE, | |
441 | }; | |
9e4b93f9 | 442 | struct vxlan_rdst remote = { }; |
e4f67add | 443 | |
e4f67add DS |
444 | memcpy(f.eth_addr, eth_addr, ETH_ALEN); |
445 | ||
4c59b7d1 | 446 | vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL); |
e4f67add DS |
447 | } |
448 | ||
d342894c | 449 | /* Hash Ethernet address */ |
450 | static u32 eth_hash(const unsigned char *addr) | |
451 | { | |
452 | u64 value = get_unaligned((u64 *)addr); | |
453 | ||
454 | /* only want 6 bytes */ | |
455 | #ifdef __BIG_ENDIAN | |
d342894c | 456 | value >>= 16; |
321fb991 | 457 | #else |
458 | value <<= 16; | |
d342894c | 459 | #endif |
460 | return hash_64(value, FDB_HASH_BITS); | |
461 | } | |
462 | ||
3ad7a4b1 RP |
463 | static u32 eth_vni_hash(const unsigned char *addr, __be32 vni) |
464 | { | |
465 | /* use 1 byte of OUI and 3 bytes of NIC */ | |
466 | u32 key = get_unaligned((u32 *)(addr + 2)); | |
467 | ||
468 | return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1); | |
469 | } | |
470 | ||
fe1e0713 L |
471 | static u32 fdb_head_index(struct vxlan_dev *vxlan, const u8 *mac, __be32 vni) |
472 | { | |
473 | if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) | |
474 | return eth_vni_hash(mac, vni); | |
475 | else | |
476 | return eth_hash(mac); | |
477 | } | |
478 | ||
d342894c | 479 | /* Hash chain to use given mac address */ |
480 | static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan, | |
3ad7a4b1 | 481 | const u8 *mac, __be32 vni) |
d342894c | 482 | { |
fe1e0713 | 483 | return &vxlan->fdb_head[fdb_head_index(vxlan, mac, vni)]; |
d342894c | 484 | } |
485 | ||
486 | /* Look up Ethernet address in forwarding table */ | |
014be2c8 | 487 | static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan, |
3ad7a4b1 | 488 | const u8 *mac, __be32 vni) |
d342894c | 489 | { |
3ad7a4b1 | 490 | struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni); |
d342894c | 491 | struct vxlan_fdb *f; |
d342894c | 492 | |
b67bfe0d | 493 | hlist_for_each_entry_rcu(f, head, hlist) { |
3ad7a4b1 | 494 | if (ether_addr_equal(mac, f->eth_addr)) { |
dc5321d7 | 495 | if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) { |
3ad7a4b1 RP |
496 | if (vni == f->vni) |
497 | return f; | |
498 | } else { | |
499 | return f; | |
500 | } | |
501 | } | |
d342894c | 502 | } |
503 | ||
504 | return NULL; | |
505 | } | |
506 | ||
014be2c8 | 507 | static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan, |
3ad7a4b1 | 508 | const u8 *mac, __be32 vni) |
014be2c8 SS |
509 | { |
510 | struct vxlan_fdb *f; | |
511 | ||
3ad7a4b1 | 512 | f = __vxlan_find_mac(vxlan, mac, vni); |
016f3d18 | 513 | if (f && f->used != jiffies) |
014be2c8 SS |
514 | f->used = jiffies; |
515 | ||
516 | return f; | |
517 | } | |
518 | ||
a5e7c10a MR |
519 | /* caller should hold vxlan->hash_lock */ |
520 | static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f, | |
e4c7ed41 | 521 | union vxlan_addr *ip, __be16 port, |
54bfd872 | 522 | __be32 vni, __u32 ifindex) |
6681712d | 523 | { |
3e61aa8f | 524 | struct vxlan_rdst *rd; |
6681712d | 525 | |
3e61aa8f | 526 | list_for_each_entry(rd, &f->remotes, list) { |
e4c7ed41 | 527 | if (vxlan_addr_equal(&rd->remote_ip, ip) && |
6681712d DS |
528 | rd->remote_port == port && |
529 | rd->remote_vni == vni && | |
530 | rd->remote_ifindex == ifindex) | |
a5e7c10a | 531 | return rd; |
6681712d | 532 | } |
3e61aa8f | 533 | |
a5e7c10a MR |
534 | return NULL; |
535 | } | |
536 | ||
1941f1d6 PM |
537 | int vxlan_fdb_find_uc(struct net_device *dev, const u8 *mac, __be32 vni, |
538 | struct switchdev_notifier_vxlan_fdb_info *fdb_info) | |
539 | { | |
540 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
541 | u8 eth_addr[ETH_ALEN + 2] = { 0 }; | |
542 | struct vxlan_rdst *rdst; | |
543 | struct vxlan_fdb *f; | |
544 | int rc = 0; | |
545 | ||
546 | if (is_multicast_ether_addr(mac) || | |
547 | is_zero_ether_addr(mac)) | |
548 | return -EINVAL; | |
549 | ||
550 | ether_addr_copy(eth_addr, mac); | |
551 | ||
552 | rcu_read_lock(); | |
553 | ||
554 | f = __vxlan_find_mac(vxlan, eth_addr, vni); | |
555 | if (!f) { | |
556 | rc = -ENOENT; | |
557 | goto out; | |
558 | } | |
559 | ||
560 | rdst = first_remote_rcu(f); | |
4c59b7d1 | 561 | vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, NULL, fdb_info); |
1941f1d6 PM |
562 | |
563 | out: | |
564 | rcu_read_unlock(); | |
565 | return rc; | |
566 | } | |
567 | EXPORT_SYMBOL_GPL(vxlan_fdb_find_uc); | |
568 | ||
4f89f5b5 PM |
569 | static int vxlan_fdb_notify_one(struct notifier_block *nb, |
570 | const struct vxlan_dev *vxlan, | |
571 | const struct vxlan_fdb *f, | |
4c59b7d1 PM |
572 | const struct vxlan_rdst *rdst, |
573 | struct netlink_ext_ack *extack) | |
4f89f5b5 PM |
574 | { |
575 | struct switchdev_notifier_vxlan_fdb_info fdb_info; | |
576 | int rc; | |
577 | ||
4c59b7d1 | 578 | vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, extack, &fdb_info); |
4f89f5b5 PM |
579 | rc = nb->notifier_call(nb, SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE, |
580 | &fdb_info); | |
581 | return notifier_to_errno(rc); | |
582 | } | |
583 | ||
584 | int vxlan_fdb_replay(const struct net_device *dev, __be32 vni, | |
4c59b7d1 PM |
585 | struct notifier_block *nb, |
586 | struct netlink_ext_ack *extack) | |
4f89f5b5 PM |
587 | { |
588 | struct vxlan_dev *vxlan; | |
589 | struct vxlan_rdst *rdst; | |
590 | struct vxlan_fdb *f; | |
591 | unsigned int h; | |
592 | int rc = 0; | |
593 | ||
594 | if (!netif_is_vxlan(dev)) | |
595 | return -EINVAL; | |
596 | vxlan = netdev_priv(dev); | |
597 | ||
4f89f5b5 | 598 | for (h = 0; h < FDB_HASH_SIZE; ++h) { |
fe1e0713 | 599 | spin_lock_bh(&vxlan->hash_lock[h]); |
4f89f5b5 PM |
600 | hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) { |
601 | if (f->vni == vni) { | |
602 | list_for_each_entry(rdst, &f->remotes, list) { | |
603 | rc = vxlan_fdb_notify_one(nb, vxlan, | |
4c59b7d1 PM |
604 | f, rdst, |
605 | extack); | |
4f89f5b5 | 606 | if (rc) |
fe1e0713 | 607 | goto unlock; |
4f89f5b5 PM |
608 | } |
609 | } | |
610 | } | |
fe1e0713 | 611 | spin_unlock_bh(&vxlan->hash_lock[h]); |
4f89f5b5 | 612 | } |
fe1e0713 | 613 | return 0; |
4f89f5b5 | 614 | |
fe1e0713 L |
615 | unlock: |
616 | spin_unlock_bh(&vxlan->hash_lock[h]); | |
4f89f5b5 PM |
617 | return rc; |
618 | } | |
619 | EXPORT_SYMBOL_GPL(vxlan_fdb_replay); | |
620 | ||
e5ff4b19 PM |
621 | void vxlan_fdb_clear_offload(const struct net_device *dev, __be32 vni) |
622 | { | |
623 | struct vxlan_dev *vxlan; | |
624 | struct vxlan_rdst *rdst; | |
625 | struct vxlan_fdb *f; | |
626 | unsigned int h; | |
627 | ||
628 | if (!netif_is_vxlan(dev)) | |
629 | return; | |
630 | vxlan = netdev_priv(dev); | |
631 | ||
e5ff4b19 | 632 | for (h = 0; h < FDB_HASH_SIZE; ++h) { |
fe1e0713 | 633 | spin_lock_bh(&vxlan->hash_lock[h]); |
e5ff4b19 PM |
634 | hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) |
635 | if (f->vni == vni) | |
636 | list_for_each_entry(rdst, &f->remotes, list) | |
637 | rdst->offloaded = false; | |
fe1e0713 | 638 | spin_unlock_bh(&vxlan->hash_lock[h]); |
e5ff4b19 | 639 | } |
fe1e0713 | 640 | |
e5ff4b19 PM |
641 | } |
642 | EXPORT_SYMBOL_GPL(vxlan_fdb_clear_offload); | |
643 | ||
906dc186 TR |
644 | /* Replace destination of unicast mac */ |
645 | static int vxlan_fdb_replace(struct vxlan_fdb *f, | |
54bfd872 | 646 | union vxlan_addr *ip, __be16 port, __be32 vni, |
ccdfd4f7 | 647 | __u32 ifindex, struct vxlan_rdst *oldrd) |
906dc186 TR |
648 | { |
649 | struct vxlan_rdst *rd; | |
650 | ||
651 | rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex); | |
652 | if (rd) | |
653 | return 0; | |
654 | ||
655 | rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list); | |
656 | if (!rd) | |
657 | return 0; | |
0c1d70af | 658 | |
ccdfd4f7 | 659 | *oldrd = *rd; |
0c1d70af | 660 | dst_cache_reset(&rd->dst_cache); |
e4c7ed41 | 661 | rd->remote_ip = *ip; |
906dc186 TR |
662 | rd->remote_port = port; |
663 | rd->remote_vni = vni; | |
664 | rd->remote_ifindex = ifindex; | |
6ad0b5a4 | 665 | rd->offloaded = false; |
906dc186 TR |
666 | return 1; |
667 | } | |
668 | ||
a5e7c10a MR |
669 | /* Add/update destinations for multicast */ |
670 | static int vxlan_fdb_append(struct vxlan_fdb *f, | |
54bfd872 | 671 | union vxlan_addr *ip, __be16 port, __be32 vni, |
9e4b93f9 | 672 | __u32 ifindex, struct vxlan_rdst **rdp) |
a5e7c10a MR |
673 | { |
674 | struct vxlan_rdst *rd; | |
675 | ||
676 | rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex); | |
677 | if (rd) | |
678 | return 0; | |
679 | ||
6681712d DS |
680 | rd = kmalloc(sizeof(*rd), GFP_ATOMIC); |
681 | if (rd == NULL) | |
682 | return -ENOBUFS; | |
0c1d70af PA |
683 | |
684 | if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) { | |
685 | kfree(rd); | |
686 | return -ENOBUFS; | |
687 | } | |
688 | ||
e4c7ed41 | 689 | rd->remote_ip = *ip; |
6681712d | 690 | rd->remote_port = port; |
0efe1173 | 691 | rd->offloaded = false; |
6681712d DS |
692 | rd->remote_vni = vni; |
693 | rd->remote_ifindex = ifindex; | |
3e61aa8f SH |
694 | |
695 | list_add_tail_rcu(&rd->list, &f->remotes); | |
696 | ||
9e4b93f9 | 697 | *rdp = rd; |
6681712d DS |
698 | return 1; |
699 | } | |
700 | ||
dfd8645e TH |
701 | static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb, |
702 | unsigned int off, | |
703 | struct vxlanhdr *vh, size_t hdrlen, | |
54bfd872 JB |
704 | __be32 vni_field, |
705 | struct gro_remcsum *grc, | |
0ace2ca8 | 706 | bool nopartial) |
dfd8645e | 707 | { |
b7fe10e5 | 708 | size_t start, offset; |
dfd8645e TH |
709 | |
710 | if (skb->remcsum_offload) | |
b7fe10e5 | 711 | return vh; |
dfd8645e TH |
712 | |
713 | if (!NAPI_GRO_CB(skb)->csum_valid) | |
714 | return NULL; | |
715 | ||
54bfd872 JB |
716 | start = vxlan_rco_start(vni_field); |
717 | offset = start + vxlan_rco_offset(vni_field); | |
dfd8645e | 718 | |
b7fe10e5 TH |
719 | vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen, |
720 | start, offset, grc, nopartial); | |
dfd8645e TH |
721 | |
722 | skb->remcsum_offload = 1; | |
723 | ||
724 | return vh; | |
725 | } | |
726 | ||
d4546c25 DM |
727 | static struct sk_buff *vxlan_gro_receive(struct sock *sk, |
728 | struct list_head *head, | |
729 | struct sk_buff *skb) | |
dc01e7d3 | 730 | { |
d4546c25 DM |
731 | struct sk_buff *pp = NULL; |
732 | struct sk_buff *p; | |
dc01e7d3 | 733 | struct vxlanhdr *vh, *vh2; |
9b174d88 | 734 | unsigned int hlen, off_vx; |
dc01e7d3 | 735 | int flush = 1; |
5602c48c | 736 | struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk); |
54bfd872 | 737 | __be32 flags; |
26c4f7da TH |
738 | struct gro_remcsum grc; |
739 | ||
740 | skb_gro_remcsum_init(&grc); | |
dc01e7d3 OG |
741 | |
742 | off_vx = skb_gro_offset(skb); | |
743 | hlen = off_vx + sizeof(*vh); | |
744 | vh = skb_gro_header_fast(skb, off_vx); | |
745 | if (skb_gro_header_hard(skb, hlen)) { | |
746 | vh = skb_gro_header_slow(skb, hlen, off_vx); | |
747 | if (unlikely(!vh)) | |
748 | goto out; | |
749 | } | |
dc01e7d3 | 750 | |
dfd8645e TH |
751 | skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr)); |
752 | ||
54bfd872 | 753 | flags = vh->vx_flags; |
dfd8645e TH |
754 | |
755 | if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) { | |
756 | vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr), | |
54bfd872 | 757 | vh->vx_vni, &grc, |
0ace2ca8 TH |
758 | !!(vs->flags & |
759 | VXLAN_F_REMCSUM_NOPARTIAL)); | |
dfd8645e TH |
760 | |
761 | if (!vh) | |
762 | goto out; | |
763 | } | |
764 | ||
b7fe10e5 TH |
765 | skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */ |
766 | ||
d4546c25 | 767 | list_for_each_entry(p, head, list) { |
dc01e7d3 OG |
768 | if (!NAPI_GRO_CB(p)->same_flow) |
769 | continue; | |
770 | ||
771 | vh2 = (struct vxlanhdr *)(p->data + off_vx); | |
3511494c TG |
772 | if (vh->vx_flags != vh2->vx_flags || |
773 | vh->vx_vni != vh2->vx_vni) { | |
dc01e7d3 OG |
774 | NAPI_GRO_CB(p)->same_flow = 0; |
775 | continue; | |
776 | } | |
dc01e7d3 OG |
777 | } |
778 | ||
fcd91dd4 | 779 | pp = call_gro_receive(eth_gro_receive, head, skb); |
c194cf93 | 780 | flush = 0; |
dc01e7d3 | 781 | |
dc01e7d3 | 782 | out: |
603d4cf8 | 783 | skb_gro_flush_final_remcsum(skb, pp, flush, &grc); |
dc01e7d3 OG |
784 | |
785 | return pp; | |
786 | } | |
787 | ||
5602c48c | 788 | static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff) |
dc01e7d3 | 789 | { |
229740c6 JR |
790 | /* Sets 'skb->inner_mac_header' since we are always called with |
791 | * 'skb->encapsulation' set. | |
792 | */ | |
9b174d88 | 793 | return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr)); |
dc01e7d3 OG |
794 | } |
795 | ||
1d7a5526 | 796 | static struct vxlan_fdb *vxlan_fdb_alloc(const u8 *mac, __u16 state, |
45598c1c | 797 | __be32 src_vni, __u16 ndm_flags) |
25e20e73 RP |
798 | { |
799 | struct vxlan_fdb *f; | |
800 | ||
801 | f = kmalloc(sizeof(*f), GFP_ATOMIC); | |
802 | if (!f) | |
803 | return NULL; | |
804 | f->state = state; | |
805 | f->flags = ndm_flags; | |
806 | f->updated = f->used = jiffies; | |
807 | f->vni = src_vni; | |
808 | INIT_LIST_HEAD(&f->remotes); | |
809 | memcpy(f->eth_addr, mac, ETH_ALEN); | |
810 | ||
811 | return f; | |
812 | } | |
813 | ||
7c31e54a TY |
814 | static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac, |
815 | __be32 src_vni, struct vxlan_fdb *f) | |
816 | { | |
817 | ++vxlan->addrcnt; | |
818 | hlist_add_head_rcu(&f->hlist, | |
819 | vxlan_fdb_head(vxlan, mac, src_vni)); | |
820 | } | |
821 | ||
d342894c | 822 | static int vxlan_fdb_create(struct vxlan_dev *vxlan, |
25e20e73 RP |
823 | const u8 *mac, union vxlan_addr *ip, |
824 | __u16 state, __be16 port, __be32 src_vni, | |
45598c1c | 825 | __be32 vni, __u32 ifindex, __u16 ndm_flags, |
25e20e73 RP |
826 | struct vxlan_fdb **fdb) |
827 | { | |
828 | struct vxlan_rdst *rd = NULL; | |
829 | struct vxlan_fdb *f; | |
830 | int rc; | |
831 | ||
832 | if (vxlan->cfg.addrmax && | |
833 | vxlan->addrcnt >= vxlan->cfg.addrmax) | |
834 | return -ENOSPC; | |
835 | ||
836 | netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip); | |
1d7a5526 | 837 | f = vxlan_fdb_alloc(mac, state, src_vni, ndm_flags); |
25e20e73 RP |
838 | if (!f) |
839 | return -ENOMEM; | |
840 | ||
841 | rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd); | |
842 | if (rc < 0) { | |
843 | kfree(f); | |
844 | return rc; | |
845 | } | |
846 | ||
25e20e73 RP |
847 | *fdb = f; |
848 | ||
849 | return 0; | |
850 | } | |
851 | ||
7c31e54a | 852 | static void __vxlan_fdb_free(struct vxlan_fdb *f) |
c2b200e0 | 853 | { |
c2b200e0 PM |
854 | struct vxlan_rdst *rd, *nd; |
855 | ||
856 | list_for_each_entry_safe(rd, nd, &f->remotes, list) { | |
857 | dst_cache_destroy(&rd->dst_cache); | |
858 | kfree(rd); | |
859 | } | |
860 | kfree(f); | |
861 | } | |
862 | ||
7c31e54a TY |
863 | static void vxlan_fdb_free(struct rcu_head *head) |
864 | { | |
865 | struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu); | |
866 | ||
867 | __vxlan_fdb_free(f); | |
868 | } | |
869 | ||
c2b200e0 PM |
870 | static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f, |
871 | bool do_notify, bool swdev_notify) | |
872 | { | |
873 | struct vxlan_rdst *rd; | |
874 | ||
875 | netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr); | |
876 | ||
877 | --vxlan->addrcnt; | |
878 | if (do_notify) | |
879 | list_for_each_entry(rd, &f->remotes, list) | |
880 | vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, | |
4c59b7d1 | 881 | swdev_notify, NULL); |
c2b200e0 PM |
882 | |
883 | hlist_del_rcu(&f->hlist); | |
884 | call_rcu(&f->rcu, vxlan_fdb_free); | |
885 | } | |
886 | ||
fc4aa1ca PM |
887 | static void vxlan_dst_free(struct rcu_head *head) |
888 | { | |
889 | struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu); | |
890 | ||
891 | dst_cache_destroy(&rd->dst_cache); | |
892 | kfree(rd); | |
893 | } | |
894 | ||
a76d1ca2 PM |
895 | static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan, |
896 | union vxlan_addr *ip, | |
897 | __u16 state, __u16 flags, | |
898 | __be16 port, __be32 vni, | |
899 | __u32 ifindex, __u16 ndm_flags, | |
900 | struct vxlan_fdb *f, | |
4c59b7d1 PM |
901 | bool swdev_notify, |
902 | struct netlink_ext_ack *extack) | |
d342894c | 903 | { |
45598c1c | 904 | __u16 fdb_flags = (ndm_flags & ~NTF_USE); |
9e4b93f9 | 905 | struct vxlan_rdst *rd = NULL; |
ccdfd4f7 | 906 | struct vxlan_rdst oldrd; |
d342894c | 907 | int notify = 0; |
61f46fe8 PM |
908 | int rc = 0; |
909 | int err; | |
d342894c | 910 | |
a76d1ca2 PM |
911 | /* Do not allow an externally learned entry to take over an entry added |
912 | * by the user. | |
913 | */ | |
914 | if (!(fdb_flags & NTF_EXT_LEARNED) || | |
915 | !(f->flags & NTF_VXLAN_ADDED_BY_USER)) { | |
916 | if (f->state != state) { | |
917 | f->state = state; | |
918 | f->updated = jiffies; | |
919 | notify = 1; | |
ae884082 | 920 | } |
a76d1ca2 PM |
921 | if (f->flags != fdb_flags) { |
922 | f->flags = fdb_flags; | |
923 | f->updated = jiffies; | |
924 | notify = 1; | |
906dc186 | 925 | } |
a76d1ca2 | 926 | } |
6681712d | 927 | |
a76d1ca2 PM |
928 | if ((flags & NLM_F_REPLACE)) { |
929 | /* Only change unicasts */ | |
930 | if (!(is_multicast_ether_addr(f->eth_addr) || | |
931 | is_zero_ether_addr(f->eth_addr))) { | |
932 | rc = vxlan_fdb_replace(f, ip, port, vni, | |
ccdfd4f7 | 933 | ifindex, &oldrd); |
6681712d | 934 | notify |= rc; |
a76d1ca2 | 935 | } else { |
906dc186 | 936 | return -EOPNOTSUPP; |
a76d1ca2 PM |
937 | } |
938 | } | |
939 | if ((flags & NLM_F_APPEND) && | |
940 | (is_multicast_ether_addr(f->eth_addr) || | |
941 | is_zero_ether_addr(f->eth_addr))) { | |
942 | rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd); | |
906dc186 | 943 | |
25e20e73 | 944 | if (rc < 0) |
17b46365 | 945 | return rc; |
a76d1ca2 | 946 | notify |= rc; |
d342894c | 947 | } |
948 | ||
a76d1ca2 PM |
949 | if (ndm_flags & NTF_USE) |
950 | f->used = jiffies; | |
951 | ||
9e4b93f9 ND |
952 | if (notify) { |
953 | if (rd == NULL) | |
954 | rd = first_remote_rtnl(f); | |
a76d1ca2 | 955 | |
61f46fe8 | 956 | err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH, |
4c59b7d1 | 957 | swdev_notify, extack); |
61f46fe8 PM |
958 | if (err) |
959 | goto err_notify; | |
9e4b93f9 | 960 | } |
d342894c | 961 | |
962 | return 0; | |
61f46fe8 PM |
963 | |
964 | err_notify: | |
965 | if ((flags & NLM_F_REPLACE) && rc) | |
966 | *rd = oldrd; | |
fc4aa1ca | 967 | else if ((flags & NLM_F_APPEND) && rc) { |
61f46fe8 | 968 | list_del_rcu(&rd->list); |
fc4aa1ca PM |
969 | call_rcu(&rd->rcu, vxlan_dst_free); |
970 | } | |
61f46fe8 | 971 | return err; |
d342894c | 972 | } |
973 | ||
a76d1ca2 PM |
974 | static int vxlan_fdb_update_create(struct vxlan_dev *vxlan, |
975 | const u8 *mac, union vxlan_addr *ip, | |
976 | __u16 state, __u16 flags, | |
977 | __be16 port, __be32 src_vni, __be32 vni, | |
978 | __u32 ifindex, __u16 ndm_flags, | |
4c59b7d1 PM |
979 | bool swdev_notify, |
980 | struct netlink_ext_ack *extack) | |
a76d1ca2 PM |
981 | { |
982 | __u16 fdb_flags = (ndm_flags & ~NTF_USE); | |
983 | struct vxlan_fdb *f; | |
984 | int rc; | |
985 | ||
986 | /* Disallow replace to add a multicast entry */ | |
987 | if ((flags & NLM_F_REPLACE) && | |
988 | (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac))) | |
989 | return -EOPNOTSUPP; | |
990 | ||
991 | netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip); | |
992 | rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni, | |
993 | vni, ifindex, fdb_flags, &f); | |
994 | if (rc < 0) | |
995 | return rc; | |
996 | ||
7c31e54a | 997 | vxlan_fdb_insert(vxlan, mac, src_vni, f); |
61f46fe8 | 998 | rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH, |
4c59b7d1 | 999 | swdev_notify, extack); |
61f46fe8 PM |
1000 | if (rc) |
1001 | goto err_notify; | |
1002 | ||
a76d1ca2 | 1003 | return 0; |
61f46fe8 PM |
1004 | |
1005 | err_notify: | |
1006 | vxlan_fdb_destroy(vxlan, f, false, false); | |
1007 | return rc; | |
a76d1ca2 PM |
1008 | } |
1009 | ||
1010 | /* Add new entry to forwarding table -- assumes lock held */ | |
1011 | static int vxlan_fdb_update(struct vxlan_dev *vxlan, | |
1012 | const u8 *mac, union vxlan_addr *ip, | |
1013 | __u16 state, __u16 flags, | |
1014 | __be16 port, __be32 src_vni, __be32 vni, | |
1015 | __u32 ifindex, __u16 ndm_flags, | |
4c59b7d1 PM |
1016 | bool swdev_notify, |
1017 | struct netlink_ext_ack *extack) | |
a76d1ca2 PM |
1018 | { |
1019 | struct vxlan_fdb *f; | |
1020 | ||
1021 | f = __vxlan_find_mac(vxlan, mac, src_vni); | |
1022 | if (f) { | |
1023 | if (flags & NLM_F_EXCL) { | |
1024 | netdev_dbg(vxlan->dev, | |
1025 | "lost race to create %pM\n", mac); | |
1026 | return -EEXIST; | |
1027 | } | |
1028 | ||
1029 | return vxlan_fdb_update_existing(vxlan, ip, state, flags, port, | |
1030 | vni, ifindex, ndm_flags, f, | |
4c59b7d1 | 1031 | swdev_notify, extack); |
a76d1ca2 PM |
1032 | } else { |
1033 | if (!(flags & NLM_F_CREATE)) | |
1034 | return -ENOENT; | |
1035 | ||
1036 | return vxlan_fdb_update_create(vxlan, mac, ip, state, flags, | |
1037 | port, src_vni, vni, ifindex, | |
4c59b7d1 | 1038 | ndm_flags, swdev_notify, extack); |
a76d1ca2 PM |
1039 | } |
1040 | } | |
1041 | ||
35cf2845 | 1042 | static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f, |
0e6160f3 | 1043 | struct vxlan_rdst *rd, bool swdev_notify) |
35cf2845 LR |
1044 | { |
1045 | list_del_rcu(&rd->list); | |
4c59b7d1 | 1046 | vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL); |
35cf2845 LR |
1047 | call_rcu(&rd->rcu, vxlan_dst_free); |
1048 | } | |
1049 | ||
f0b074be | 1050 | static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan, |
3ad7a4b1 RP |
1051 | union vxlan_addr *ip, __be16 *port, __be32 *src_vni, |
1052 | __be32 *vni, u32 *ifindex) | |
d342894c | 1053 | { |
6681712d | 1054 | struct net *net = dev_net(vxlan->dev); |
e4c7ed41 | 1055 | int err; |
d342894c | 1056 | |
f0b074be | 1057 | if (tb[NDA_DST]) { |
e4c7ed41 CW |
1058 | err = vxlan_nla_get_addr(ip, tb[NDA_DST]); |
1059 | if (err) | |
1060 | return err; | |
f0b074be | 1061 | } else { |
e4c7ed41 | 1062 | union vxlan_addr *remote = &vxlan->default_dst.remote_ip; |
98c81476 | 1063 | |
e4c7ed41 CW |
1064 | if (remote->sa.sa_family == AF_INET) { |
1065 | ip->sin.sin_addr.s_addr = htonl(INADDR_ANY); | |
1066 | ip->sa.sa_family = AF_INET; | |
1067 | #if IS_ENABLED(CONFIG_IPV6) | |
1068 | } else { | |
1069 | ip->sin6.sin6_addr = in6addr_any; | |
1070 | ip->sa.sa_family = AF_INET6; | |
1071 | #endif | |
1072 | } | |
f0b074be | 1073 | } |
d342894c | 1074 | |
6681712d | 1075 | if (tb[NDA_PORT]) { |
73cf3317 | 1076 | if (nla_len(tb[NDA_PORT]) != sizeof(__be16)) |
6681712d | 1077 | return -EINVAL; |
f0b074be MR |
1078 | *port = nla_get_be16(tb[NDA_PORT]); |
1079 | } else { | |
0dfbdf41 | 1080 | *port = vxlan->cfg.dst_port; |
f0b074be | 1081 | } |
6681712d DS |
1082 | |
1083 | if (tb[NDA_VNI]) { | |
1084 | if (nla_len(tb[NDA_VNI]) != sizeof(u32)) | |
1085 | return -EINVAL; | |
54bfd872 | 1086 | *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI])); |
f0b074be MR |
1087 | } else { |
1088 | *vni = vxlan->default_dst.remote_vni; | |
1089 | } | |
6681712d | 1090 | |
3ad7a4b1 RP |
1091 | if (tb[NDA_SRC_VNI]) { |
1092 | if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32)) | |
1093 | return -EINVAL; | |
1094 | *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI])); | |
1095 | } else { | |
1096 | *src_vni = vxlan->default_dst.remote_vni; | |
1097 | } | |
1098 | ||
6681712d | 1099 | if (tb[NDA_IFINDEX]) { |
5abb0029 | 1100 | struct net_device *tdev; |
6681712d DS |
1101 | |
1102 | if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32)) | |
1103 | return -EINVAL; | |
f0b074be | 1104 | *ifindex = nla_get_u32(tb[NDA_IFINDEX]); |
73763949 | 1105 | tdev = __dev_get_by_index(net, *ifindex); |
5abb0029 | 1106 | if (!tdev) |
6681712d | 1107 | return -EADDRNOTAVAIL; |
f0b074be MR |
1108 | } else { |
1109 | *ifindex = 0; | |
1110 | } | |
1111 | ||
1112 | return 0; | |
1113 | } | |
1114 | ||
1115 | /* Add static entry (via netlink) */ | |
1116 | static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[], | |
1117 | struct net_device *dev, | |
87b0984e PM |
1118 | const unsigned char *addr, u16 vid, u16 flags, |
1119 | struct netlink_ext_ack *extack) | |
f0b074be MR |
1120 | { |
1121 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1122 | /* struct net *net = dev_net(vxlan->dev); */ | |
e4c7ed41 | 1123 | union vxlan_addr ip; |
f0b074be | 1124 | __be16 port; |
3ad7a4b1 | 1125 | __be32 src_vni, vni; |
54bfd872 | 1126 | u32 ifindex; |
fe1e0713 | 1127 | u32 hash_index; |
f0b074be MR |
1128 | int err; |
1129 | ||
1130 | if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) { | |
1131 | pr_info("RTM_NEWNEIGH with invalid state %#x\n", | |
1132 | ndm->ndm_state); | |
1133 | return -EINVAL; | |
1134 | } | |
1135 | ||
1136 | if (tb[NDA_DST] == NULL) | |
1137 | return -EINVAL; | |
1138 | ||
3ad7a4b1 | 1139 | err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex); |
f0b074be MR |
1140 | if (err) |
1141 | return err; | |
6681712d | 1142 | |
5933a7bb MR |
1143 | if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family) |
1144 | return -EAFNOSUPPORT; | |
1145 | ||
fe1e0713 L |
1146 | hash_index = fdb_head_index(vxlan, addr, src_vni); |
1147 | spin_lock_bh(&vxlan->hash_lock[hash_index]); | |
25e20e73 | 1148 | err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags, |
45598c1c PM |
1149 | port, src_vni, vni, ifindex, |
1150 | ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER, | |
4c59b7d1 | 1151 | true, extack); |
fe1e0713 | 1152 | spin_unlock_bh(&vxlan->hash_lock[hash_index]); |
d342894c | 1153 | |
1154 | return err; | |
1155 | } | |
1156 | ||
3ad7a4b1 RP |
1157 | static int __vxlan_fdb_delete(struct vxlan_dev *vxlan, |
1158 | const unsigned char *addr, union vxlan_addr ip, | |
fc39c38b | 1159 | __be16 port, __be32 src_vni, __be32 vni, |
0e6160f3 | 1160 | u32 ifindex, bool swdev_notify) |
d342894c | 1161 | { |
d342894c | 1162 | struct vxlan_fdb *f; |
bc7892ba | 1163 | struct vxlan_rdst *rd = NULL; |
3ad7a4b1 | 1164 | int err = -ENOENT; |
bc7892ba | 1165 | |
3ad7a4b1 | 1166 | f = vxlan_find_mac(vxlan, addr, src_vni); |
bc7892ba | 1167 | if (!f) |
3ad7a4b1 | 1168 | return err; |
bc7892ba | 1169 | |
e4c7ed41 CW |
1170 | if (!vxlan_addr_any(&ip)) { |
1171 | rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex); | |
bc7892ba MR |
1172 | if (!rd) |
1173 | goto out; | |
1174 | } | |
1175 | ||
bc7892ba MR |
1176 | /* remove a destination if it's not the only one on the list, |
1177 | * otherwise destroy the fdb entry | |
1178 | */ | |
1179 | if (rd && !list_is_singular(&f->remotes)) { | |
0e6160f3 | 1180 | vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify); |
bc7892ba | 1181 | goto out; |
d342894c | 1182 | } |
bc7892ba | 1183 | |
0e6160f3 | 1184 | vxlan_fdb_destroy(vxlan, f, true, swdev_notify); |
bc7892ba MR |
1185 | |
1186 | out: | |
3ad7a4b1 RP |
1187 | return 0; |
1188 | } | |
1189 | ||
1190 | /* Delete entry (via netlink) */ | |
1191 | static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[], | |
1192 | struct net_device *dev, | |
1193 | const unsigned char *addr, u16 vid) | |
1194 | { | |
1195 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1196 | union vxlan_addr ip; | |
1197 | __be32 src_vni, vni; | |
1198 | __be16 port; | |
1199 | u32 ifindex; | |
fe1e0713 | 1200 | u32 hash_index; |
3ad7a4b1 RP |
1201 | int err; |
1202 | ||
1203 | err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex); | |
1204 | if (err) | |
1205 | return err; | |
1206 | ||
fe1e0713 L |
1207 | hash_index = fdb_head_index(vxlan, addr, src_vni); |
1208 | spin_lock_bh(&vxlan->hash_lock[hash_index]); | |
0e6160f3 PM |
1209 | err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex, |
1210 | true); | |
fe1e0713 | 1211 | spin_unlock_bh(&vxlan->hash_lock[hash_index]); |
d342894c | 1212 | |
1213 | return err; | |
1214 | } | |
1215 | ||
1216 | /* Dump forwarding table */ | |
1217 | static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb, | |
5d5eacb3 | 1218 | struct net_device *dev, |
d297653d | 1219 | struct net_device *filter_dev, int *idx) |
d342894c | 1220 | { |
1221 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1222 | unsigned int h; | |
d297653d | 1223 | int err = 0; |
d342894c | 1224 | |
1225 | for (h = 0; h < FDB_HASH_SIZE; ++h) { | |
1226 | struct vxlan_fdb *f; | |
d342894c | 1227 | |
b67bfe0d | 1228 | hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) { |
6681712d | 1229 | struct vxlan_rdst *rd; |
6681712d | 1230 | |
3e61aa8f | 1231 | list_for_each_entry_rcu(rd, &f->remotes, list) { |
d297653d | 1232 | if (*idx < cb->args[2]) |
07a51cd3 AW |
1233 | goto skip; |
1234 | ||
6681712d DS |
1235 | err = vxlan_fdb_info(skb, vxlan, f, |
1236 | NETLINK_CB(cb->skb).portid, | |
1237 | cb->nlh->nlmsg_seq, | |
1238 | RTM_NEWNEIGH, | |
1239 | NLM_F_MULTI, rd); | |
d297653d | 1240 | if (err < 0) |
3e61aa8f | 1241 | goto out; |
3e61aa8f | 1242 | skip: |
d297653d | 1243 | *idx += 1; |
07a51cd3 | 1244 | } |
d342894c | 1245 | } |
1246 | } | |
3e61aa8f | 1247 | out: |
d297653d | 1248 | return err; |
d342894c | 1249 | } |
1250 | ||
474c3c89 RP |
1251 | static int vxlan_fdb_get(struct sk_buff *skb, |
1252 | struct nlattr *tb[], | |
1253 | struct net_device *dev, | |
1254 | const unsigned char *addr, | |
1255 | u16 vid, u32 portid, u32 seq, | |
1256 | struct netlink_ext_ack *extack) | |
1257 | { | |
1258 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1259 | struct vxlan_fdb *f; | |
1260 | __be32 vni; | |
1261 | int err; | |
1262 | ||
1263 | if (tb[NDA_VNI]) | |
1264 | vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI])); | |
1265 | else | |
1266 | vni = vxlan->default_dst.remote_vni; | |
1267 | ||
1268 | rcu_read_lock(); | |
1269 | ||
1270 | f = __vxlan_find_mac(vxlan, addr, vni); | |
1271 | if (!f) { | |
1272 | NL_SET_ERR_MSG(extack, "Fdb entry not found"); | |
1273 | err = -ENOENT; | |
1274 | goto errout; | |
1275 | } | |
1276 | ||
1277 | err = vxlan_fdb_info(skb, vxlan, f, portid, seq, | |
1278 | RTM_NEWNEIGH, 0, first_remote_rcu(f)); | |
1279 | errout: | |
1280 | rcu_read_unlock(); | |
1281 | return err; | |
1282 | } | |
1283 | ||
d342894c | 1284 | /* Watch incoming packets to learn mapping between Ethernet address |
1285 | * and Tunnel endpoint. | |
c4b49512 | 1286 | * Return true if packet is bogus and should be dropped. |
d342894c | 1287 | */ |
26a41ae6 | 1288 | static bool vxlan_snoop(struct net_device *dev, |
3ad7a4b1 | 1289 | union vxlan_addr *src_ip, const u8 *src_mac, |
87613de9 | 1290 | u32 src_ifindex, __be32 vni) |
d342894c | 1291 | { |
1292 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1293 | struct vxlan_fdb *f; | |
87613de9 MS |
1294 | u32 ifindex = 0; |
1295 | ||
1296 | #if IS_ENABLED(CONFIG_IPV6) | |
1297 | if (src_ip->sa.sa_family == AF_INET6 && | |
1298 | (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)) | |
1299 | ifindex = src_ifindex; | |
1300 | #endif | |
d342894c | 1301 | |
3ad7a4b1 | 1302 | f = vxlan_find_mac(vxlan, src_mac, vni); |
d342894c | 1303 | if (likely(f)) { |
5ca5461c | 1304 | struct vxlan_rdst *rdst = first_remote_rcu(f); |
3e61aa8f | 1305 | |
87613de9 MS |
1306 | if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) && |
1307 | rdst->remote_ifindex == ifindex)) | |
26a41ae6 | 1308 | return false; |
1309 | ||
1310 | /* Don't migrate static entries, drop packets */ | |
e0090a9e | 1311 | if (f->state & (NUD_PERMANENT | NUD_NOARP)) |
26a41ae6 | 1312 | return true; |
d342894c | 1313 | |
1314 | if (net_ratelimit()) | |
1315 | netdev_info(dev, | |
e4c7ed41 | 1316 | "%pM migrated from %pIS to %pIS\n", |
a4870f79 | 1317 | src_mac, &rdst->remote_ip.sa, &src_ip->sa); |
d342894c | 1318 | |
e4c7ed41 | 1319 | rdst->remote_ip = *src_ip; |
d342894c | 1320 | f->updated = jiffies; |
4c59b7d1 | 1321 | vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL); |
d342894c | 1322 | } else { |
fe1e0713 L |
1323 | u32 hash_index = fdb_head_index(vxlan, src_mac, vni); |
1324 | ||
d342894c | 1325 | /* learned new entry */ |
fe1e0713 | 1326 | spin_lock(&vxlan->hash_lock[hash_index]); |
3bf74b1a | 1327 | |
1328 | /* close off race between vxlan_flush and incoming packets */ | |
1329 | if (netif_running(dev)) | |
25e20e73 | 1330 | vxlan_fdb_update(vxlan, src_mac, src_ip, |
3bf74b1a | 1331 | NUD_REACHABLE, |
1332 | NLM_F_EXCL|NLM_F_CREATE, | |
0dfbdf41 | 1333 | vxlan->cfg.dst_port, |
3ad7a4b1 | 1334 | vni, |
3bf74b1a | 1335 | vxlan->default_dst.remote_vni, |
4c59b7d1 | 1336 | ifindex, NTF_SELF, true, NULL); |
fe1e0713 | 1337 | spin_unlock(&vxlan->hash_lock[hash_index]); |
d342894c | 1338 | } |
26a41ae6 | 1339 | |
1340 | return false; | |
d342894c | 1341 | } |
1342 | ||
d342894c | 1343 | /* See if multicast group is already in use by other ID */ |
95ab0991 | 1344 | static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev) |
d342894c | 1345 | { |
553675fb | 1346 | struct vxlan_dev *vxlan; |
c6fcc4fc | 1347 | struct vxlan_sock *sock4; |
4053ab1b AB |
1348 | #if IS_ENABLED(CONFIG_IPV6) |
1349 | struct vxlan_sock *sock6; | |
1350 | #endif | |
b1be00a6 | 1351 | unsigned short family = dev->default_dst.remote_ip.sa.sa_family; |
d342894c | 1352 | |
c6fcc4fc | 1353 | sock4 = rtnl_dereference(dev->vn4_sock); |
1354 | ||
95ab0991 G |
1355 | /* The vxlan_sock is only used by dev, leaving group has |
1356 | * no effect on other vxlan devices. | |
1357 | */ | |
66af846f | 1358 | if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1) |
95ab0991 | 1359 | return false; |
b1be00a6 | 1360 | #if IS_ENABLED(CONFIG_IPV6) |
c6fcc4fc | 1361 | sock6 = rtnl_dereference(dev->vn6_sock); |
66af846f | 1362 | if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1) |
b1be00a6 JB |
1363 | return false; |
1364 | #endif | |
95ab0991 | 1365 | |
553675fb | 1366 | list_for_each_entry(vxlan, &vn->vxlan_list, next) { |
95ab0991 | 1367 | if (!netif_running(vxlan->dev) || vxlan == dev) |
553675fb | 1368 | continue; |
d342894c | 1369 | |
c6fcc4fc | 1370 | if (family == AF_INET && |
1371 | rtnl_dereference(vxlan->vn4_sock) != sock4) | |
95ab0991 | 1372 | continue; |
b1be00a6 | 1373 | #if IS_ENABLED(CONFIG_IPV6) |
c6fcc4fc | 1374 | if (family == AF_INET6 && |
1375 | rtnl_dereference(vxlan->vn6_sock) != sock6) | |
b1be00a6 JB |
1376 | continue; |
1377 | #endif | |
95ab0991 G |
1378 | |
1379 | if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip, | |
1380 | &dev->default_dst.remote_ip)) | |
1381 | continue; | |
1382 | ||
1383 | if (vxlan->default_dst.remote_ifindex != | |
1384 | dev->default_dst.remote_ifindex) | |
1385 | continue; | |
1386 | ||
1387 | return true; | |
553675fb | 1388 | } |
d342894c | 1389 | |
1390 | return false; | |
1391 | } | |
1392 | ||
544a773a | 1393 | static bool __vxlan_sock_release_prep(struct vxlan_sock *vs) |
7c47cedf | 1394 | { |
b1be00a6 | 1395 | struct vxlan_net *vn; |
012a5729 | 1396 | |
b1be00a6 | 1397 | if (!vs) |
544a773a | 1398 | return false; |
66af846f | 1399 | if (!refcount_dec_and_test(&vs->refcnt)) |
544a773a | 1400 | return false; |
d342894c | 1401 | |
b1be00a6 | 1402 | vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id); |
1c51a915 | 1403 | spin_lock(&vn->sock_lock); |
7c47cedf | 1404 | hlist_del_rcu(&vs->hlist); |
e7b3db5e | 1405 | udp_tunnel_notify_del_rx_port(vs->sock, |
b9adcd69 AD |
1406 | (vs->flags & VXLAN_F_GPE) ? |
1407 | UDP_TUNNEL_TYPE_VXLAN_GPE : | |
e7b3db5e | 1408 | UDP_TUNNEL_TYPE_VXLAN); |
1c51a915 SH |
1409 | spin_unlock(&vn->sock_lock); |
1410 | ||
544a773a | 1411 | return true; |
d342894c | 1412 | } |
1413 | ||
b1be00a6 JB |
1414 | static void vxlan_sock_release(struct vxlan_dev *vxlan) |
1415 | { | |
c6fcc4fc | 1416 | struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock); |
b1be00a6 | 1417 | #if IS_ENABLED(CONFIG_IPV6) |
c6fcc4fc | 1418 | struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock); |
1419 | ||
57d88182 | 1420 | RCU_INIT_POINTER(vxlan->vn6_sock, NULL); |
544a773a HFS |
1421 | #endif |
1422 | ||
57d88182 | 1423 | RCU_INIT_POINTER(vxlan->vn4_sock, NULL); |
544a773a HFS |
1424 | synchronize_net(); |
1425 | ||
a53cb29b MB |
1426 | vxlan_vs_del_dev(vxlan); |
1427 | ||
c6fcc4fc | 1428 | if (__vxlan_sock_release_prep(sock4)) { |
1429 | udp_tunnel_sock_release(sock4->sock); | |
1430 | kfree(sock4); | |
544a773a HFS |
1431 | } |
1432 | ||
1433 | #if IS_ENABLED(CONFIG_IPV6) | |
c6fcc4fc | 1434 | if (__vxlan_sock_release_prep(sock6)) { |
1435 | udp_tunnel_sock_release(sock6->sock); | |
1436 | kfree(sock6); | |
544a773a | 1437 | } |
b1be00a6 JB |
1438 | #endif |
1439 | } | |
1440 | ||
56ef9c90 | 1441 | /* Update multicast group membership when first VNI on |
c4b49512 | 1442 | * multicast address is brought up |
7c47cedf | 1443 | */ |
56ef9c90 | 1444 | static int vxlan_igmp_join(struct vxlan_dev *vxlan) |
d342894c | 1445 | { |
b1be00a6 | 1446 | struct sock *sk; |
e4c7ed41 CW |
1447 | union vxlan_addr *ip = &vxlan->default_dst.remote_ip; |
1448 | int ifindex = vxlan->default_dst.remote_ifindex; | |
149d7549 | 1449 | int ret = -EINVAL; |
d342894c | 1450 | |
e4c7ed41 | 1451 | if (ip->sa.sa_family == AF_INET) { |
c6fcc4fc | 1452 | struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock); |
e4c7ed41 CW |
1453 | struct ip_mreqn mreq = { |
1454 | .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr, | |
1455 | .imr_ifindex = ifindex, | |
1456 | }; | |
1457 | ||
c6fcc4fc | 1458 | sk = sock4->sock->sk; |
b1be00a6 | 1459 | lock_sock(sk); |
56ef9c90 | 1460 | ret = ip_mc_join_group(sk, &mreq); |
b1be00a6 | 1461 | release_sock(sk); |
e4c7ed41 CW |
1462 | #if IS_ENABLED(CONFIG_IPV6) |
1463 | } else { | |
c6fcc4fc | 1464 | struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock); |
1465 | ||
1466 | sk = sock6->sock->sk; | |
b1be00a6 | 1467 | lock_sock(sk); |
56ef9c90 MRL |
1468 | ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex, |
1469 | &ip->sin6.sin6_addr); | |
b1be00a6 | 1470 | release_sock(sk); |
e4c7ed41 CW |
1471 | #endif |
1472 | } | |
3fc2de2f | 1473 | |
56ef9c90 | 1474 | return ret; |
3fc2de2f | 1475 | } |
1476 | ||
1477 | /* Inverse of vxlan_igmp_join when last VNI is brought down */ | |
56ef9c90 | 1478 | static int vxlan_igmp_leave(struct vxlan_dev *vxlan) |
3fc2de2f | 1479 | { |
b1be00a6 | 1480 | struct sock *sk; |
e4c7ed41 CW |
1481 | union vxlan_addr *ip = &vxlan->default_dst.remote_ip; |
1482 | int ifindex = vxlan->default_dst.remote_ifindex; | |
149d7549 | 1483 | int ret = -EINVAL; |
3fc2de2f | 1484 | |
e4c7ed41 | 1485 | if (ip->sa.sa_family == AF_INET) { |
c6fcc4fc | 1486 | struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock); |
e4c7ed41 CW |
1487 | struct ip_mreqn mreq = { |
1488 | .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr, | |
1489 | .imr_ifindex = ifindex, | |
1490 | }; | |
1491 | ||
c6fcc4fc | 1492 | sk = sock4->sock->sk; |
b1be00a6 | 1493 | lock_sock(sk); |
56ef9c90 | 1494 | ret = ip_mc_leave_group(sk, &mreq); |
b1be00a6 | 1495 | release_sock(sk); |
e4c7ed41 CW |
1496 | #if IS_ENABLED(CONFIG_IPV6) |
1497 | } else { | |
c6fcc4fc | 1498 | struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock); |
1499 | ||
1500 | sk = sock6->sock->sk; | |
b1be00a6 | 1501 | lock_sock(sk); |
56ef9c90 MRL |
1502 | ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex, |
1503 | &ip->sin6.sin6_addr); | |
b1be00a6 | 1504 | release_sock(sk); |
e4c7ed41 CW |
1505 | #endif |
1506 | } | |
d342894c | 1507 | |
56ef9c90 | 1508 | return ret; |
d342894c | 1509 | } |
1510 | ||
f14ecebb JB |
1511 | static bool vxlan_remcsum(struct vxlanhdr *unparsed, |
1512 | struct sk_buff *skb, u32 vxflags) | |
dfd8645e | 1513 | { |
7d34fa75 | 1514 | size_t start, offset; |
dfd8645e | 1515 | |
f14ecebb JB |
1516 | if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload) |
1517 | goto out; | |
b7fe10e5 | 1518 | |
f14ecebb JB |
1519 | start = vxlan_rco_start(unparsed->vx_vni); |
1520 | offset = start + vxlan_rco_offset(unparsed->vx_vni); | |
dfd8645e | 1521 | |
7d34fa75 | 1522 | if (!pskb_may_pull(skb, offset + sizeof(u16))) |
be5cfeab | 1523 | return false; |
dfd8645e | 1524 | |
be5cfeab JB |
1525 | skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset, |
1526 | !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL)); | |
f14ecebb JB |
1527 | out: |
1528 | unparsed->vx_flags &= ~VXLAN_HF_RCO; | |
1529 | unparsed->vx_vni &= VXLAN_VNI_MASK; | |
be5cfeab | 1530 | return true; |
dfd8645e TH |
1531 | } |
1532 | ||
f14ecebb | 1533 | static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed, |
64f87d36 | 1534 | struct sk_buff *skb, u32 vxflags, |
10a5af23 | 1535 | struct vxlan_metadata *md) |
3288af08 | 1536 | { |
f14ecebb | 1537 | struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed; |
10a5af23 | 1538 | struct metadata_dst *tun_dst; |
f14ecebb JB |
1539 | |
1540 | if (!(unparsed->vx_flags & VXLAN_HF_GBP)) | |
1541 | goto out; | |
3288af08 | 1542 | |
3288af08 JB |
1543 | md->gbp = ntohs(gbp->policy_id); |
1544 | ||
10a5af23 | 1545 | tun_dst = (struct metadata_dst *)skb_dst(skb); |
810813c4 | 1546 | if (tun_dst) { |
3288af08 | 1547 | tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT; |
810813c4 DM |
1548 | tun_dst->u.tun_info.options_len = sizeof(*md); |
1549 | } | |
3288af08 JB |
1550 | if (gbp->dont_learn) |
1551 | md->gbp |= VXLAN_GBP_DONT_LEARN; | |
1552 | ||
1553 | if (gbp->policy_applied) | |
1554 | md->gbp |= VXLAN_GBP_POLICY_APPLIED; | |
f14ecebb | 1555 | |
64f87d36 JB |
1556 | /* In flow-based mode, GBP is carried in dst_metadata */ |
1557 | if (!(vxflags & VXLAN_F_COLLECT_METADATA)) | |
1558 | skb->mark = md->gbp; | |
f14ecebb JB |
1559 | out: |
1560 | unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS; | |
3288af08 JB |
1561 | } |
1562 | ||
e1e5314d | 1563 | static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed, |
61618eea | 1564 | __be16 *protocol, |
e1e5314d JB |
1565 | struct sk_buff *skb, u32 vxflags) |
1566 | { | |
1567 | struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed; | |
1568 | ||
1569 | /* Need to have Next Protocol set for interfaces in GPE mode. */ | |
1570 | if (!gpe->np_applied) | |
1571 | return false; | |
1572 | /* "The initial version is 0. If a receiver does not support the | |
1573 | * version indicated it MUST drop the packet. | |
1574 | */ | |
1575 | if (gpe->version != 0) | |
1576 | return false; | |
1577 | /* "When the O bit is set to 1, the packet is an OAM packet and OAM | |
1578 | * processing MUST occur." However, we don't implement OAM | |
1579 | * processing, thus drop the packet. | |
1580 | */ | |
1581 | if (gpe->oam_flag) | |
1582 | return false; | |
1583 | ||
fa20e0e3 JB |
1584 | *protocol = tun_p_to_eth_p(gpe->next_protocol); |
1585 | if (!*protocol) | |
e1e5314d | 1586 | return false; |
e1e5314d JB |
1587 | |
1588 | unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS; | |
1589 | return true; | |
1590 | } | |
1591 | ||
1ab016e2 JB |
1592 | static bool vxlan_set_mac(struct vxlan_dev *vxlan, |
1593 | struct vxlan_sock *vs, | |
3ad7a4b1 | 1594 | struct sk_buff *skb, __be32 vni) |
614732ea | 1595 | { |
614732ea | 1596 | union vxlan_addr saddr; |
87613de9 | 1597 | u32 ifindex = skb->dev->ifindex; |
614732ea | 1598 | |
614732ea | 1599 | skb_reset_mac_header(skb); |
614732ea TG |
1600 | skb->protocol = eth_type_trans(skb, vxlan->dev); |
1601 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
1602 | ||
1603 | /* Ignore packet loops (and multicast echo) */ | |
1604 | if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr)) | |
1ab016e2 | 1605 | return false; |
614732ea | 1606 | |
760c6805 | 1607 | /* Get address from the outer IP header */ |
ce212d0f | 1608 | if (vxlan_get_sk_family(vs) == AF_INET) { |
1ab016e2 | 1609 | saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr; |
614732ea TG |
1610 | saddr.sa.sa_family = AF_INET; |
1611 | #if IS_ENABLED(CONFIG_IPV6) | |
1612 | } else { | |
1ab016e2 | 1613 | saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr; |
614732ea TG |
1614 | saddr.sa.sa_family = AF_INET6; |
1615 | #endif | |
1616 | } | |
1617 | ||
dc5321d7 | 1618 | if ((vxlan->cfg.flags & VXLAN_F_LEARN) && |
87613de9 | 1619 | vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni)) |
1ab016e2 JB |
1620 | return false; |
1621 | ||
1622 | return true; | |
1623 | } | |
1624 | ||
760c6805 JB |
1625 | static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph, |
1626 | struct sk_buff *skb) | |
1627 | { | |
1628 | int err = 0; | |
1629 | ||
1630 | if (vxlan_get_sk_family(vs) == AF_INET) | |
1631 | err = IP_ECN_decapsulate(oiph, skb); | |
1632 | #if IS_ENABLED(CONFIG_IPV6) | |
1633 | else | |
1634 | err = IP6_ECN_decapsulate(oiph, skb); | |
1635 | #endif | |
1636 | ||
1637 | if (unlikely(err) && log_ecn_error) { | |
1638 | if (vxlan_get_sk_family(vs) == AF_INET) | |
1639 | net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n", | |
1640 | &((struct iphdr *)oiph)->saddr, | |
1641 | ((struct iphdr *)oiph)->tos); | |
1642 | else | |
1643 | net_info_ratelimited("non-ECT from %pI6\n", | |
1644 | &((struct ipv6hdr *)oiph)->saddr); | |
1645 | } | |
1646 | return err <= 1; | |
1647 | } | |
1648 | ||
d342894c | 1649 | /* Callback from net/ipv4/udp.c to receive packets */ |
f2d1968e | 1650 | static int vxlan_rcv(struct sock *sk, struct sk_buff *skb) |
d342894c | 1651 | { |
f2d1968e | 1652 | struct pcpu_sw_netstats *stats; |
c9e78efb | 1653 | struct vxlan_dev *vxlan; |
5cfccc5a | 1654 | struct vxlan_sock *vs; |
f14ecebb | 1655 | struct vxlanhdr unparsed; |
ee122c79 TG |
1656 | struct vxlan_metadata _md; |
1657 | struct vxlan_metadata *md = &_md; | |
61618eea | 1658 | __be16 protocol = htons(ETH_P_TEB); |
e1e5314d | 1659 | bool raw_proto = false; |
f2d1968e | 1660 | void *oiph; |
3ad7a4b1 | 1661 | __be32 vni = 0; |
d342894c | 1662 | |
e1e5314d | 1663 | /* Need UDP and VXLAN header to be present */ |
7ce04758 | 1664 | if (!pskb_may_pull(skb, VXLAN_HLEN)) |
e5aed006 | 1665 | goto drop; |
d342894c | 1666 | |
f14ecebb | 1667 | unparsed = *vxlan_hdr(skb); |
288b01c8 JB |
1668 | /* VNI flag always required to be set */ |
1669 | if (!(unparsed.vx_flags & VXLAN_HF_VNI)) { | |
1670 | netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n", | |
1671 | ntohl(vxlan_hdr(skb)->vx_flags), | |
1672 | ntohl(vxlan_hdr(skb)->vx_vni)); | |
1673 | /* Return non vxlan pkt */ | |
e5aed006 | 1674 | goto drop; |
d342894c | 1675 | } |
288b01c8 JB |
1676 | unparsed.vx_flags &= ~VXLAN_HF_VNI; |
1677 | unparsed.vx_vni &= ~VXLAN_VNI_MASK; | |
d342894c | 1678 | |
559835ea | 1679 | vs = rcu_dereference_sk_user_data(sk); |
5cfccc5a | 1680 | if (!vs) |
d342894c | 1681 | goto drop; |
d342894c | 1682 | |
3ad7a4b1 RP |
1683 | vni = vxlan_vni(vxlan_hdr(skb)->vx_vni); |
1684 | ||
49f810f0 | 1685 | vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni); |
c9e78efb JB |
1686 | if (!vxlan) |
1687 | goto drop; | |
1688 | ||
e1e5314d JB |
1689 | /* For backwards compatibility, only allow reserved fields to be |
1690 | * used by VXLAN extensions if explicitly requested. | |
1691 | */ | |
1692 | if (vs->flags & VXLAN_F_GPE) { | |
1693 | if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags)) | |
1694 | goto drop; | |
1695 | raw_proto = true; | |
1696 | } | |
1697 | ||
1698 | if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto, | |
1699 | !net_eq(vxlan->net, dev_net(vxlan->dev)))) | |
98c81476 | 1700 | goto drop; |
c9e78efb | 1701 | |
ee122c79 | 1702 | if (vxlan_collect_metadata(vs)) { |
10a5af23 | 1703 | struct metadata_dst *tun_dst; |
07dabf20 | 1704 | |
c29a70d2 | 1705 | tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY, |
d817f432 | 1706 | key32_to_tunnel_id(vni), sizeof(*md)); |
c29a70d2 | 1707 | |
ee122c79 TG |
1708 | if (!tun_dst) |
1709 | goto drop; | |
1710 | ||
0f1b7354 | 1711 | md = ip_tunnel_info_opts(&tun_dst->u.tun_info); |
10a5af23 JB |
1712 | |
1713 | skb_dst_set(skb, (struct dst_entry *)tun_dst); | |
ee122c79 TG |
1714 | } else { |
1715 | memset(md, 0, sizeof(*md)); | |
1716 | } | |
1717 | ||
f14ecebb JB |
1718 | if (vs->flags & VXLAN_F_REMCSUM_RX) |
1719 | if (!vxlan_remcsum(&unparsed, skb, vs->flags)) | |
1720 | goto drop; | |
1721 | if (vs->flags & VXLAN_F_GBP) | |
10a5af23 | 1722 | vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md); |
e1e5314d JB |
1723 | /* Note that GBP and GPE can never be active together. This is |
1724 | * ensured in vxlan_dev_configure. | |
1725 | */ | |
3511494c | 1726 | |
f14ecebb | 1727 | if (unparsed.vx_flags || unparsed.vx_vni) { |
3bf39475 TH |
1728 | /* If there are any unprocessed flags remaining treat |
1729 | * this as a malformed packet. This behavior diverges from | |
1730 | * VXLAN RFC (RFC7348) which stipulates that bits in reserved | |
1731 | * in reserved fields are to be ignored. The approach here | |
c4b49512 | 1732 | * maintains compatibility with previous stack code, and also |
3bf39475 TH |
1733 | * is more robust and provides a little more security in |
1734 | * adding extensions to VXLAN. | |
1735 | */ | |
288b01c8 | 1736 | goto drop; |
3bf39475 TH |
1737 | } |
1738 | ||
e1e5314d | 1739 | if (!raw_proto) { |
3ad7a4b1 | 1740 | if (!vxlan_set_mac(vxlan, vs, skb, vni)) |
e1e5314d JB |
1741 | goto drop; |
1742 | } else { | |
8be0cfa4 | 1743 | skb_reset_mac_header(skb); |
e1e5314d JB |
1744 | skb->dev = vxlan->dev; |
1745 | skb->pkt_type = PACKET_HOST; | |
1746 | } | |
f2d1968e | 1747 | |
f2d1968e JB |
1748 | oiph = skb_network_header(skb); |
1749 | skb_reset_network_header(skb); | |
1750 | ||
1751 | if (!vxlan_ecn_decapsulate(vs, oiph, skb)) { | |
1752 | ++vxlan->dev->stats.rx_frame_errors; | |
1753 | ++vxlan->dev->stats.rx_errors; | |
1754 | goto drop; | |
1755 | } | |
1756 | ||
59cbf56f ED |
1757 | rcu_read_lock(); |
1758 | ||
1759 | if (unlikely(!(vxlan->dev->flags & IFF_UP))) { | |
1760 | rcu_read_unlock(); | |
1761 | atomic_long_inc(&vxlan->dev->rx_dropped); | |
1762 | goto drop; | |
1763 | } | |
1764 | ||
f2d1968e JB |
1765 | stats = this_cpu_ptr(vxlan->dev->tstats); |
1766 | u64_stats_update_begin(&stats->syncp); | |
1767 | stats->rx_packets++; | |
1768 | stats->rx_bytes += skb->len; | |
1769 | u64_stats_update_end(&stats->syncp); | |
1770 | ||
1771 | gro_cells_receive(&vxlan->gro_cells, skb); | |
59cbf56f ED |
1772 | |
1773 | rcu_read_unlock(); | |
1774 | ||
5cfccc5a PS |
1775 | return 0; |
1776 | ||
1777 | drop: | |
288b01c8 JB |
1778 | /* Consume bad packet */ |
1779 | kfree_skb(skb); | |
1780 | return 0; | |
5cfccc5a PS |
1781 | } |
1782 | ||
c3a43b9f SB |
1783 | /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */ |
1784 | static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb) | |
1785 | { | |
1786 | struct vxlan_dev *vxlan; | |
1787 | struct vxlan_sock *vs; | |
1788 | struct vxlanhdr *hdr; | |
1789 | __be32 vni; | |
1790 | ||
8399a693 | 1791 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN)) |
c3a43b9f SB |
1792 | return -EINVAL; |
1793 | ||
1794 | hdr = vxlan_hdr(skb); | |
1795 | ||
1796 | if (!(hdr->vx_flags & VXLAN_HF_VNI)) | |
1797 | return -EINVAL; | |
1798 | ||
1799 | vs = rcu_dereference_sk_user_data(sk); | |
1800 | if (!vs) | |
1801 | return -ENOENT; | |
1802 | ||
1803 | vni = vxlan_vni(hdr->vx_vni); | |
1804 | vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni); | |
1805 | if (!vxlan) | |
1806 | return -ENOENT; | |
1807 | ||
1808 | return 0; | |
1809 | } | |
1810 | ||
3ad7a4b1 | 1811 | static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni) |
e4f67add DS |
1812 | { |
1813 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
1814 | struct arphdr *parp; | |
1815 | u8 *arpptr, *sha; | |
1816 | __be32 sip, tip; | |
1817 | struct neighbour *n; | |
1818 | ||
1819 | if (dev->flags & IFF_NOARP) | |
1820 | goto out; | |
1821 | ||
1822 | if (!pskb_may_pull(skb, arp_hdr_len(dev))) { | |
1823 | dev->stats.tx_dropped++; | |
1824 | goto out; | |
1825 | } | |
1826 | parp = arp_hdr(skb); | |
1827 | ||
1828 | if ((parp->ar_hrd != htons(ARPHRD_ETHER) && | |
1829 | parp->ar_hrd != htons(ARPHRD_IEEE802)) || | |
1830 | parp->ar_pro != htons(ETH_P_IP) || | |
1831 | parp->ar_op != htons(ARPOP_REQUEST) || | |
1832 | parp->ar_hln != dev->addr_len || | |
1833 | parp->ar_pln != 4) | |
1834 | goto out; | |
1835 | arpptr = (u8 *)parp + sizeof(struct arphdr); | |
1836 | sha = arpptr; | |
1837 | arpptr += dev->addr_len; /* sha */ | |
1838 | memcpy(&sip, arpptr, sizeof(sip)); | |
1839 | arpptr += sizeof(sip); | |
1840 | arpptr += dev->addr_len; /* tha */ | |
1841 | memcpy(&tip, arpptr, sizeof(tip)); | |
1842 | ||
1843 | if (ipv4_is_loopback(tip) || | |
1844 | ipv4_is_multicast(tip)) | |
1845 | goto out; | |
1846 | ||
1847 | n = neigh_lookup(&arp_tbl, &tip, dev); | |
1848 | ||
1849 | if (n) { | |
e4f67add DS |
1850 | struct vxlan_fdb *f; |
1851 | struct sk_buff *reply; | |
1852 | ||
1853 | if (!(n->nud_state & NUD_CONNECTED)) { | |
1854 | neigh_release(n); | |
1855 | goto out; | |
1856 | } | |
1857 | ||
3ad7a4b1 | 1858 | f = vxlan_find_mac(vxlan, n->ha, vni); |
e4c7ed41 | 1859 | if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) { |
e4f67add DS |
1860 | /* bridge-local neighbor */ |
1861 | neigh_release(n); | |
1862 | goto out; | |
1863 | } | |
1864 | ||
1865 | reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha, | |
1866 | n->ha, sha); | |
1867 | ||
1868 | neigh_release(n); | |
1869 | ||
7346135d DS |
1870 | if (reply == NULL) |
1871 | goto out; | |
1872 | ||
e4f67add DS |
1873 | skb_reset_mac_header(reply); |
1874 | __skb_pull(reply, skb_network_offset(reply)); | |
1875 | reply->ip_summed = CHECKSUM_UNNECESSARY; | |
1876 | reply->pkt_type = PACKET_HOST; | |
1877 | ||
1878 | if (netif_rx_ni(reply) == NET_RX_DROP) | |
1879 | dev->stats.rx_dropped++; | |
dc5321d7 | 1880 | } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) { |
e4c7ed41 CW |
1881 | union vxlan_addr ipa = { |
1882 | .sin.sin_addr.s_addr = tip, | |
a45e92a5 | 1883 | .sin.sin_family = AF_INET, |
e4c7ed41 CW |
1884 | }; |
1885 | ||
1886 | vxlan_ip_miss(dev, &ipa); | |
1887 | } | |
e4f67add DS |
1888 | out: |
1889 | consume_skb(skb); | |
1890 | return NETDEV_TX_OK; | |
1891 | } | |
1892 | ||
f564f45c | 1893 | #if IS_ENABLED(CONFIG_IPV6) |
4b29dba9 DS |
1894 | static struct sk_buff *vxlan_na_create(struct sk_buff *request, |
1895 | struct neighbour *n, bool isrouter) | |
1896 | { | |
1897 | struct net_device *dev = request->dev; | |
1898 | struct sk_buff *reply; | |
1899 | struct nd_msg *ns, *na; | |
1900 | struct ipv6hdr *pip6; | |
1901 | u8 *daddr; | |
1902 | int na_olen = 8; /* opt hdr + ETH_ALEN for target */ | |
1903 | int ns_olen; | |
1904 | int i, len; | |
1905 | ||
f1fb08f6 | 1906 | if (dev == NULL || !pskb_may_pull(request, request->len)) |
4b29dba9 DS |
1907 | return NULL; |
1908 | ||
1909 | len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) + | |
1910 | sizeof(*na) + na_olen + dev->needed_tailroom; | |
1911 | reply = alloc_skb(len, GFP_ATOMIC); | |
1912 | if (reply == NULL) | |
1913 | return NULL; | |
1914 | ||
1915 | reply->protocol = htons(ETH_P_IPV6); | |
1916 | reply->dev = dev; | |
1917 | skb_reserve(reply, LL_RESERVED_SPACE(request->dev)); | |
1918 | skb_push(reply, sizeof(struct ethhdr)); | |
6297b91c | 1919 | skb_reset_mac_header(reply); |
4b29dba9 | 1920 | |
f1fb08f6 | 1921 | ns = (struct nd_msg *)(ipv6_hdr(request) + 1); |
4b29dba9 DS |
1922 | |
1923 | daddr = eth_hdr(request)->h_source; | |
f1fb08f6 VB |
1924 | ns_olen = request->len - skb_network_offset(request) - |
1925 | sizeof(struct ipv6hdr) - sizeof(*ns); | |
4b29dba9 DS |
1926 | for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) { |
1927 | if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) { | |
1928 | daddr = ns->opt + i + sizeof(struct nd_opt_hdr); | |
1929 | break; | |
1930 | } | |
1931 | } | |
1932 | ||
1933 | /* Ethernet header */ | |
1934 | ether_addr_copy(eth_hdr(reply)->h_dest, daddr); | |
1935 | ether_addr_copy(eth_hdr(reply)->h_source, n->ha); | |
1936 | eth_hdr(reply)->h_proto = htons(ETH_P_IPV6); | |
1937 | reply->protocol = htons(ETH_P_IPV6); | |
1938 | ||
1939 | skb_pull(reply, sizeof(struct ethhdr)); | |
6297b91c | 1940 | skb_reset_network_header(reply); |
4b29dba9 DS |
1941 | skb_put(reply, sizeof(struct ipv6hdr)); |
1942 | ||
1943 | /* IPv6 header */ | |
1944 | ||
1945 | pip6 = ipv6_hdr(reply); | |
1946 | memset(pip6, 0, sizeof(struct ipv6hdr)); | |
1947 | pip6->version = 6; | |
1948 | pip6->priority = ipv6_hdr(request)->priority; | |
1949 | pip6->nexthdr = IPPROTO_ICMPV6; | |
1950 | pip6->hop_limit = 255; | |
1951 | pip6->daddr = ipv6_hdr(request)->saddr; | |
1952 | pip6->saddr = *(struct in6_addr *)n->primary_key; | |
1953 | ||
1954 | skb_pull(reply, sizeof(struct ipv6hdr)); | |
6297b91c | 1955 | skb_reset_transport_header(reply); |
4b29dba9 | 1956 | |
4b29dba9 | 1957 | /* Neighbor Advertisement */ |
b080db58 | 1958 | na = skb_put_zero(reply, sizeof(*na) + na_olen); |
4b29dba9 DS |
1959 | na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT; |
1960 | na->icmph.icmp6_router = isrouter; | |
1961 | na->icmph.icmp6_override = 1; | |
1962 | na->icmph.icmp6_solicited = 1; | |
1963 | na->target = ns->target; | |
1964 | ether_addr_copy(&na->opt[2], n->ha); | |
1965 | na->opt[0] = ND_OPT_TARGET_LL_ADDR; | |
1966 | na->opt[1] = na_olen >> 3; | |
1967 | ||
1968 | na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr, | |
1969 | &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6, | |
1970 | csum_partial(na, sizeof(*na)+na_olen, 0)); | |
1971 | ||
1972 | pip6->payload_len = htons(sizeof(*na)+na_olen); | |
1973 | ||
1974 | skb_push(reply, sizeof(struct ipv6hdr)); | |
1975 | ||
1976 | reply->ip_summed = CHECKSUM_UNNECESSARY; | |
1977 | ||
1978 | return reply; | |
1979 | } | |
1980 | ||
3ad7a4b1 | 1981 | static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni) |
f564f45c CW |
1982 | { |
1983 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
8dcd81a9 | 1984 | const struct in6_addr *daddr; |
8bff3685 | 1985 | const struct ipv6hdr *iphdr; |
4b29dba9 | 1986 | struct inet6_dev *in6_dev; |
8bff3685 XL |
1987 | struct neighbour *n; |
1988 | struct nd_msg *msg; | |
f564f45c CW |
1989 | |
1990 | in6_dev = __in6_dev_get(dev); | |
1991 | if (!in6_dev) | |
1992 | goto out; | |
1993 | ||
f564f45c | 1994 | iphdr = ipv6_hdr(skb); |
f564f45c | 1995 | daddr = &iphdr->daddr; |
f1fb08f6 | 1996 | msg = (struct nd_msg *)(iphdr + 1); |
f564f45c | 1997 | |
4b29dba9 DS |
1998 | if (ipv6_addr_loopback(daddr) || |
1999 | ipv6_addr_is_multicast(&msg->target)) | |
2000 | goto out; | |
2001 | ||
2002 | n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev); | |
f564f45c CW |
2003 | |
2004 | if (n) { | |
2005 | struct vxlan_fdb *f; | |
4b29dba9 | 2006 | struct sk_buff *reply; |
f564f45c CW |
2007 | |
2008 | if (!(n->nud_state & NUD_CONNECTED)) { | |
2009 | neigh_release(n); | |
2010 | goto out; | |
2011 | } | |
2012 | ||
3ad7a4b1 | 2013 | f = vxlan_find_mac(vxlan, n->ha, vni); |
f564f45c CW |
2014 | if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) { |
2015 | /* bridge-local neighbor */ | |
2016 | neigh_release(n); | |
2017 | goto out; | |
2018 | } | |
2019 | ||
4b29dba9 DS |
2020 | reply = vxlan_na_create(skb, n, |
2021 | !!(f ? f->flags & NTF_ROUTER : 0)); | |
2022 | ||
f564f45c | 2023 | neigh_release(n); |
4b29dba9 DS |
2024 | |
2025 | if (reply == NULL) | |
2026 | goto out; | |
2027 | ||
2028 | if (netif_rx_ni(reply) == NET_RX_DROP) | |
2029 | dev->stats.rx_dropped++; | |
2030 | ||
dc5321d7 | 2031 | } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) { |
4b29dba9 DS |
2032 | union vxlan_addr ipa = { |
2033 | .sin6.sin6_addr = msg->target, | |
a45e92a5 | 2034 | .sin6.sin6_family = AF_INET6, |
4b29dba9 DS |
2035 | }; |
2036 | ||
f564f45c CW |
2037 | vxlan_ip_miss(dev, &ipa); |
2038 | } | |
2039 | ||
2040 | out: | |
2041 | consume_skb(skb); | |
2042 | return NETDEV_TX_OK; | |
2043 | } | |
2044 | #endif | |
2045 | ||
e4f67add DS |
2046 | static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb) |
2047 | { | |
2048 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
2049 | struct neighbour *n; | |
e4f67add DS |
2050 | |
2051 | if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) | |
2052 | return false; | |
2053 | ||
2054 | n = NULL; | |
2055 | switch (ntohs(eth_hdr(skb)->h_proto)) { | |
2056 | case ETH_P_IP: | |
e15a00aa CW |
2057 | { |
2058 | struct iphdr *pip; | |
2059 | ||
e4f67add DS |
2060 | if (!pskb_may_pull(skb, sizeof(struct iphdr))) |
2061 | return false; | |
2062 | pip = ip_hdr(skb); | |
2063 | n = neigh_lookup(&arp_tbl, &pip->daddr, dev); | |
dc5321d7 | 2064 | if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) { |
e4c7ed41 CW |
2065 | union vxlan_addr ipa = { |
2066 | .sin.sin_addr.s_addr = pip->daddr, | |
a45e92a5 | 2067 | .sin.sin_family = AF_INET, |
e4c7ed41 CW |
2068 | }; |
2069 | ||
2070 | vxlan_ip_miss(dev, &ipa); | |
2071 | return false; | |
2072 | } | |
2073 | ||
e4f67add | 2074 | break; |
e15a00aa CW |
2075 | } |
2076 | #if IS_ENABLED(CONFIG_IPV6) | |
2077 | case ETH_P_IPV6: | |
2078 | { | |
2079 | struct ipv6hdr *pip6; | |
2080 | ||
2081 | if (!pskb_may_pull(skb, sizeof(struct ipv6hdr))) | |
2082 | return false; | |
2083 | pip6 = ipv6_hdr(skb); | |
2084 | n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev); | |
dc5321d7 | 2085 | if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) { |
e15a00aa CW |
2086 | union vxlan_addr ipa = { |
2087 | .sin6.sin6_addr = pip6->daddr, | |
a45e92a5 | 2088 | .sin6.sin6_family = AF_INET6, |
e15a00aa CW |
2089 | }; |
2090 | ||
2091 | vxlan_ip_miss(dev, &ipa); | |
2092 | return false; | |
2093 | } | |
2094 | ||
2095 | break; | |
2096 | } | |
2097 | #endif | |
e4f67add DS |
2098 | default: |
2099 | return false; | |
2100 | } | |
2101 | ||
2102 | if (n) { | |
2103 | bool diff; | |
2104 | ||
7367d0b5 | 2105 | diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha); |
e4f67add DS |
2106 | if (diff) { |
2107 | memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest, | |
2108 | dev->addr_len); | |
2109 | memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len); | |
2110 | } | |
2111 | neigh_release(n); | |
2112 | return diff; | |
e4c7ed41 CW |
2113 | } |
2114 | ||
e4f67add DS |
2115 | return false; |
2116 | } | |
2117 | ||
af33c1ad | 2118 | static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags, |
3511494c TG |
2119 | struct vxlan_metadata *md) |
2120 | { | |
2121 | struct vxlanhdr_gbp *gbp; | |
2122 | ||
db79a621 TG |
2123 | if (!md->gbp) |
2124 | return; | |
2125 | ||
3511494c | 2126 | gbp = (struct vxlanhdr_gbp *)vxh; |
54bfd872 | 2127 | vxh->vx_flags |= VXLAN_HF_GBP; |
3511494c TG |
2128 | |
2129 | if (md->gbp & VXLAN_GBP_DONT_LEARN) | |
2130 | gbp->dont_learn = 1; | |
2131 | ||
2132 | if (md->gbp & VXLAN_GBP_POLICY_APPLIED) | |
2133 | gbp->policy_applied = 1; | |
2134 | ||
2135 | gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK); | |
2136 | } | |
2137 | ||
e1e5314d JB |
2138 | static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags, |
2139 | __be16 protocol) | |
2140 | { | |
2141 | struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh; | |
2142 | ||
2143 | gpe->np_applied = 1; | |
fa20e0e3 JB |
2144 | gpe->next_protocol = tun_p_from_eth_p(protocol); |
2145 | if (!gpe->next_protocol) | |
2146 | return -EPFNOSUPPORT; | |
2147 | return 0; | |
e1e5314d JB |
2148 | } |
2149 | ||
f491e56d JB |
2150 | static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst, |
2151 | int iphdr_len, __be32 vni, | |
2152 | struct vxlan_metadata *md, u32 vxflags, | |
b4ed5cad | 2153 | bool udp_sum) |
e4c7ed41 | 2154 | { |
e4c7ed41 | 2155 | struct vxlanhdr *vxh; |
e4c7ed41 CW |
2156 | int min_headroom; |
2157 | int err; | |
dfd8645e | 2158 | int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL; |
e1e5314d | 2159 | __be16 inner_protocol = htons(ETH_P_TEB); |
dfd8645e | 2160 | |
af33c1ad | 2161 | if ((vxflags & VXLAN_F_REMCSUM_TX) && |
dfd8645e TH |
2162 | skb->ip_summed == CHECKSUM_PARTIAL) { |
2163 | int csum_start = skb_checksum_start_offset(skb); | |
2164 | ||
2165 | if (csum_start <= VXLAN_MAX_REMCSUM_START && | |
2166 | !(csum_start & VXLAN_RCO_SHIFT_MASK) && | |
2167 | (skb->csum_offset == offsetof(struct udphdr, check) || | |
b5708501 | 2168 | skb->csum_offset == offsetof(struct tcphdr, check))) |
dfd8645e | 2169 | type |= SKB_GSO_TUNNEL_REMCSUM; |
dfd8645e | 2170 | } |
e4c7ed41 | 2171 | |
e4c7ed41 | 2172 | min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len |
4a4f86cc | 2173 | + VXLAN_HLEN + iphdr_len; |
649c5b8b PS |
2174 | |
2175 | /* Need space for new headers (invalidates iph ptr) */ | |
2176 | err = skb_cow_head(skb, min_headroom); | |
e1e5314d | 2177 | if (unlikely(err)) |
c46b7897 | 2178 | return err; |
649c5b8b | 2179 | |
aed069df AD |
2180 | err = iptunnel_handle_offloads(skb, type); |
2181 | if (err) | |
c46b7897 | 2182 | return err; |
b736a623 | 2183 | |
d58ff351 | 2184 | vxh = __skb_push(skb, sizeof(*vxh)); |
54bfd872 JB |
2185 | vxh->vx_flags = VXLAN_HF_VNI; |
2186 | vxh->vx_vni = vxlan_vni_field(vni); | |
49560532 | 2187 | |
dfd8645e | 2188 | if (type & SKB_GSO_TUNNEL_REMCSUM) { |
54bfd872 | 2189 | unsigned int start; |
dfd8645e | 2190 | |
54bfd872 JB |
2191 | start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr); |
2192 | vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset); | |
2193 | vxh->vx_flags |= VXLAN_HF_RCO; | |
dfd8645e TH |
2194 | |
2195 | if (!skb_is_gso(skb)) { | |
2196 | skb->ip_summed = CHECKSUM_NONE; | |
2197 | skb->encapsulation = 0; | |
2198 | } | |
2199 | } | |
2200 | ||
af33c1ad TH |
2201 | if (vxflags & VXLAN_F_GBP) |
2202 | vxlan_build_gbp_hdr(vxh, vxflags, md); | |
e1e5314d JB |
2203 | if (vxflags & VXLAN_F_GPE) { |
2204 | err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol); | |
2205 | if (err < 0) | |
c46b7897 | 2206 | return err; |
e1e5314d JB |
2207 | inner_protocol = skb->protocol; |
2208 | } | |
3511494c | 2209 | |
e1e5314d | 2210 | skb_set_inner_protocol(skb, inner_protocol); |
039f5062 | 2211 | return 0; |
49560532 | 2212 | } |
49560532 | 2213 | |
655c3de1 | 2214 | static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev, |
2215 | struct vxlan_sock *sock4, | |
1a8496ba | 2216 | struct sk_buff *skb, int oif, u8 tos, |
4ecb1d83 | 2217 | __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport, |
0c1d70af | 2218 | struct dst_cache *dst_cache, |
db3c6139 | 2219 | const struct ip_tunnel_info *info) |
1a8496ba | 2220 | { |
db3c6139 | 2221 | bool use_cache = ip_tunnel_dst_cache_usable(skb, info); |
1a8496ba JB |
2222 | struct rtable *rt = NULL; |
2223 | struct flowi4 fl4; | |
2224 | ||
655c3de1 | 2225 | if (!sock4) |
2226 | return ERR_PTR(-EIO); | |
2227 | ||
db3c6139 DB |
2228 | if (tos && !info) |
2229 | use_cache = false; | |
2230 | if (use_cache) { | |
0c1d70af PA |
2231 | rt = dst_cache_get_ip4(dst_cache, saddr); |
2232 | if (rt) | |
2233 | return rt; | |
2234 | } | |
2235 | ||
1a8496ba JB |
2236 | memset(&fl4, 0, sizeof(fl4)); |
2237 | fl4.flowi4_oif = oif; | |
2238 | fl4.flowi4_tos = RT_TOS(tos); | |
2239 | fl4.flowi4_mark = skb->mark; | |
2240 | fl4.flowi4_proto = IPPROTO_UDP; | |
2241 | fl4.daddr = daddr; | |
272d96a5 | 2242 | fl4.saddr = *saddr; |
4ecb1d83 MP |
2243 | fl4.fl4_dport = dport; |
2244 | fl4.fl4_sport = sport; | |
1a8496ba JB |
2245 | |
2246 | rt = ip_route_output_key(vxlan->net, &fl4); | |
478db1f1 | 2247 | if (!IS_ERR(rt)) { |
655c3de1 | 2248 | if (rt->dst.dev == dev) { |
2249 | netdev_dbg(dev, "circular route to %pI4\n", &daddr); | |
2250 | ip_rt_put(rt); | |
2251 | return ERR_PTR(-ELOOP); | |
2252 | } | |
2253 | ||
1a8496ba | 2254 | *saddr = fl4.saddr; |
0c1d70af PA |
2255 | if (use_cache) |
2256 | dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr); | |
655c3de1 | 2257 | } else { |
2258 | netdev_dbg(dev, "no route to %pI4\n", &daddr); | |
2259 | return ERR_PTR(-ENETUNREACH); | |
0c1d70af | 2260 | } |
1a8496ba JB |
2261 | return rt; |
2262 | } | |
2263 | ||
e5d4b29f JB |
2264 | #if IS_ENABLED(CONFIG_IPV6) |
2265 | static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan, | |
655c3de1 | 2266 | struct net_device *dev, |
03dc52a8 | 2267 | struct vxlan_sock *sock6, |
1400615d | 2268 | struct sk_buff *skb, int oif, u8 tos, |
e7f70af1 | 2269 | __be32 label, |
e5d4b29f | 2270 | const struct in6_addr *daddr, |
0c1d70af | 2271 | struct in6_addr *saddr, |
4ecb1d83 | 2272 | __be16 dport, __be16 sport, |
db3c6139 DB |
2273 | struct dst_cache *dst_cache, |
2274 | const struct ip_tunnel_info *info) | |
e5d4b29f | 2275 | { |
db3c6139 | 2276 | bool use_cache = ip_tunnel_dst_cache_usable(skb, info); |
e5d4b29f JB |
2277 | struct dst_entry *ndst; |
2278 | struct flowi6 fl6; | |
e5d4b29f | 2279 | |
c6fcc4fc | 2280 | if (!sock6) |
2281 | return ERR_PTR(-EIO); | |
2282 | ||
1400615d DB |
2283 | if (tos && !info) |
2284 | use_cache = false; | |
db3c6139 | 2285 | if (use_cache) { |
0c1d70af PA |
2286 | ndst = dst_cache_get_ip6(dst_cache, saddr); |
2287 | if (ndst) | |
2288 | return ndst; | |
2289 | } | |
2290 | ||
e5d4b29f JB |
2291 | memset(&fl6, 0, sizeof(fl6)); |
2292 | fl6.flowi6_oif = oif; | |
2293 | fl6.daddr = *daddr; | |
272d96a5 | 2294 | fl6.saddr = *saddr; |
eaa93bf4 | 2295 | fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label); |
e5d4b29f JB |
2296 | fl6.flowi6_mark = skb->mark; |
2297 | fl6.flowi6_proto = IPPROTO_UDP; | |
4ecb1d83 MP |
2298 | fl6.fl6_dport = dport; |
2299 | fl6.fl6_sport = sport; | |
e5d4b29f | 2300 | |
6c8991f4 SD |
2301 | ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk, |
2302 | &fl6, NULL); | |
2303 | if (unlikely(IS_ERR(ndst))) { | |
655c3de1 | 2304 | netdev_dbg(dev, "no route to %pI6\n", daddr); |
2305 | return ERR_PTR(-ENETUNREACH); | |
2306 | } | |
2307 | ||
2308 | if (unlikely(ndst->dev == dev)) { | |
2309 | netdev_dbg(dev, "circular route to %pI6\n", daddr); | |
2310 | dst_release(ndst); | |
2311 | return ERR_PTR(-ELOOP); | |
2312 | } | |
e5d4b29f JB |
2313 | |
2314 | *saddr = fl6.saddr; | |
db3c6139 | 2315 | if (use_cache) |
0c1d70af | 2316 | dst_cache_set_ip6(dst_cache, ndst, saddr); |
e5d4b29f JB |
2317 | return ndst; |
2318 | } | |
2319 | #endif | |
2320 | ||
9dcc71e1 SS |
2321 | /* Bypass encapsulation if the destination is local */ |
2322 | static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan, | |
3ad7a4b1 | 2323 | struct vxlan_dev *dst_vxlan, __be32 vni) |
9dcc71e1 | 2324 | { |
8f84985f | 2325 | struct pcpu_sw_netstats *tx_stats, *rx_stats; |
e4c7ed41 CW |
2326 | union vxlan_addr loopback; |
2327 | union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip; | |
4179cb5a | 2328 | struct net_device *dev; |
ce6502a8 | 2329 | int len = skb->len; |
9dcc71e1 | 2330 | |
8f84985f LR |
2331 | tx_stats = this_cpu_ptr(src_vxlan->dev->tstats); |
2332 | rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats); | |
9dcc71e1 SS |
2333 | skb->pkt_type = PACKET_HOST; |
2334 | skb->encapsulation = 0; | |
2335 | skb->dev = dst_vxlan->dev; | |
2336 | __skb_pull(skb, skb_network_offset(skb)); | |
2337 | ||
e4c7ed41 CW |
2338 | if (remote_ip->sa.sa_family == AF_INET) { |
2339 | loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK); | |
2340 | loopback.sa.sa_family = AF_INET; | |
2341 | #if IS_ENABLED(CONFIG_IPV6) | |
2342 | } else { | |
2343 | loopback.sin6.sin6_addr = in6addr_loopback; | |
2344 | loopback.sa.sa_family = AF_INET6; | |
2345 | #endif | |
2346 | } | |
2347 | ||
4179cb5a ED |
2348 | rcu_read_lock(); |
2349 | dev = skb->dev; | |
2350 | if (unlikely(!(dev->flags & IFF_UP))) { | |
2351 | kfree_skb(skb); | |
2352 | goto drop; | |
2353 | } | |
2354 | ||
dc5321d7 | 2355 | if (dst_vxlan->cfg.flags & VXLAN_F_LEARN) |
4179cb5a | 2356 | vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni); |
9dcc71e1 SS |
2357 | |
2358 | u64_stats_update_begin(&tx_stats->syncp); | |
2359 | tx_stats->tx_packets++; | |
ce6502a8 | 2360 | tx_stats->tx_bytes += len; |
9dcc71e1 SS |
2361 | u64_stats_update_end(&tx_stats->syncp); |
2362 | ||
2363 | if (netif_rx(skb) == NET_RX_SUCCESS) { | |
2364 | u64_stats_update_begin(&rx_stats->syncp); | |
2365 | rx_stats->rx_packets++; | |
ce6502a8 | 2366 | rx_stats->rx_bytes += len; |
9dcc71e1 SS |
2367 | u64_stats_update_end(&rx_stats->syncp); |
2368 | } else { | |
4179cb5a | 2369 | drop: |
ce6502a8 | 2370 | dev->stats.rx_dropped++; |
9dcc71e1 | 2371 | } |
4179cb5a | 2372 | rcu_read_unlock(); |
9dcc71e1 SS |
2373 | } |
2374 | ||
fee1fad7 | 2375 | static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev, |
49f810f0 MS |
2376 | struct vxlan_dev *vxlan, |
2377 | union vxlan_addr *daddr, | |
2378 | __be16 dst_port, int dst_ifindex, __be32 vni, | |
2379 | struct dst_entry *dst, | |
fee1fad7 | 2380 | u32 rt_flags) |
2381 | { | |
2382 | #if IS_ENABLED(CONFIG_IPV6) | |
2383 | /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of | |
2384 | * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple | |
2385 | * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry. | |
2386 | */ | |
2387 | BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL); | |
2388 | #endif | |
2389 | /* Bypass encapsulation if the destination is local */ | |
2390 | if (rt_flags & RTCF_LOCAL && | |
2391 | !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) { | |
2392 | struct vxlan_dev *dst_vxlan; | |
2393 | ||
2394 | dst_release(dst); | |
49f810f0 | 2395 | dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni, |
fee1fad7 | 2396 | daddr->sa.sa_family, dst_port, |
dc5321d7 | 2397 | vxlan->cfg.flags); |
fee1fad7 | 2398 | if (!dst_vxlan) { |
2399 | dev->stats.tx_errors++; | |
2400 | kfree_skb(skb); | |
2401 | ||
2402 | return -ENOENT; | |
2403 | } | |
3ad7a4b1 | 2404 | vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni); |
fee1fad7 | 2405 | return 1; |
2406 | } | |
2407 | ||
2408 | return 0; | |
2409 | } | |
2410 | ||
4ad16930 | 2411 | static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev, |
3ad7a4b1 RP |
2412 | __be32 default_vni, struct vxlan_rdst *rdst, |
2413 | bool did_rsc) | |
d342894c | 2414 | { |
d71785ff | 2415 | struct dst_cache *dst_cache; |
3093fbe7 | 2416 | struct ip_tunnel_info *info; |
d342894c | 2417 | struct vxlan_dev *vxlan = netdev_priv(dev); |
0770b53b | 2418 | const struct iphdr *old_iph = ip_hdr(skb); |
e4c7ed41 | 2419 | union vxlan_addr *dst; |
272d96a5 | 2420 | union vxlan_addr remote_ip, local_ip; |
ee122c79 TG |
2421 | struct vxlan_metadata _md; |
2422 | struct vxlan_metadata *md = &_md; | |
e4c7ed41 | 2423 | __be16 src_port = 0, dst_port; |
655c3de1 | 2424 | struct dst_entry *ndst = NULL; |
e7f70af1 | 2425 | __be32 vni, label; |
d342894c | 2426 | __u8 tos, ttl; |
49f810f0 | 2427 | int ifindex; |
0e6fbc5b | 2428 | int err; |
dc5321d7 | 2429 | u32 flags = vxlan->cfg.flags; |
b4ed5cad | 2430 | bool udp_sum = false; |
f491e56d | 2431 | bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev)); |
d342894c | 2432 | |
61adedf3 | 2433 | info = skb_tunnel_info(skb); |
3093fbe7 | 2434 | |
ee122c79 | 2435 | if (rdst) { |
0770b53b | 2436 | dst = &rdst->remote_ip; |
2437 | if (vxlan_addr_any(dst)) { | |
2438 | if (did_rsc) { | |
2439 | /* short-circuited back to local bridge */ | |
3ad7a4b1 | 2440 | vxlan_encap_bypass(skb, vxlan, vxlan, default_vni); |
0770b53b | 2441 | return; |
2442 | } | |
2443 | goto drop; | |
2444 | } | |
2445 | ||
0dfbdf41 | 2446 | dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port; |
3ad7a4b1 | 2447 | vni = (rdst->remote_vni) ? : default_vni; |
49f810f0 | 2448 | ifindex = rdst->remote_ifindex; |
1158632b | 2449 | local_ip = vxlan->cfg.saddr; |
d71785ff | 2450 | dst_cache = &rdst->dst_cache; |
0770b53b | 2451 | md->gbp = skb->mark; |
72f6d71e HL |
2452 | if (flags & VXLAN_F_TTL_INHERIT) { |
2453 | ttl = ip_tunnel_get_ttl(old_iph, skb); | |
2454 | } else { | |
2455 | ttl = vxlan->cfg.ttl; | |
2456 | if (!ttl && vxlan_addr_multicast(dst)) | |
2457 | ttl = 1; | |
2458 | } | |
0770b53b | 2459 | |
2460 | tos = vxlan->cfg.tos; | |
2461 | if (tos == 1) | |
2462 | tos = ip_tunnel_get_dsfield(old_iph, skb); | |
2463 | ||
2464 | if (dst->sa.sa_family == AF_INET) | |
2465 | udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX); | |
2466 | else | |
2467 | udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX); | |
2468 | label = vxlan->cfg.label; | |
ee122c79 TG |
2469 | } else { |
2470 | if (!info) { | |
2471 | WARN_ONCE(1, "%s: Missing encapsulation instructions\n", | |
2472 | dev->name); | |
2473 | goto drop; | |
2474 | } | |
b1be00a6 | 2475 | remote_ip.sa.sa_family = ip_tunnel_info_af(info); |
272d96a5 | 2476 | if (remote_ip.sa.sa_family == AF_INET) { |
a725e514 | 2477 | remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst; |
272d96a5 | 2478 | local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src; |
2479 | } else { | |
a725e514 | 2480 | remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst; |
272d96a5 | 2481 | local_ip.sin6.sin6_addr = info->key.u.ipv6.src; |
2482 | } | |
ee122c79 | 2483 | dst = &remote_ip; |
0770b53b | 2484 | dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port; |
2485 | vni = tunnel_id_to_key32(info->key.tun_id); | |
49f810f0 | 2486 | ifindex = 0; |
d71785ff | 2487 | dst_cache = &info->dst_cache; |
eadf52cf XL |
2488 | if (info->key.tun_flags & TUNNEL_VXLAN_OPT) { |
2489 | if (info->options_len < sizeof(*md)) | |
2490 | goto drop; | |
0770b53b | 2491 | md = ip_tunnel_info_opts(info); |
eadf52cf | 2492 | } |
a725e514 JB |
2493 | ttl = info->key.ttl; |
2494 | tos = info->key.tos; | |
e7f70af1 | 2495 | label = info->key.label; |
b4ed5cad | 2496 | udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM); |
a725e514 | 2497 | } |
0770b53b | 2498 | src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min, |
2499 | vxlan->cfg.port_max, true); | |
a725e514 | 2500 | |
56de859e | 2501 | rcu_read_lock(); |
e4c7ed41 | 2502 | if (dst->sa.sa_family == AF_INET) { |
c6fcc4fc | 2503 | struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock); |
c46b7897 | 2504 | struct rtable *rt; |
0770b53b | 2505 | __be16 df = 0; |
c6fcc4fc | 2506 | |
aab8cc36 AB |
2507 | if (!ifindex) |
2508 | ifindex = sock4->sock->sk->sk_bound_dev_if; | |
2509 | ||
49f810f0 | 2510 | rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos, |
272d96a5 | 2511 | dst->sin.sin_addr.s_addr, |
1158632b | 2512 | &local_ip.sin.sin_addr.s_addr, |
4ecb1d83 | 2513 | dst_port, src_port, |
d71785ff | 2514 | dst_cache, info); |
8ebd115b DM |
2515 | if (IS_ERR(rt)) { |
2516 | err = PTR_ERR(rt); | |
c46b7897 | 2517 | goto tx_error; |
8ebd115b | 2518 | } |
e4c7ed41 | 2519 | |
fee1fad7 | 2520 | if (!info) { |
b4d30697 | 2521 | /* Bypass encapsulation if the destination is local */ |
fee1fad7 | 2522 | err = encap_bypass_if_local(skb, dev, vxlan, dst, |
49f810f0 MS |
2523 | dst_port, ifindex, vni, |
2524 | &rt->dst, rt->rt_flags); | |
fee1fad7 | 2525 | if (err) |
56de859e | 2526 | goto out_unlock; |
b4d30697 SB |
2527 | |
2528 | if (vxlan->cfg.df == VXLAN_DF_SET) { | |
2529 | df = htons(IP_DF); | |
2530 | } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) { | |
2531 | struct ethhdr *eth = eth_hdr(skb); | |
2532 | ||
2533 | if (ntohs(eth->h_proto) == ETH_P_IPV6 || | |
2534 | (ntohs(eth->h_proto) == ETH_P_IP && | |
2535 | old_iph->frag_off & htons(IP_DF))) | |
2536 | df = htons(IP_DF); | |
2537 | } | |
fee1fad7 | 2538 | } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) { |
6ceb31ca | 2539 | df = htons(IP_DF); |
fee1fad7 | 2540 | } |
6ceb31ca | 2541 | |
c46b7897 | 2542 | ndst = &rt->dst; |
6b4f92af | 2543 | skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM); |
a93bf0ff | 2544 | |
71130f29 | 2545 | tos = ip_tunnel_ecn_encap(RT_TOS(tos), old_iph, skb); |
e4c7ed41 | 2546 | ttl = ttl ? : ip4_dst_hoplimit(&rt->dst); |
c46b7897 | 2547 | err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr), |
54bfd872 | 2548 | vni, md, flags, udp_sum); |
f491e56d | 2549 | if (err < 0) |
c46b7897 | 2550 | goto tx_error; |
f491e56d | 2551 | |
1158632b | 2552 | udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr, |
f491e56d JB |
2553 | dst->sin.sin_addr.s_addr, tos, ttl, df, |
2554 | src_port, dst_port, xnet, !udp_sum); | |
e4c7ed41 CW |
2555 | #if IS_ENABLED(CONFIG_IPV6) |
2556 | } else { | |
c6fcc4fc | 2557 | struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock); |
e4c7ed41 | 2558 | |
aab8cc36 AB |
2559 | if (!ifindex) |
2560 | ifindex = sock6->sock->sk->sk_bound_dev_if; | |
2561 | ||
49f810f0 | 2562 | ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos, |
272d96a5 | 2563 | label, &dst->sin6.sin6_addr, |
1158632b | 2564 | &local_ip.sin6.sin6_addr, |
4ecb1d83 | 2565 | dst_port, src_port, |
db3c6139 | 2566 | dst_cache, info); |
e5d4b29f | 2567 | if (IS_ERR(ndst)) { |
8ebd115b | 2568 | err = PTR_ERR(ndst); |
c46b7897 | 2569 | ndst = NULL; |
9dcc71e1 | 2570 | goto tx_error; |
e4c7ed41 | 2571 | } |
655c3de1 | 2572 | |
fee1fad7 | 2573 | if (!info) { |
2574 | u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags; | |
49560532 | 2575 | |
fee1fad7 | 2576 | err = encap_bypass_if_local(skb, dev, vxlan, dst, |
49f810f0 MS |
2577 | dst_port, ifindex, vni, |
2578 | ndst, rt6i_flags); | |
fee1fad7 | 2579 | if (err) |
56de859e | 2580 | goto out_unlock; |
fee1fad7 | 2581 | } |
35e2d115 | 2582 | |
6b4f92af | 2583 | skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM); |
a93bf0ff | 2584 | |
71130f29 | 2585 | tos = ip_tunnel_ecn_encap(RT_TOS(tos), old_iph, skb); |
e4c7ed41 | 2586 | ttl = ttl ? : ip6_dst_hoplimit(ndst); |
f491e56d JB |
2587 | skb_scrub_packet(skb, xnet); |
2588 | err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr), | |
54bfd872 | 2589 | vni, md, flags, udp_sum); |
c46b7897 | 2590 | if (err < 0) |
2591 | goto tx_error; | |
2592 | ||
0770b53b | 2593 | udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev, |
1158632b | 2594 | &local_ip.sin6.sin6_addr, |
272d96a5 | 2595 | &dst->sin6.sin6_addr, tos, ttl, |
e7f70af1 | 2596 | label, src_port, dst_port, !udp_sum); |
e4c7ed41 CW |
2597 | #endif |
2598 | } | |
56de859e JK |
2599 | out_unlock: |
2600 | rcu_read_unlock(); | |
4ad16930 | 2601 | return; |
d342894c | 2602 | |
2603 | drop: | |
2604 | dev->stats.tx_dropped++; | |
c46b7897 | 2605 | dev_kfree_skb(skb); |
2606 | return; | |
d342894c | 2607 | |
2608 | tx_error: | |
56de859e | 2609 | rcu_read_unlock(); |
655c3de1 | 2610 | if (err == -ELOOP) |
2611 | dev->stats.collisions++; | |
2612 | else if (err == -ENETUNREACH) | |
2613 | dev->stats.tx_carrier_errors++; | |
c46b7897 | 2614 | dst_release(ndst); |
d342894c | 2615 | dev->stats.tx_errors++; |
c46b7897 | 2616 | kfree_skb(skb); |
d342894c | 2617 | } |
2618 | ||
6681712d DS |
2619 | /* Transmit local packets over Vxlan |
2620 | * | |
2621 | * Outer IP header inherits ECN and DF from inner header. | |
2622 | * Outer UDP destination is the VXLAN assigned port. | |
2623 | * source port is based on hash of flow | |
2624 | */ | |
2625 | static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev) | |
2626 | { | |
2627 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
8bff3685 | 2628 | struct vxlan_rdst *rdst, *fdst = NULL; |
3093fbe7 | 2629 | const struct ip_tunnel_info *info; |
6681712d | 2630 | bool did_rsc = false; |
6681712d | 2631 | struct vxlan_fdb *f; |
8bff3685 | 2632 | struct ethhdr *eth; |
3ad7a4b1 | 2633 | __be32 vni = 0; |
6681712d | 2634 | |
61adedf3 | 2635 | info = skb_tunnel_info(skb); |
3093fbe7 | 2636 | |
6681712d | 2637 | skb_reset_mac_header(skb); |
6681712d | 2638 | |
dc5321d7 | 2639 | if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) { |
3ad7a4b1 RP |
2640 | if (info && info->mode & IP_TUNNEL_INFO_BRIDGE && |
2641 | info->mode & IP_TUNNEL_INFO_TX) { | |
2642 | vni = tunnel_id_to_key32(info->key.tun_id); | |
2643 | } else { | |
2644 | if (info && info->mode & IP_TUNNEL_INFO_TX) | |
2645 | vxlan_xmit_one(skb, dev, vni, NULL, false); | |
2646 | else | |
2647 | kfree_skb(skb); | |
2648 | return NETDEV_TX_OK; | |
2649 | } | |
47e5d1b0 JB |
2650 | } |
2651 | ||
dc5321d7 | 2652 | if (vxlan->cfg.flags & VXLAN_F_PROXY) { |
47e5d1b0 | 2653 | eth = eth_hdr(skb); |
f564f45c | 2654 | if (ntohs(eth->h_proto) == ETH_P_ARP) |
3ad7a4b1 | 2655 | return arp_reduce(dev, skb, vni); |
f564f45c | 2656 | #if IS_ENABLED(CONFIG_IPV6) |
8bff3685 XL |
2657 | else if (ntohs(eth->h_proto) == ETH_P_IPV6 && |
2658 | pskb_may_pull(skb, sizeof(struct ipv6hdr) + | |
2659 | sizeof(struct nd_msg)) && | |
2660 | ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) { | |
2661 | struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1); | |
2662 | ||
2663 | if (m->icmph.icmp6_code == 0 && | |
2664 | m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) | |
f1fb08f6 | 2665 | return neigh_reduce(dev, skb, vni); |
f564f45c CW |
2666 | } |
2667 | #endif | |
2668 | } | |
6681712d | 2669 | |
47e5d1b0 | 2670 | eth = eth_hdr(skb); |
3ad7a4b1 | 2671 | f = vxlan_find_mac(vxlan, eth->h_dest, vni); |
ae884082 DS |
2672 | did_rsc = false; |
2673 | ||
dc5321d7 | 2674 | if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) && |
e15a00aa CW |
2675 | (ntohs(eth->h_proto) == ETH_P_IP || |
2676 | ntohs(eth->h_proto) == ETH_P_IPV6)) { | |
ae884082 DS |
2677 | did_rsc = route_shortcircuit(dev, skb); |
2678 | if (did_rsc) | |
3ad7a4b1 | 2679 | f = vxlan_find_mac(vxlan, eth->h_dest, vni); |
ae884082 DS |
2680 | } |
2681 | ||
6681712d | 2682 | if (f == NULL) { |
3ad7a4b1 | 2683 | f = vxlan_find_mac(vxlan, all_zeros_mac, vni); |
afbd8bae | 2684 | if (f == NULL) { |
dc5321d7 | 2685 | if ((vxlan->cfg.flags & VXLAN_F_L2MISS) && |
afbd8bae MR |
2686 | !is_multicast_ether_addr(eth->h_dest)) |
2687 | vxlan_fdb_miss(vxlan, eth->h_dest); | |
2688 | ||
2689 | dev->stats.tx_dropped++; | |
8f646c92 | 2690 | kfree_skb(skb); |
afbd8bae MR |
2691 | return NETDEV_TX_OK; |
2692 | } | |
2693 | } | |
6681712d | 2694 | |
afbd8bae MR |
2695 | list_for_each_entry_rcu(rdst, &f->remotes, list) { |
2696 | struct sk_buff *skb1; | |
6681712d | 2697 | |
8f646c92 ED |
2698 | if (!fdst) { |
2699 | fdst = rdst; | |
2700 | continue; | |
2701 | } | |
afbd8bae MR |
2702 | skb1 = skb_clone(skb, GFP_ATOMIC); |
2703 | if (skb1) | |
3ad7a4b1 | 2704 | vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc); |
6681712d DS |
2705 | } |
2706 | ||
8f646c92 | 2707 | if (fdst) |
3ad7a4b1 | 2708 | vxlan_xmit_one(skb, dev, vni, fdst, did_rsc); |
8f646c92 ED |
2709 | else |
2710 | kfree_skb(skb); | |
4ad16930 | 2711 | return NETDEV_TX_OK; |
6681712d DS |
2712 | } |
2713 | ||
d342894c | 2714 | /* Walk the forwarding table and purge stale entries */ |
df7e828c | 2715 | static void vxlan_cleanup(struct timer_list *t) |
d342894c | 2716 | { |
df7e828c | 2717 | struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer); |
d342894c | 2718 | unsigned long next_timer = jiffies + FDB_AGE_INTERVAL; |
2719 | unsigned int h; | |
2720 | ||
2721 | if (!netif_running(vxlan->dev)) | |
2722 | return; | |
2723 | ||
d342894c | 2724 | for (h = 0; h < FDB_HASH_SIZE; ++h) { |
2725 | struct hlist_node *p, *n; | |
14e1d0fa | 2726 | |
fe1e0713 | 2727 | spin_lock(&vxlan->hash_lock[h]); |
d342894c | 2728 | hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) { |
2729 | struct vxlan_fdb *f | |
2730 | = container_of(p, struct vxlan_fdb, hlist); | |
2731 | unsigned long timeout; | |
2732 | ||
efb5f68f | 2733 | if (f->state & (NUD_PERMANENT | NUD_NOARP)) |
d342894c | 2734 | continue; |
2735 | ||
def499c9 RP |
2736 | if (f->flags & NTF_EXT_LEARNED) |
2737 | continue; | |
2738 | ||
0dfbdf41 | 2739 | timeout = f->used + vxlan->cfg.age_interval * HZ; |
d342894c | 2740 | if (time_before_eq(timeout, jiffies)) { |
2741 | netdev_dbg(vxlan->dev, | |
2742 | "garbage collect %pM\n", | |
2743 | f->eth_addr); | |
2744 | f->state = NUD_STALE; | |
0e6160f3 | 2745 | vxlan_fdb_destroy(vxlan, f, true, true); |
d342894c | 2746 | } else if (time_before(timeout, next_timer)) |
2747 | next_timer = timeout; | |
2748 | } | |
fe1e0713 | 2749 | spin_unlock(&vxlan->hash_lock[h]); |
d342894c | 2750 | } |
d342894c | 2751 | |
2752 | mod_timer(&vxlan->age_timer, next_timer); | |
2753 | } | |
2754 | ||
a53cb29b MB |
2755 | static void vxlan_vs_del_dev(struct vxlan_dev *vxlan) |
2756 | { | |
2757 | struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); | |
2758 | ||
2759 | spin_lock(&vn->sock_lock); | |
69e76661 JB |
2760 | hlist_del_init_rcu(&vxlan->hlist4.hlist); |
2761 | #if IS_ENABLED(CONFIG_IPV6) | |
2762 | hlist_del_init_rcu(&vxlan->hlist6.hlist); | |
2763 | #endif | |
a53cb29b MB |
2764 | spin_unlock(&vn->sock_lock); |
2765 | } | |
2766 | ||
69e76661 JB |
2767 | static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan, |
2768 | struct vxlan_dev_node *node) | |
9c2e24e1 | 2769 | { |
56ef9c90 | 2770 | struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); |
54bfd872 | 2771 | __be32 vni = vxlan->default_dst.remote_vni; |
9c2e24e1 | 2772 | |
69e76661 | 2773 | node->vxlan = vxlan; |
56ef9c90 | 2774 | spin_lock(&vn->sock_lock); |
69e76661 | 2775 | hlist_add_head_rcu(&node->hlist, vni_head(vs, vni)); |
56ef9c90 | 2776 | spin_unlock(&vn->sock_lock); |
9c2e24e1 PS |
2777 | } |
2778 | ||
d342894c | 2779 | /* Setup stats when device is created */ |
2780 | static int vxlan_init(struct net_device *dev) | |
2781 | { | |
384d91c2 TY |
2782 | struct vxlan_dev *vxlan = netdev_priv(dev); |
2783 | int err; | |
2784 | ||
1c213bd2 | 2785 | dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); |
e8171045 | 2786 | if (!dev->tstats) |
d342894c | 2787 | return -ENOMEM; |
2788 | ||
384d91c2 TY |
2789 | err = gro_cells_init(&vxlan->gro_cells, dev); |
2790 | if (err) { | |
2791 | free_percpu(dev->tstats); | |
2792 | return err; | |
2793 | } | |
2794 | ||
d342894c | 2795 | return 0; |
2796 | } | |
2797 | ||
3ad7a4b1 | 2798 | static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni) |
afbd8bae MR |
2799 | { |
2800 | struct vxlan_fdb *f; | |
fe1e0713 | 2801 | u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni); |
afbd8bae | 2802 | |
fe1e0713 | 2803 | spin_lock_bh(&vxlan->hash_lock[hash_index]); |
3ad7a4b1 | 2804 | f = __vxlan_find_mac(vxlan, all_zeros_mac, vni); |
afbd8bae | 2805 | if (f) |
0e6160f3 | 2806 | vxlan_fdb_destroy(vxlan, f, true, true); |
fe1e0713 | 2807 | spin_unlock_bh(&vxlan->hash_lock[hash_index]); |
afbd8bae MR |
2808 | } |
2809 | ||
ebf4063e SH |
2810 | static void vxlan_uninit(struct net_device *dev) |
2811 | { | |
2812 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
ebf4063e | 2813 | |
ad6c9986 SB |
2814 | gro_cells_destroy(&vxlan->gro_cells); |
2815 | ||
3ad7a4b1 | 2816 | vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni); |
afbd8bae | 2817 | |
ebf4063e SH |
2818 | free_percpu(dev->tstats); |
2819 | } | |
2820 | ||
d342894c | 2821 | /* Start ageing timer and join group when device is brought up */ |
2822 | static int vxlan_open(struct net_device *dev) | |
2823 | { | |
2824 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
205f356d | 2825 | int ret; |
56ef9c90 | 2826 | |
205f356d JB |
2827 | ret = vxlan_sock_add(vxlan); |
2828 | if (ret < 0) | |
2829 | return ret; | |
d342894c | 2830 | |
79d4a94f | 2831 | if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) { |
56ef9c90 | 2832 | ret = vxlan_igmp_join(vxlan); |
bef0057b MRL |
2833 | if (ret == -EADDRINUSE) |
2834 | ret = 0; | |
56ef9c90 | 2835 | if (ret) { |
205f356d | 2836 | vxlan_sock_release(vxlan); |
56ef9c90 MRL |
2837 | return ret; |
2838 | } | |
d342894c | 2839 | } |
2840 | ||
0dfbdf41 | 2841 | if (vxlan->cfg.age_interval) |
d342894c | 2842 | mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL); |
2843 | ||
56ef9c90 | 2844 | return ret; |
d342894c | 2845 | } |
2846 | ||
2847 | /* Purge the forwarding table */ | |
8b3f9337 | 2848 | static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all) |
d342894c | 2849 | { |
31fec5aa | 2850 | unsigned int h; |
d342894c | 2851 | |
d342894c | 2852 | for (h = 0; h < FDB_HASH_SIZE; ++h) { |
2853 | struct hlist_node *p, *n; | |
fe1e0713 L |
2854 | |
2855 | spin_lock_bh(&vxlan->hash_lock[h]); | |
d342894c | 2856 | hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) { |
2857 | struct vxlan_fdb *f | |
2858 | = container_of(p, struct vxlan_fdb, hlist); | |
8b3f9337 RP |
2859 | if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP))) |
2860 | continue; | |
afbd8bae MR |
2861 | /* the all_zeros_mac entry is deleted at vxlan_uninit */ |
2862 | if (!is_zero_ether_addr(f->eth_addr)) | |
0e6160f3 | 2863 | vxlan_fdb_destroy(vxlan, f, true, true); |
d342894c | 2864 | } |
fe1e0713 | 2865 | spin_unlock_bh(&vxlan->hash_lock[h]); |
d342894c | 2866 | } |
d342894c | 2867 | } |
2868 | ||
2869 | /* Cleanup timer and forwarding table on shutdown */ | |
2870 | static int vxlan_stop(struct net_device *dev) | |
2871 | { | |
2872 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
f01ec1c0 | 2873 | struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); |
56ef9c90 | 2874 | int ret = 0; |
d342894c | 2875 | |
24c0e683 | 2876 | if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) && |
f13b1689 | 2877 | !vxlan_group_used(vn, vxlan)) |
56ef9c90 | 2878 | ret = vxlan_igmp_leave(vxlan); |
d342894c | 2879 | |
2880 | del_timer_sync(&vxlan->age_timer); | |
2881 | ||
8b3f9337 | 2882 | vxlan_flush(vxlan, false); |
205f356d | 2883 | vxlan_sock_release(vxlan); |
d342894c | 2884 | |
56ef9c90 | 2885 | return ret; |
d342894c | 2886 | } |
2887 | ||
d342894c | 2888 | /* Stub, nothing needs to be done. */ |
2889 | static void vxlan_set_multicast_list(struct net_device *dev) | |
2890 | { | |
2891 | } | |
2892 | ||
91572088 | 2893 | static int vxlan_change_mtu(struct net_device *dev, int new_mtu) |
345010b5 | 2894 | { |
91572088 JW |
2895 | struct vxlan_dev *vxlan = netdev_priv(dev); |
2896 | struct vxlan_rdst *dst = &vxlan->default_dst; | |
2897 | struct net_device *lowerdev = __dev_get_by_index(vxlan->net, | |
2898 | dst->remote_ifindex); | |
ce44a4ae | 2899 | bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6); |
345010b5 | 2900 | |
91572088 JW |
2901 | /* This check is different than dev->max_mtu, because it looks at |
2902 | * the lowerdev->mtu, rather than the static dev->max_mtu | |
2903 | */ | |
2904 | if (lowerdev) { | |
2905 | int max_mtu = lowerdev->mtu - | |
2906 | (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM); | |
2907 | if (new_mtu > max_mtu) | |
72564b59 | 2908 | return -EINVAL; |
72564b59 DW |
2909 | } |
2910 | ||
345010b5 DB |
2911 | dev->mtu = new_mtu; |
2912 | return 0; | |
2913 | } | |
2914 | ||
fc4099f1 PS |
2915 | static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) |
2916 | { | |
2917 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
2918 | struct ip_tunnel_info *info = skb_tunnel_info(skb); | |
2919 | __be16 sport, dport; | |
2920 | ||
2921 | sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min, | |
2922 | vxlan->cfg.port_max, true); | |
2923 | dport = info->key.tp_dst ? : vxlan->cfg.dst_port; | |
2924 | ||
239e944f | 2925 | if (ip_tunnel_info_af(info) == AF_INET) { |
c6fcc4fc | 2926 | struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock); |
1a8496ba JB |
2927 | struct rtable *rt; |
2928 | ||
655c3de1 | 2929 | rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos, |
1a8496ba | 2930 | info->key.u.ipv4.dst, |
22f0708a PA |
2931 | &info->key.u.ipv4.src, dport, sport, |
2932 | &info->dst_cache, info); | |
1a8496ba JB |
2933 | if (IS_ERR(rt)) |
2934 | return PTR_ERR(rt); | |
2935 | ip_rt_put(rt); | |
239e944f JB |
2936 | } else { |
2937 | #if IS_ENABLED(CONFIG_IPV6) | |
03dc52a8 | 2938 | struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock); |
239e944f JB |
2939 | struct dst_entry *ndst; |
2940 | ||
655c3de1 | 2941 | ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos, |
e7f70af1 | 2942 | info->key.label, &info->key.u.ipv6.dst, |
22f0708a PA |
2943 | &info->key.u.ipv6.src, dport, sport, |
2944 | &info->dst_cache, info); | |
239e944f JB |
2945 | if (IS_ERR(ndst)) |
2946 | return PTR_ERR(ndst); | |
2947 | dst_release(ndst); | |
239e944f JB |
2948 | #else /* !CONFIG_IPV6 */ |
2949 | return -EPFNOSUPPORT; | |
2950 | #endif | |
2951 | } | |
1a8496ba JB |
2952 | info->key.tp_src = sport; |
2953 | info->key.tp_dst = dport; | |
239e944f | 2954 | return 0; |
fc4099f1 PS |
2955 | } |
2956 | ||
0c867c9b | 2957 | static const struct net_device_ops vxlan_netdev_ether_ops = { |
d342894c | 2958 | .ndo_init = vxlan_init, |
ebf4063e | 2959 | .ndo_uninit = vxlan_uninit, |
d342894c | 2960 | .ndo_open = vxlan_open, |
2961 | .ndo_stop = vxlan_stop, | |
2962 | .ndo_start_xmit = vxlan_xmit, | |
e8171045 | 2963 | .ndo_get_stats64 = ip_tunnel_get_stats64, |
d342894c | 2964 | .ndo_set_rx_mode = vxlan_set_multicast_list, |
345010b5 | 2965 | .ndo_change_mtu = vxlan_change_mtu, |
d342894c | 2966 | .ndo_validate_addr = eth_validate_addr, |
2967 | .ndo_set_mac_address = eth_mac_addr, | |
2968 | .ndo_fdb_add = vxlan_fdb_add, | |
2969 | .ndo_fdb_del = vxlan_fdb_delete, | |
2970 | .ndo_fdb_dump = vxlan_fdb_dump, | |
474c3c89 | 2971 | .ndo_fdb_get = vxlan_fdb_get, |
fc4099f1 | 2972 | .ndo_fill_metadata_dst = vxlan_fill_metadata_dst, |
8f1af75d | 2973 | .ndo_change_proto_down = dev_change_proto_down_generic, |
d342894c | 2974 | }; |
2975 | ||
e1e5314d JB |
2976 | static const struct net_device_ops vxlan_netdev_raw_ops = { |
2977 | .ndo_init = vxlan_init, | |
2978 | .ndo_uninit = vxlan_uninit, | |
2979 | .ndo_open = vxlan_open, | |
2980 | .ndo_stop = vxlan_stop, | |
2981 | .ndo_start_xmit = vxlan_xmit, | |
2982 | .ndo_get_stats64 = ip_tunnel_get_stats64, | |
2983 | .ndo_change_mtu = vxlan_change_mtu, | |
2984 | .ndo_fill_metadata_dst = vxlan_fill_metadata_dst, | |
2985 | }; | |
2986 | ||
d342894c | 2987 | /* Info for udev, that this is a virtual tunnel endpoint */ |
2988 | static struct device_type vxlan_type = { | |
2989 | .name = "vxlan", | |
2990 | }; | |
2991 | ||
e5de25dc | 2992 | /* Calls the ndo_udp_tunnel_add of the caller in order to |
35e42379 | 2993 | * supply the listening VXLAN udp ports. Callers are expected |
e5de25dc | 2994 | * to implement the ndo_udp_tunnel_add. |
53cf5275 | 2995 | */ |
2d2b13fc | 2996 | static void vxlan_offload_rx_ports(struct net_device *dev, bool push) |
53cf5275 JG |
2997 | { |
2998 | struct vxlan_sock *vs; | |
2999 | struct net *net = dev_net(dev); | |
3000 | struct vxlan_net *vn = net_generic(net, vxlan_net_id); | |
35e42379 | 3001 | unsigned int i; |
53cf5275 JG |
3002 | |
3003 | spin_lock(&vn->sock_lock); | |
3004 | for (i = 0; i < PORT_HASH_SIZE; ++i) { | |
2d2b13fc SD |
3005 | hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) { |
3006 | unsigned short type; | |
3007 | ||
3008 | if (vs->flags & VXLAN_F_GPE) | |
3009 | type = UDP_TUNNEL_TYPE_VXLAN_GPE; | |
3010 | else | |
3011 | type = UDP_TUNNEL_TYPE_VXLAN; | |
3012 | ||
3013 | if (push) | |
3014 | udp_tunnel_push_rx_port(dev, vs->sock, type); | |
3015 | else | |
3016 | udp_tunnel_drop_rx_port(dev, vs->sock, type); | |
3017 | } | |
53cf5275 JG |
3018 | } |
3019 | spin_unlock(&vn->sock_lock); | |
3020 | } | |
53cf5275 | 3021 | |
d342894c | 3022 | /* Initialize the device structure. */ |
3023 | static void vxlan_setup(struct net_device *dev) | |
3024 | { | |
3025 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
31fec5aa | 3026 | unsigned int h; |
d342894c | 3027 | |
65226ef8 JB |
3028 | eth_hw_addr_random(dev); |
3029 | ether_setup(dev); | |
3030 | ||
cf124db5 | 3031 | dev->needs_free_netdev = true; |
d342894c | 3032 | SET_NETDEV_DEVTYPE(dev, &vxlan_type); |
3033 | ||
d342894c | 3034 | dev->features |= NETIF_F_LLTX; |
d6727fe3 | 3035 | dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM; |
0afb1666 | 3036 | dev->features |= NETIF_F_RXCSUM; |
05c0db08 | 3037 | dev->features |= NETIF_F_GSO_SOFTWARE; |
0afb1666 | 3038 | |
1eaa8178 | 3039 | dev->vlan_features = dev->features; |
0afb1666 | 3040 | dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM; |
05c0db08 | 3041 | dev->hw_features |= NETIF_F_GSO_SOFTWARE; |
02875878 | 3042 | netif_keep_dst(dev); |
0c867c9b | 3043 | dev->priv_flags |= IFF_NO_QUEUE; |
d342894c | 3044 | |
a985343b MS |
3045 | /* MTU range: 68 - 65535 */ |
3046 | dev->min_mtu = ETH_MIN_MTU; | |
3047 | dev->max_mtu = ETH_MAX_MTU; | |
3048 | ||
553675fb | 3049 | INIT_LIST_HEAD(&vxlan->next); |
d342894c | 3050 | |
df7e828c | 3051 | timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE); |
d342894c | 3052 | |
3053 | vxlan->dev = dev; | |
3054 | ||
fe1e0713 L |
3055 | for (h = 0; h < FDB_HASH_SIZE; ++h) { |
3056 | spin_lock_init(&vxlan->hash_lock[h]); | |
d342894c | 3057 | INIT_HLIST_HEAD(&vxlan->fdb_head[h]); |
fe1e0713 | 3058 | } |
d342894c | 3059 | } |
3060 | ||
0c867c9b JB |
3061 | static void vxlan_ether_setup(struct net_device *dev) |
3062 | { | |
0c867c9b JB |
3063 | dev->priv_flags &= ~IFF_TX_SKB_SHARING; |
3064 | dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; | |
3065 | dev->netdev_ops = &vxlan_netdev_ether_ops; | |
3066 | } | |
3067 | ||
e1e5314d JB |
3068 | static void vxlan_raw_setup(struct net_device *dev) |
3069 | { | |
65226ef8 | 3070 | dev->header_ops = NULL; |
e1e5314d JB |
3071 | dev->type = ARPHRD_NONE; |
3072 | dev->hard_header_len = 0; | |
3073 | dev->addr_len = 0; | |
e1e5314d JB |
3074 | dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST; |
3075 | dev->netdev_ops = &vxlan_netdev_raw_ops; | |
3076 | } | |
3077 | ||
d342894c | 3078 | static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = { |
3079 | [IFLA_VXLAN_ID] = { .type = NLA_U32 }, | |
c593642c | 3080 | [IFLA_VXLAN_GROUP] = { .len = sizeof_field(struct iphdr, daddr) }, |
e4c7ed41 | 3081 | [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) }, |
d342894c | 3082 | [IFLA_VXLAN_LINK] = { .type = NLA_U32 }, |
c593642c | 3083 | [IFLA_VXLAN_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) }, |
e4c7ed41 | 3084 | [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) }, |
d342894c | 3085 | [IFLA_VXLAN_TOS] = { .type = NLA_U8 }, |
3086 | [IFLA_VXLAN_TTL] = { .type = NLA_U8 }, | |
e7f70af1 | 3087 | [IFLA_VXLAN_LABEL] = { .type = NLA_U32 }, |
d342894c | 3088 | [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 }, |
3089 | [IFLA_VXLAN_AGEING] = { .type = NLA_U32 }, | |
3090 | [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 }, | |
05f47d69 | 3091 | [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) }, |
e4f67add DS |
3092 | [IFLA_VXLAN_PROXY] = { .type = NLA_U8 }, |
3093 | [IFLA_VXLAN_RSC] = { .type = NLA_U8 }, | |
3094 | [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 }, | |
3095 | [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 }, | |
f8a9b1bc | 3096 | [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 }, |
823aa873 | 3097 | [IFLA_VXLAN_PORT] = { .type = NLA_U16 }, |
5c91ae08 TH |
3098 | [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 }, |
3099 | [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, | |
3100 | [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, | |
dfd8645e TH |
3101 | [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 }, |
3102 | [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 }, | |
3511494c | 3103 | [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, }, |
e1e5314d | 3104 | [IFLA_VXLAN_GPE] = { .type = NLA_FLAG, }, |
0ace2ca8 | 3105 | [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG }, |
72f6d71e | 3106 | [IFLA_VXLAN_TTL_INHERIT] = { .type = NLA_FLAG }, |
b4d30697 | 3107 | [IFLA_VXLAN_DF] = { .type = NLA_U8 }, |
d342894c | 3108 | }; |
3109 | ||
a8b8a889 MS |
3110 | static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[], |
3111 | struct netlink_ext_ack *extack) | |
d342894c | 3112 | { |
3113 | if (tb[IFLA_ADDRESS]) { | |
3114 | if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { | |
653ef6a3 GM |
3115 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], |
3116 | "Provided link layer address is not Ethernet"); | |
d342894c | 3117 | return -EINVAL; |
3118 | } | |
3119 | ||
3120 | if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { | |
653ef6a3 GM |
3121 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], |
3122 | "Provided Ethernet address is not unicast"); | |
d342894c | 3123 | return -EADDRNOTAVAIL; |
3124 | } | |
3125 | } | |
3126 | ||
a985343b | 3127 | if (tb[IFLA_MTU]) { |
019b13ae | 3128 | u32 mtu = nla_get_u32(tb[IFLA_MTU]); |
a985343b | 3129 | |
653ef6a3 GM |
3130 | if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) { |
3131 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU], | |
3132 | "MTU must be between 68 and 65535"); | |
a985343b | 3133 | return -EINVAL; |
653ef6a3 | 3134 | } |
a985343b MS |
3135 | } |
3136 | ||
653ef6a3 GM |
3137 | if (!data) { |
3138 | NL_SET_ERR_MSG(extack, | |
3139 | "Required attributes not provided to perform the operation"); | |
d342894c | 3140 | return -EINVAL; |
653ef6a3 | 3141 | } |
d342894c | 3142 | |
3143 | if (data[IFLA_VXLAN_ID]) { | |
a985343b MS |
3144 | u32 id = nla_get_u32(data[IFLA_VXLAN_ID]); |
3145 | ||
653ef6a3 GM |
3146 | if (id >= VXLAN_N_VID) { |
3147 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], | |
3148 | "VXLAN ID must be lower than 16777216"); | |
d342894c | 3149 | return -ERANGE; |
653ef6a3 | 3150 | } |
d342894c | 3151 | } |
3152 | ||
05f47d69 | 3153 | if (data[IFLA_VXLAN_PORT_RANGE]) { |
3154 | const struct ifla_vxlan_port_range *p | |
3155 | = nla_data(data[IFLA_VXLAN_PORT_RANGE]); | |
3156 | ||
3157 | if (ntohs(p->high) < ntohs(p->low)) { | |
653ef6a3 GM |
3158 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE], |
3159 | "Invalid source port range"); | |
05f47d69 | 3160 | return -EINVAL; |
3161 | } | |
3162 | } | |
3163 | ||
b4d30697 SB |
3164 | if (data[IFLA_VXLAN_DF]) { |
3165 | enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]); | |
3166 | ||
3167 | if (df < 0 || df > VXLAN_DF_MAX) { | |
3168 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_DF], | |
3169 | "Invalid DF attribute"); | |
3170 | return -EINVAL; | |
3171 | } | |
3172 | } | |
3173 | ||
d342894c | 3174 | return 0; |
3175 | } | |
3176 | ||
1b13c97f YB |
3177 | static void vxlan_get_drvinfo(struct net_device *netdev, |
3178 | struct ethtool_drvinfo *drvinfo) | |
3179 | { | |
3180 | strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version)); | |
3181 | strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver)); | |
3182 | } | |
3183 | ||
36fe3a61 MS |
3184 | static int vxlan_get_link_ksettings(struct net_device *dev, |
3185 | struct ethtool_link_ksettings *cmd) | |
3186 | { | |
3187 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
3188 | struct vxlan_rdst *dst = &vxlan->default_dst; | |
3189 | struct net_device *lowerdev = __dev_get_by_index(vxlan->net, | |
3190 | dst->remote_ifindex); | |
3191 | ||
3192 | if (!lowerdev) { | |
3193 | cmd->base.duplex = DUPLEX_UNKNOWN; | |
3194 | cmd->base.port = PORT_OTHER; | |
3195 | cmd->base.speed = SPEED_UNKNOWN; | |
3196 | ||
3197 | return 0; | |
3198 | } | |
3199 | ||
3200 | return __ethtool_get_link_ksettings(lowerdev, cmd); | |
3201 | } | |
3202 | ||
1b13c97f | 3203 | static const struct ethtool_ops vxlan_ethtool_ops = { |
36fe3a61 MS |
3204 | .get_drvinfo = vxlan_get_drvinfo, |
3205 | .get_link = ethtool_op_get_link, | |
3206 | .get_link_ksettings = vxlan_get_link_ksettings, | |
1b13c97f YB |
3207 | }; |
3208 | ||
3ee64f39 | 3209 | static struct socket *vxlan_create_sock(struct net *net, bool ipv6, |
aab8cc36 | 3210 | __be16 port, u32 flags, int ifindex) |
553675fb | 3211 | { |
e4c7ed41 | 3212 | struct socket *sock; |
3ee64f39 TH |
3213 | struct udp_port_cfg udp_conf; |
3214 | int err; | |
e4c7ed41 | 3215 | |
3ee64f39 | 3216 | memset(&udp_conf, 0, sizeof(udp_conf)); |
e4c7ed41 | 3217 | |
3ee64f39 TH |
3218 | if (ipv6) { |
3219 | udp_conf.family = AF_INET6; | |
3ee64f39 | 3220 | udp_conf.use_udp6_rx_checksums = |
3dc2b6a8 | 3221 | !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX); |
a43a9ef6 | 3222 | udp_conf.ipv6_v6only = 1; |
3ee64f39 TH |
3223 | } else { |
3224 | udp_conf.family = AF_INET; | |
e4c7ed41 | 3225 | } |
e4c7ed41 | 3226 | |
3ee64f39 | 3227 | udp_conf.local_udp_port = port; |
aab8cc36 | 3228 | udp_conf.bind_ifindex = ifindex; |
553675fb | 3229 | |
3ee64f39 TH |
3230 | /* Open UDP socket */ |
3231 | err = udp_sock_create(net, &udp_conf, &sock); | |
3232 | if (err < 0) | |
3233 | return ERR_PTR(err); | |
e4c7ed41 | 3234 | |
39deb2c7 | 3235 | return sock; |
e4c7ed41 CW |
3236 | } |
3237 | ||
3238 | /* Create new listen socket if needed */ | |
b1be00a6 | 3239 | static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6, |
aab8cc36 AB |
3240 | __be16 port, u32 flags, |
3241 | int ifindex) | |
e4c7ed41 CW |
3242 | { |
3243 | struct vxlan_net *vn = net_generic(net, vxlan_net_id); | |
3244 | struct vxlan_sock *vs; | |
3245 | struct socket *sock; | |
e4c7ed41 | 3246 | unsigned int h; |
acbf74a7 | 3247 | struct udp_tunnel_sock_cfg tunnel_cfg; |
e4c7ed41 | 3248 | |
dc01e7d3 | 3249 | vs = kzalloc(sizeof(*vs), GFP_KERNEL); |
e4c7ed41 CW |
3250 | if (!vs) |
3251 | return ERR_PTR(-ENOMEM); | |
3252 | ||
3253 | for (h = 0; h < VNI_HASH_SIZE; ++h) | |
3254 | INIT_HLIST_HEAD(&vs->vni_list[h]); | |
3255 | ||
aab8cc36 | 3256 | sock = vxlan_create_sock(net, ipv6, port, flags, ifindex); |
39deb2c7 | 3257 | if (IS_ERR(sock)) { |
553675fb | 3258 | kfree(vs); |
e50fddc8 | 3259 | return ERR_CAST(sock); |
553675fb | 3260 | } |
e4c7ed41 CW |
3261 | |
3262 | vs->sock = sock; | |
66af846f | 3263 | refcount_set(&vs->refcnt, 1); |
af33c1ad | 3264 | vs->flags = (flags & VXLAN_F_RCV_FLAGS); |
553675fb | 3265 | |
9c2e24e1 PS |
3266 | spin_lock(&vn->sock_lock); |
3267 | hlist_add_head_rcu(&vs->hlist, vs_head(net, port)); | |
e7b3db5e | 3268 | udp_tunnel_notify_add_rx_port(sock, |
b9adcd69 AD |
3269 | (vs->flags & VXLAN_F_GPE) ? |
3270 | UDP_TUNNEL_TYPE_VXLAN_GPE : | |
e7b3db5e | 3271 | UDP_TUNNEL_TYPE_VXLAN); |
9c2e24e1 | 3272 | spin_unlock(&vn->sock_lock); |
553675fb | 3273 | |
3274 | /* Mark socket as an encapsulation socket. */ | |
5602c48c | 3275 | memset(&tunnel_cfg, 0, sizeof(tunnel_cfg)); |
acbf74a7 AZ |
3276 | tunnel_cfg.sk_user_data = vs; |
3277 | tunnel_cfg.encap_type = 1; | |
f2d1968e | 3278 | tunnel_cfg.encap_rcv = vxlan_rcv; |
c3a43b9f | 3279 | tunnel_cfg.encap_err_lookup = vxlan_err_lookup; |
acbf74a7 | 3280 | tunnel_cfg.encap_destroy = NULL; |
5602c48c TH |
3281 | tunnel_cfg.gro_receive = vxlan_gro_receive; |
3282 | tunnel_cfg.gro_complete = vxlan_gro_complete; | |
acbf74a7 AZ |
3283 | |
3284 | setup_udp_tunnel_sock(net, sock, &tunnel_cfg); | |
e4c7ed41 | 3285 | |
9c2e24e1 PS |
3286 | return vs; |
3287 | } | |
3288 | ||
b1be00a6 | 3289 | static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6) |
9c2e24e1 | 3290 | { |
205f356d JB |
3291 | struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); |
3292 | struct vxlan_sock *vs = NULL; | |
69e76661 | 3293 | struct vxlan_dev_node *node; |
aab8cc36 AB |
3294 | int l3mdev_index = 0; |
3295 | ||
3296 | if (vxlan->cfg.remote_ifindex) | |
3297 | l3mdev_index = l3mdev_master_upper_ifindex_by_index( | |
3298 | vxlan->net, vxlan->cfg.remote_ifindex); | |
9c2e24e1 | 3299 | |
205f356d | 3300 | if (!vxlan->cfg.no_share) { |
56ef9c90 | 3301 | spin_lock(&vn->sock_lock); |
205f356d | 3302 | vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET, |
aab8cc36 AB |
3303 | vxlan->cfg.dst_port, vxlan->cfg.flags, |
3304 | l3mdev_index); | |
66af846f | 3305 | if (vs && !refcount_inc_not_zero(&vs->refcnt)) { |
56ef9c90 | 3306 | spin_unlock(&vn->sock_lock); |
205f356d | 3307 | return -EBUSY; |
56ef9c90 MRL |
3308 | } |
3309 | spin_unlock(&vn->sock_lock); | |
3310 | } | |
205f356d | 3311 | if (!vs) |
b1be00a6 | 3312 | vs = vxlan_socket_create(vxlan->net, ipv6, |
aab8cc36 AB |
3313 | vxlan->cfg.dst_port, vxlan->cfg.flags, |
3314 | l3mdev_index); | |
205f356d JB |
3315 | if (IS_ERR(vs)) |
3316 | return PTR_ERR(vs); | |
b1be00a6 | 3317 | #if IS_ENABLED(CONFIG_IPV6) |
69e76661 | 3318 | if (ipv6) { |
c6fcc4fc | 3319 | rcu_assign_pointer(vxlan->vn6_sock, vs); |
69e76661 JB |
3320 | node = &vxlan->hlist6; |
3321 | } else | |
b1be00a6 | 3322 | #endif |
69e76661 | 3323 | { |
c6fcc4fc | 3324 | rcu_assign_pointer(vxlan->vn4_sock, vs); |
69e76661 JB |
3325 | node = &vxlan->hlist4; |
3326 | } | |
3327 | vxlan_vs_add_dev(vs, vxlan, node); | |
205f356d | 3328 | return 0; |
553675fb | 3329 | } |
3330 | ||
b1be00a6 JB |
3331 | static int vxlan_sock_add(struct vxlan_dev *vxlan) |
3332 | { | |
dc5321d7 MS |
3333 | bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA; |
3334 | bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata; | |
d074bf96 | 3335 | bool ipv4 = !ipv6 || metadata; |
b1be00a6 JB |
3336 | int ret = 0; |
3337 | ||
c6fcc4fc | 3338 | RCU_INIT_POINTER(vxlan->vn4_sock, NULL); |
b1be00a6 | 3339 | #if IS_ENABLED(CONFIG_IPV6) |
c6fcc4fc | 3340 | RCU_INIT_POINTER(vxlan->vn6_sock, NULL); |
d074bf96 | 3341 | if (ipv6) { |
b1be00a6 | 3342 | ret = __vxlan_sock_add(vxlan, true); |
d074bf96 JB |
3343 | if (ret < 0 && ret != -EAFNOSUPPORT) |
3344 | ipv4 = false; | |
3345 | } | |
b1be00a6 | 3346 | #endif |
d074bf96 | 3347 | if (ipv4) |
b1be00a6 JB |
3348 | ret = __vxlan_sock_add(vxlan, false); |
3349 | if (ret < 0) | |
3350 | vxlan_sock_release(vxlan); | |
3351 | return ret; | |
3352 | } | |
3353 | ||
a985343b MS |
3354 | static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf, |
3355 | struct net_device **lower, | |
653ef6a3 GM |
3356 | struct vxlan_dev *old, |
3357 | struct netlink_ext_ack *extack) | |
d342894c | 3358 | { |
33564bbb | 3359 | struct vxlan_net *vn = net_generic(src_net, vxlan_net_id); |
a985343b | 3360 | struct vxlan_dev *tmp; |
e4c7ed41 | 3361 | bool use_ipv6 = false; |
d342894c | 3362 | |
a985343b MS |
3363 | if (conf->flags & VXLAN_F_GPE) { |
3364 | /* For now, allow GPE only together with | |
3365 | * COLLECT_METADATA. This can be relaxed later; in such | |
3366 | * case, the other side of the PtP link will have to be | |
3367 | * provided. | |
3368 | */ | |
3369 | if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) || | |
3370 | !(conf->flags & VXLAN_F_COLLECT_METADATA)) { | |
653ef6a3 GM |
3371 | NL_SET_ERR_MSG(extack, |
3372 | "VXLAN GPE does not support this combination of attributes"); | |
a985343b | 3373 | return -EINVAL; |
3555621d | 3374 | } |
e1e5314d | 3375 | } |
0c867c9b | 3376 | |
ce44a4ae MS |
3377 | if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) { |
3378 | /* Unless IPv6 is explicitly requested, assume IPv4 */ | |
a985343b | 3379 | conf->remote_ip.sa.sa_family = AF_INET; |
ce44a4ae MS |
3380 | conf->saddr.sa.sa_family = AF_INET; |
3381 | } else if (!conf->remote_ip.sa.sa_family) { | |
3382 | conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family; | |
3383 | } else if (!conf->saddr.sa.sa_family) { | |
3384 | conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family; | |
3385 | } | |
3386 | ||
653ef6a3 GM |
3387 | if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) { |
3388 | NL_SET_ERR_MSG(extack, | |
3389 | "Local and remote address must be from the same family"); | |
ce44a4ae | 3390 | return -EINVAL; |
653ef6a3 | 3391 | } |
e4c7ed41 | 3392 | |
653ef6a3 GM |
3393 | if (vxlan_addr_multicast(&conf->saddr)) { |
3394 | NL_SET_ERR_MSG(extack, "Local address cannot be multicast"); | |
0f22a3c6 | 3395 | return -EINVAL; |
653ef6a3 | 3396 | } |
0f22a3c6 | 3397 | |
ce44a4ae | 3398 | if (conf->saddr.sa.sa_family == AF_INET6) { |
653ef6a3 GM |
3399 | if (!IS_ENABLED(CONFIG_IPV6)) { |
3400 | NL_SET_ERR_MSG(extack, | |
3401 | "IPv6 support not enabled in the kernel"); | |
057ba29b | 3402 | return -EPFNOSUPPORT; |
653ef6a3 | 3403 | } |
e4c7ed41 | 3404 | use_ipv6 = true; |
a985343b | 3405 | conf->flags |= VXLAN_F_IPV6; |
0f22a3c6 MS |
3406 | |
3407 | if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) { | |
3408 | int local_type = | |
3409 | ipv6_addr_type(&conf->saddr.sin6.sin6_addr); | |
3410 | int remote_type = | |
3411 | ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr); | |
3412 | ||
3413 | if (local_type & IPV6_ADDR_LINKLOCAL) { | |
3414 | if (!(remote_type & IPV6_ADDR_LINKLOCAL) && | |
653ef6a3 GM |
3415 | (remote_type != IPV6_ADDR_ANY)) { |
3416 | NL_SET_ERR_MSG(extack, | |
3417 | "Invalid combination of local and remote address scopes"); | |
0f22a3c6 | 3418 | return -EINVAL; |
653ef6a3 | 3419 | } |
0f22a3c6 MS |
3420 | |
3421 | conf->flags |= VXLAN_F_IPV6_LINKLOCAL; | |
3422 | } else { | |
3423 | if (remote_type == | |
653ef6a3 GM |
3424 | (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) { |
3425 | NL_SET_ERR_MSG(extack, | |
3426 | "Invalid combination of local and remote address scopes"); | |
0f22a3c6 | 3427 | return -EINVAL; |
653ef6a3 | 3428 | } |
0f22a3c6 MS |
3429 | |
3430 | conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL; | |
3431 | } | |
3432 | } | |
057ba29b | 3433 | } |
d342894c | 3434 | |
653ef6a3 GM |
3435 | if (conf->label && !use_ipv6) { |
3436 | NL_SET_ERR_MSG(extack, | |
3437 | "Label attribute only applies to IPv6 VXLAN devices"); | |
e7f70af1 | 3438 | return -EINVAL; |
653ef6a3 | 3439 | } |
e7f70af1 | 3440 | |
a985343b MS |
3441 | if (conf->remote_ifindex) { |
3442 | struct net_device *lowerdev; | |
34e02aa1 | 3443 | |
a985343b | 3444 | lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex); |
653ef6a3 GM |
3445 | if (!lowerdev) { |
3446 | NL_SET_ERR_MSG(extack, | |
3447 | "Invalid local interface, device not found"); | |
34e02aa1 | 3448 | return -ENODEV; |
653ef6a3 | 3449 | } |
d342894c | 3450 | |
e4c7ed41 CW |
3451 | #if IS_ENABLED(CONFIG_IPV6) |
3452 | if (use_ipv6) { | |
3453 | struct inet6_dev *idev = __in6_dev_get(lowerdev); | |
653ef6a3 GM |
3454 | if (idev && idev->cnf.disable_ipv6) { |
3455 | NL_SET_ERR_MSG(extack, | |
3456 | "IPv6 support disabled by administrator"); | |
e4c7ed41 | 3457 | return -EPERM; |
653ef6a3 | 3458 | } |
e4c7ed41 CW |
3459 | } |
3460 | #endif | |
3461 | ||
a985343b MS |
3462 | *lower = lowerdev; |
3463 | } else { | |
653ef6a3 GM |
3464 | if (vxlan_addr_multicast(&conf->remote_ip)) { |
3465 | NL_SET_ERR_MSG(extack, | |
3466 | "Local interface required for multicast remote destination"); | |
3467 | ||
a985343b | 3468 | return -EINVAL; |
653ef6a3 | 3469 | } |
1ba56fb4 | 3470 | |
0f22a3c6 | 3471 | #if IS_ENABLED(CONFIG_IPV6) |
653ef6a3 GM |
3472 | if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) { |
3473 | NL_SET_ERR_MSG(extack, | |
3474 | "Local interface required for link-local local/remote addresses"); | |
0f22a3c6 | 3475 | return -EINVAL; |
653ef6a3 | 3476 | } |
0f22a3c6 MS |
3477 | #endif |
3478 | ||
a985343b | 3479 | *lower = NULL; |
9dc2ad10 | 3480 | } |
d342894c | 3481 | |
a985343b MS |
3482 | if (!conf->dst_port) { |
3483 | if (conf->flags & VXLAN_F_GPE) | |
3484 | conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */ | |
3485 | else | |
3486 | conf->dst_port = htons(vxlan_port); | |
d6acfeb1 FM |
3487 | } |
3488 | ||
a985343b MS |
3489 | if (!conf->age_interval) |
3490 | conf->age_interval = FDB_AGE_DEFAULT; | |
91572088 | 3491 | |
a985343b MS |
3492 | list_for_each_entry(tmp, &vn->vxlan_list, next) { |
3493 | if (tmp == old) | |
3494 | continue; | |
91572088 | 3495 | |
49f810f0 MS |
3496 | if (tmp->cfg.vni != conf->vni) |
3497 | continue; | |
3498 | if (tmp->cfg.dst_port != conf->dst_port) | |
3499 | continue; | |
3500 | if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) != | |
ce44a4ae | 3501 | (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6))) |
49f810f0 MS |
3502 | continue; |
3503 | ||
3504 | if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) && | |
3505 | tmp->cfg.remote_ifindex != conf->remote_ifindex) | |
3506 | continue; | |
3507 | ||
653ef6a3 GM |
3508 | NL_SET_ERR_MSG(extack, |
3509 | "A VXLAN device with the specified VNI already exists"); | |
49f810f0 | 3510 | return -EEXIST; |
a985343b | 3511 | } |
91572088 | 3512 | |
a985343b MS |
3513 | return 0; |
3514 | } | |
3515 | ||
3516 | static void vxlan_config_apply(struct net_device *dev, | |
3517 | struct vxlan_config *conf, | |
889ce937 SD |
3518 | struct net_device *lowerdev, |
3519 | struct net *src_net, | |
3520 | bool changelink) | |
a985343b MS |
3521 | { |
3522 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
3523 | struct vxlan_rdst *dst = &vxlan->default_dst; | |
3524 | unsigned short needed_headroom = ETH_HLEN; | |
3525 | bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6); | |
3526 | int max_mtu = ETH_MAX_MTU; | |
3527 | ||
3528 | if (!changelink) { | |
3529 | if (conf->flags & VXLAN_F_GPE) | |
3530 | vxlan_raw_setup(dev); | |
3531 | else | |
3532 | vxlan_ether_setup(dev); | |
3533 | ||
3534 | if (conf->mtu) | |
3535 | dev->mtu = conf->mtu; | |
889ce937 SD |
3536 | |
3537 | vxlan->net = src_net; | |
a985343b MS |
3538 | } |
3539 | ||
3540 | dst->remote_vni = conf->vni; | |
3541 | ||
3542 | memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip)); | |
3543 | ||
3544 | if (lowerdev) { | |
3545 | dst->remote_ifindex = conf->remote_ifindex; | |
3546 | ||
3547 | dev->gso_max_size = lowerdev->gso_max_size; | |
3548 | dev->gso_max_segs = lowerdev->gso_max_segs; | |
91572088 | 3549 | |
a985343b | 3550 | needed_headroom = lowerdev->hard_header_len; |
91572088 | 3551 | |
a985343b MS |
3552 | max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : |
3553 | VXLAN_HEADROOM); | |
f870c1ff AK |
3554 | if (max_mtu < ETH_MIN_MTU) |
3555 | max_mtu = ETH_MIN_MTU; | |
3556 | ||
3557 | if (!changelink && !conf->mtu) | |
3558 | dev->mtu = max_mtu; | |
7e059158 DW |
3559 | } |
3560 | ||
a985343b MS |
3561 | if (dev->mtu > max_mtu) |
3562 | dev->mtu = max_mtu; | |
3563 | ||
b1be00a6 JB |
3564 | if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA) |
3565 | needed_headroom += VXLAN6_HEADROOM; | |
3566 | else | |
3567 | needed_headroom += VXLAN_HEADROOM; | |
3568 | dev->needed_headroom = needed_headroom; | |
3569 | ||
0dfbdf41 | 3570 | memcpy(&vxlan->cfg, conf, sizeof(*conf)); |
a985343b | 3571 | } |
0dfbdf41 | 3572 | |
a985343b | 3573 | static int vxlan_dev_configure(struct net *src_net, struct net_device *dev, |
653ef6a3 GM |
3574 | struct vxlan_config *conf, bool changelink, |
3575 | struct netlink_ext_ack *extack) | |
a985343b MS |
3576 | { |
3577 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
3578 | struct net_device *lowerdev; | |
3579 | int ret; | |
0dfbdf41 | 3580 | |
653ef6a3 | 3581 | ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack); |
a985343b MS |
3582 | if (ret) |
3583 | return ret; | |
8bcdc4f3 | 3584 | |
889ce937 | 3585 | vxlan_config_apply(dev, conf, lowerdev, src_net, changelink); |
0dfbdf41 | 3586 | |
0dfbdf41 TG |
3587 | return 0; |
3588 | } | |
3589 | ||
c80498e3 | 3590 | static int __vxlan_dev_create(struct net *net, struct net_device *dev, |
653ef6a3 GM |
3591 | struct vxlan_config *conf, |
3592 | struct netlink_ext_ack *extack) | |
c80498e3 ND |
3593 | { |
3594 | struct vxlan_net *vn = net_generic(net, vxlan_net_id); | |
3595 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
0ce1822c | 3596 | struct net_device *remote_dev = NULL; |
0241b836 | 3597 | struct vxlan_fdb *f = NULL; |
6db92468 | 3598 | bool unregister = false; |
0ce1822c | 3599 | struct vxlan_rdst *dst; |
c80498e3 ND |
3600 | int err; |
3601 | ||
0ce1822c | 3602 | dst = &vxlan->default_dst; |
653ef6a3 | 3603 | err = vxlan_dev_configure(net, dev, conf, false, extack); |
c80498e3 ND |
3604 | if (err) |
3605 | return err; | |
3606 | ||
3607 | dev->ethtool_ops = &vxlan_ethtool_ops; | |
3608 | ||
3609 | /* create an fdb entry for a valid default destination */ | |
0ce1822c | 3610 | if (!vxlan_addr_any(&dst->remote_ip)) { |
0241b836 | 3611 | err = vxlan_fdb_create(vxlan, all_zeros_mac, |
0ce1822c | 3612 | &dst->remote_ip, |
c80498e3 | 3613 | NUD_REACHABLE | NUD_PERMANENT, |
c80498e3 | 3614 | vxlan->cfg.dst_port, |
0ce1822c TY |
3615 | dst->remote_vni, |
3616 | dst->remote_vni, | |
3617 | dst->remote_ifindex, | |
0241b836 | 3618 | NTF_SELF, &f); |
c80498e3 ND |
3619 | if (err) |
3620 | return err; | |
3621 | } | |
3622 | ||
3623 | err = register_netdevice(dev); | |
0241b836 RP |
3624 | if (err) |
3625 | goto errout; | |
6db92468 | 3626 | unregister = true; |
0241b836 | 3627 | |
0ce1822c TY |
3628 | if (dst->remote_ifindex) { |
3629 | remote_dev = __dev_get_by_index(net, dst->remote_ifindex); | |
3630 | if (!remote_dev) | |
3631 | goto errout; | |
3632 | ||
3633 | err = netdev_upper_dev_link(remote_dev, dev, extack); | |
3634 | if (err) | |
3635 | goto errout; | |
3636 | } | |
3637 | ||
0241b836 | 3638 | err = rtnl_configure_link(dev, NULL); |
6db92468 | 3639 | if (err) |
0ce1822c | 3640 | goto unlink; |
c80498e3 | 3641 | |
61f46fe8 | 3642 | if (f) { |
0ce1822c | 3643 | vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f); |
7c31e54a TY |
3644 | |
3645 | /* notify default fdb entry */ | |
61f46fe8 | 3646 | err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), |
4c59b7d1 | 3647 | RTM_NEWNEIGH, true, extack); |
7c31e54a TY |
3648 | if (err) { |
3649 | vxlan_fdb_destroy(vxlan, f, false, false); | |
0ce1822c TY |
3650 | if (remote_dev) |
3651 | netdev_upper_dev_unlink(remote_dev, dev); | |
7c31e54a TY |
3652 | goto unregister; |
3653 | } | |
61f46fe8 | 3654 | } |
0241b836 | 3655 | |
c80498e3 | 3656 | list_add(&vxlan->next, &vn->vxlan_list); |
0ce1822c TY |
3657 | if (remote_dev) |
3658 | dst->remote_dev = remote_dev; | |
c80498e3 | 3659 | return 0; |
0ce1822c TY |
3660 | unlink: |
3661 | if (remote_dev) | |
3662 | netdev_upper_dev_unlink(remote_dev, dev); | |
0241b836 | 3663 | errout: |
6db92468 PM |
3664 | /* unregister_netdevice() destroys the default FDB entry with deletion |
3665 | * notification. But the addition notification was not sent yet, so | |
3666 | * destroy the entry by hand here. | |
3667 | */ | |
0241b836 | 3668 | if (f) |
7c31e54a TY |
3669 | __vxlan_fdb_free(f); |
3670 | unregister: | |
6db92468 PM |
3671 | if (unregister) |
3672 | unregister_netdevice(dev); | |
0241b836 | 3673 | return err; |
c80498e3 ND |
3674 | } |
3675 | ||
70fb0828 RP |
3676 | /* Set/clear flags based on attribute */ |
3677 | static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[], | |
3678 | int attrtype, unsigned long mask, bool changelink, | |
3679 | bool changelink_supported, | |
3680 | struct netlink_ext_ack *extack) | |
3681 | { | |
3682 | unsigned long flags; | |
3683 | ||
3684 | if (!tb[attrtype]) | |
3685 | return 0; | |
3686 | ||
3687 | if (changelink && !changelink_supported) { | |
3688 | vxlan_flag_attr_error(attrtype, extack); | |
3689 | return -EOPNOTSUPP; | |
3690 | } | |
3691 | ||
3692 | if (vxlan_policy[attrtype].type == NLA_FLAG) | |
3693 | flags = conf->flags | mask; | |
3694 | else if (nla_get_u8(tb[attrtype])) | |
3695 | flags = conf->flags | mask; | |
3696 | else | |
3697 | flags = conf->flags & ~mask; | |
3698 | ||
3699 | conf->flags = flags; | |
3700 | ||
3701 | return 0; | |
3702 | } | |
3703 | ||
8bcdc4f3 RP |
3704 | static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[], |
3705 | struct net_device *dev, struct vxlan_config *conf, | |
70fb0828 | 3706 | bool changelink, struct netlink_ext_ack *extack) |
0dfbdf41 | 3707 | { |
8bcdc4f3 | 3708 | struct vxlan_dev *vxlan = netdev_priv(dev); |
70fb0828 | 3709 | int err = 0; |
0dfbdf41 | 3710 | |
8bcdc4f3 | 3711 | memset(conf, 0, sizeof(*conf)); |
e277de5f | 3712 | |
8bcdc4f3 RP |
3713 | /* if changelink operation, start with old existing cfg */ |
3714 | if (changelink) | |
3715 | memcpy(conf, &vxlan->cfg, sizeof(*conf)); | |
3716 | ||
3717 | if (data[IFLA_VXLAN_ID]) { | |
3718 | __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID])); | |
3719 | ||
70fb0828 RP |
3720 | if (changelink && (vni != conf->vni)) { |
3721 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI"); | |
8bcdc4f3 | 3722 | return -EOPNOTSUPP; |
70fb0828 | 3723 | } |
8bcdc4f3 RP |
3724 | conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID])); |
3725 | } | |
0dfbdf41 TG |
3726 | |
3727 | if (data[IFLA_VXLAN_GROUP]) { | |
70fb0828 RP |
3728 | if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) { |
3729 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group"); | |
ce44a4ae | 3730 | return -EOPNOTSUPP; |
70fb0828 | 3731 | } |
ce44a4ae | 3732 | |
8bcdc4f3 | 3733 | conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]); |
ce44a4ae | 3734 | conf->remote_ip.sa.sa_family = AF_INET; |
0dfbdf41 | 3735 | } else if (data[IFLA_VXLAN_GROUP6]) { |
70fb0828 RP |
3736 | if (!IS_ENABLED(CONFIG_IPV6)) { |
3737 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel"); | |
0dfbdf41 | 3738 | return -EPFNOSUPPORT; |
70fb0828 | 3739 | } |
0dfbdf41 | 3740 | |
70fb0828 RP |
3741 | if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) { |
3742 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group"); | |
ce44a4ae | 3743 | return -EOPNOTSUPP; |
70fb0828 | 3744 | } |
ce44a4ae | 3745 | |
8bcdc4f3 RP |
3746 | conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]); |
3747 | conf->remote_ip.sa.sa_family = AF_INET6; | |
0dfbdf41 TG |
3748 | } |
3749 | ||
3750 | if (data[IFLA_VXLAN_LOCAL]) { | |
70fb0828 RP |
3751 | if (changelink && (conf->saddr.sa.sa_family != AF_INET)) { |
3752 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old"); | |
ce44a4ae | 3753 | return -EOPNOTSUPP; |
70fb0828 | 3754 | } |
ce44a4ae | 3755 | |
8bcdc4f3 RP |
3756 | conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]); |
3757 | conf->saddr.sa.sa_family = AF_INET; | |
0dfbdf41 | 3758 | } else if (data[IFLA_VXLAN_LOCAL6]) { |
70fb0828 RP |
3759 | if (!IS_ENABLED(CONFIG_IPV6)) { |
3760 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel"); | |
0dfbdf41 | 3761 | return -EPFNOSUPPORT; |
70fb0828 | 3762 | } |
0dfbdf41 | 3763 | |
70fb0828 RP |
3764 | if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) { |
3765 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old"); | |
ce44a4ae | 3766 | return -EOPNOTSUPP; |
70fb0828 | 3767 | } |
ce44a4ae | 3768 | |
0dfbdf41 | 3769 | /* TODO: respect scope id */ |
8bcdc4f3 RP |
3770 | conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]); |
3771 | conf->saddr.sa.sa_family = AF_INET6; | |
0dfbdf41 TG |
3772 | } |
3773 | ||
3774 | if (data[IFLA_VXLAN_LINK]) | |
8bcdc4f3 | 3775 | conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]); |
0dfbdf41 | 3776 | |
d342894c | 3777 | if (data[IFLA_VXLAN_TOS]) |
8bcdc4f3 | 3778 | conf->tos = nla_get_u8(data[IFLA_VXLAN_TOS]); |
d342894c | 3779 | |
afb97186 | 3780 | if (data[IFLA_VXLAN_TTL]) |
8bcdc4f3 | 3781 | conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]); |
afb97186 | 3782 | |
72f6d71e | 3783 | if (data[IFLA_VXLAN_TTL_INHERIT]) { |
70fb0828 RP |
3784 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT, |
3785 | VXLAN_F_TTL_INHERIT, changelink, false, | |
3786 | extack); | |
3787 | if (err) | |
3788 | return err; | |
3789 | ||
72f6d71e HL |
3790 | } |
3791 | ||
e7f70af1 | 3792 | if (data[IFLA_VXLAN_LABEL]) |
8bcdc4f3 | 3793 | conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) & |
e7f70af1 DB |
3794 | IPV6_FLOWLABEL_MASK; |
3795 | ||
8bcdc4f3 | 3796 | if (data[IFLA_VXLAN_LEARNING]) { |
70fb0828 RP |
3797 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING, |
3798 | VXLAN_F_LEARN, changelink, true, | |
3799 | extack); | |
3800 | if (err) | |
3801 | return err; | |
8bcdc4f3 RP |
3802 | } else if (!changelink) { |
3803 | /* default to learn on a new device */ | |
3804 | conf->flags |= VXLAN_F_LEARN; | |
3805 | } | |
d342894c | 3806 | |
40051c4d | 3807 | if (data[IFLA_VXLAN_AGEING]) |
8bcdc4f3 | 3808 | conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]); |
d342894c | 3809 | |
8bcdc4f3 | 3810 | if (data[IFLA_VXLAN_PROXY]) { |
70fb0828 RP |
3811 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY, |
3812 | VXLAN_F_PROXY, changelink, false, | |
3813 | extack); | |
3814 | if (err) | |
3815 | return err; | |
8bcdc4f3 | 3816 | } |
e4f67add | 3817 | |
8bcdc4f3 | 3818 | if (data[IFLA_VXLAN_RSC]) { |
70fb0828 RP |
3819 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC, |
3820 | VXLAN_F_RSC, changelink, false, | |
3821 | extack); | |
3822 | if (err) | |
3823 | return err; | |
8bcdc4f3 | 3824 | } |
e4f67add | 3825 | |
8bcdc4f3 | 3826 | if (data[IFLA_VXLAN_L2MISS]) { |
70fb0828 RP |
3827 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS, |
3828 | VXLAN_F_L2MISS, changelink, false, | |
3829 | extack); | |
3830 | if (err) | |
3831 | return err; | |
8bcdc4f3 | 3832 | } |
e4f67add | 3833 | |
8bcdc4f3 | 3834 | if (data[IFLA_VXLAN_L3MISS]) { |
70fb0828 RP |
3835 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS, |
3836 | VXLAN_F_L3MISS, changelink, false, | |
3837 | extack); | |
3838 | if (err) | |
3839 | return err; | |
8bcdc4f3 | 3840 | } |
e4f67add | 3841 | |
8bcdc4f3 | 3842 | if (data[IFLA_VXLAN_LIMIT]) { |
70fb0828 RP |
3843 | if (changelink) { |
3844 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT], | |
3845 | "Cannot change limit"); | |
8bcdc4f3 | 3846 | return -EOPNOTSUPP; |
70fb0828 | 3847 | } |
8bcdc4f3 RP |
3848 | conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]); |
3849 | } | |
d342894c | 3850 | |
8bcdc4f3 | 3851 | if (data[IFLA_VXLAN_COLLECT_METADATA]) { |
70fb0828 RP |
3852 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA, |
3853 | VXLAN_F_COLLECT_METADATA, changelink, false, | |
3854 | extack); | |
3855 | if (err) | |
3856 | return err; | |
8bcdc4f3 | 3857 | } |
f8a9b1bc | 3858 | |
05f47d69 | 3859 | if (data[IFLA_VXLAN_PORT_RANGE]) { |
8bcdc4f3 RP |
3860 | if (!changelink) { |
3861 | const struct ifla_vxlan_port_range *p | |
3862 | = nla_data(data[IFLA_VXLAN_PORT_RANGE]); | |
3863 | conf->port_min = ntohs(p->low); | |
3864 | conf->port_max = ntohs(p->high); | |
3865 | } else { | |
70fb0828 RP |
3866 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE], |
3867 | "Cannot change port range"); | |
8bcdc4f3 RP |
3868 | return -EOPNOTSUPP; |
3869 | } | |
05f47d69 | 3870 | } |
3871 | ||
8bcdc4f3 | 3872 | if (data[IFLA_VXLAN_PORT]) { |
70fb0828 RP |
3873 | if (changelink) { |
3874 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT], | |
3875 | "Cannot change port"); | |
8bcdc4f3 | 3876 | return -EOPNOTSUPP; |
70fb0828 | 3877 | } |
8bcdc4f3 RP |
3878 | conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]); |
3879 | } | |
823aa873 | 3880 | |
8bcdc4f3 | 3881 | if (data[IFLA_VXLAN_UDP_CSUM]) { |
70fb0828 RP |
3882 | if (changelink) { |
3883 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM], | |
3884 | "Cannot change UDP_CSUM flag"); | |
8bcdc4f3 | 3885 | return -EOPNOTSUPP; |
70fb0828 | 3886 | } |
8bcdc4f3 RP |
3887 | if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM])) |
3888 | conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX; | |
3889 | } | |
359a0ea9 | 3890 | |
8bcdc4f3 | 3891 | if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) { |
70fb0828 RP |
3892 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX, |
3893 | VXLAN_F_UDP_ZERO_CSUM6_TX, changelink, | |
3894 | false, extack); | |
3895 | if (err) | |
3896 | return err; | |
8bcdc4f3 | 3897 | } |
359a0ea9 | 3898 | |
8bcdc4f3 | 3899 | if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) { |
70fb0828 RP |
3900 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, |
3901 | VXLAN_F_UDP_ZERO_CSUM6_RX, changelink, | |
3902 | false, extack); | |
3903 | if (err) | |
3904 | return err; | |
8bcdc4f3 | 3905 | } |
359a0ea9 | 3906 | |
8bcdc4f3 | 3907 | if (data[IFLA_VXLAN_REMCSUM_TX]) { |
70fb0828 RP |
3908 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX, |
3909 | VXLAN_F_REMCSUM_TX, changelink, false, | |
3910 | extack); | |
3911 | if (err) | |
3912 | return err; | |
8bcdc4f3 | 3913 | } |
dfd8645e | 3914 | |
8bcdc4f3 | 3915 | if (data[IFLA_VXLAN_REMCSUM_RX]) { |
70fb0828 RP |
3916 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX, |
3917 | VXLAN_F_REMCSUM_RX, changelink, false, | |
3918 | extack); | |
3919 | if (err) | |
3920 | return err; | |
8bcdc4f3 RP |
3921 | } |
3922 | ||
3923 | if (data[IFLA_VXLAN_GBP]) { | |
70fb0828 RP |
3924 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP, |
3925 | VXLAN_F_GBP, changelink, false, extack); | |
3926 | if (err) | |
3927 | return err; | |
8bcdc4f3 RP |
3928 | } |
3929 | ||
3930 | if (data[IFLA_VXLAN_GPE]) { | |
70fb0828 RP |
3931 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE, |
3932 | VXLAN_F_GPE, changelink, false, | |
3933 | extack); | |
3934 | if (err) | |
3935 | return err; | |
8bcdc4f3 RP |
3936 | } |
3937 | ||
3938 | if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) { | |
70fb0828 RP |
3939 | err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL, |
3940 | VXLAN_F_REMCSUM_NOPARTIAL, changelink, | |
3941 | false, extack); | |
3942 | if (err) | |
3943 | return err; | |
8bcdc4f3 RP |
3944 | } |
3945 | ||
3946 | if (tb[IFLA_MTU]) { | |
70fb0828 RP |
3947 | if (changelink) { |
3948 | NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU], | |
3949 | "Cannot change mtu"); | |
8bcdc4f3 | 3950 | return -EOPNOTSUPP; |
70fb0828 | 3951 | } |
8bcdc4f3 RP |
3952 | conf->mtu = nla_get_u32(tb[IFLA_MTU]); |
3953 | } | |
3954 | ||
b4d30697 SB |
3955 | if (data[IFLA_VXLAN_DF]) |
3956 | conf->df = nla_get_u8(data[IFLA_VXLAN_DF]); | |
3957 | ||
8bcdc4f3 RP |
3958 | return 0; |
3959 | } | |
3960 | ||
3961 | static int vxlan_newlink(struct net *src_net, struct net_device *dev, | |
7a3f4a18 MS |
3962 | struct nlattr *tb[], struct nlattr *data[], |
3963 | struct netlink_ext_ack *extack) | |
8bcdc4f3 | 3964 | { |
8bcdc4f3 RP |
3965 | struct vxlan_config conf; |
3966 | int err; | |
3967 | ||
70fb0828 | 3968 | err = vxlan_nl2conf(tb, data, dev, &conf, false, extack); |
8bcdc4f3 RP |
3969 | if (err) |
3970 | return err; | |
3971 | ||
653ef6a3 | 3972 | return __vxlan_dev_create(src_net, dev, &conf, extack); |
8bcdc4f3 | 3973 | } |
dfd8645e | 3974 | |
8bcdc4f3 | 3975 | static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[], |
ad744b22 MS |
3976 | struct nlattr *data[], |
3977 | struct netlink_ext_ack *extack) | |
8bcdc4f3 RP |
3978 | { |
3979 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
8db9427d | 3980 | struct net_device *lowerdev; |
8bcdc4f3 | 3981 | struct vxlan_config conf; |
0ce1822c | 3982 | struct vxlan_rdst *dst; |
8bcdc4f3 RP |
3983 | int err; |
3984 | ||
0ce1822c | 3985 | dst = &vxlan->default_dst; |
70fb0828 | 3986 | err = vxlan_nl2conf(tb, data, dev, &conf, true, extack); |
8bcdc4f3 RP |
3987 | if (err) |
3988 | return err; | |
3511494c | 3989 | |
8db9427d PM |
3990 | err = vxlan_config_validate(vxlan->net, &conf, &lowerdev, |
3991 | vxlan, extack); | |
8bcdc4f3 RP |
3992 | if (err) |
3993 | return err; | |
0ace2ca8 | 3994 | |
c6761cf5 TY |
3995 | if (dst->remote_dev == lowerdev) |
3996 | lowerdev = NULL; | |
3997 | ||
0ce1822c TY |
3998 | err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev, |
3999 | extack); | |
4000 | if (err) | |
4001 | return err; | |
4002 | ||
8bcdc4f3 | 4003 | /* handle default dst entry */ |
038a5a99 | 4004 | if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) { |
fe1e0713 L |
4005 | u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni); |
4006 | ||
4007 | spin_lock_bh(&vxlan->hash_lock[hash_index]); | |
038a5a99 | 4008 | if (!vxlan_addr_any(&conf.remote_ip)) { |
ce5e098f | 4009 | err = vxlan_fdb_update(vxlan, all_zeros_mac, |
038a5a99 | 4010 | &conf.remote_ip, |
8bcdc4f3 | 4011 | NUD_REACHABLE | NUD_PERMANENT, |
ce5e098f | 4012 | NLM_F_APPEND | NLM_F_CREATE, |
8bcdc4f3 | 4013 | vxlan->cfg.dst_port, |
038a5a99 PM |
4014 | conf.vni, conf.vni, |
4015 | conf.remote_ifindex, | |
4c59b7d1 | 4016 | NTF_SELF, true, extack); |
8bcdc4f3 | 4017 | if (err) { |
fe1e0713 | 4018 | spin_unlock_bh(&vxlan->hash_lock[hash_index]); |
0ce1822c TY |
4019 | netdev_adjacent_change_abort(dst->remote_dev, |
4020 | lowerdev, dev); | |
8bcdc4f3 RP |
4021 | return err; |
4022 | } | |
4023 | } | |
1cdc98c2 PM |
4024 | if (!vxlan_addr_any(&dst->remote_ip)) |
4025 | __vxlan_fdb_delete(vxlan, all_zeros_mac, | |
4026 | dst->remote_ip, | |
4027 | vxlan->cfg.dst_port, | |
4028 | dst->remote_vni, | |
4029 | dst->remote_vni, | |
4030 | dst->remote_ifindex, | |
4031 | true); | |
fe1e0713 | 4032 | spin_unlock_bh(&vxlan->hash_lock[hash_index]); |
8bcdc4f3 | 4033 | } |
ce577668 | 4034 | |
038a5a99 PM |
4035 | if (conf.age_interval != vxlan->cfg.age_interval) |
4036 | mod_timer(&vxlan->age_timer, jiffies); | |
4037 | ||
0ce1822c | 4038 | netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev); |
c6761cf5 | 4039 | if (lowerdev && lowerdev != dst->remote_dev) { |
0ce1822c | 4040 | dst->remote_dev = lowerdev; |
c6761cf5 TY |
4041 | netdev_update_lockdep_key(lowerdev); |
4042 | } | |
038a5a99 | 4043 | vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true); |
8bcdc4f3 | 4044 | return 0; |
d342894c | 4045 | } |
4046 | ||
4047 | static void vxlan_dellink(struct net_device *dev, struct list_head *head) | |
4048 | { | |
4049 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
4050 | ||
8b3f9337 RP |
4051 | vxlan_flush(vxlan, true); |
4052 | ||
553675fb | 4053 | list_del(&vxlan->next); |
d342894c | 4054 | unregister_netdevice_queue(dev, head); |
0ce1822c TY |
4055 | if (vxlan->default_dst.remote_dev) |
4056 | netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev); | |
d342894c | 4057 | } |
4058 | ||
4059 | static size_t vxlan_get_size(const struct net_device *dev) | |
4060 | { | |
4061 | ||
4062 | return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */ | |
e4c7ed41 | 4063 | nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */ |
d342894c | 4064 | nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */ |
e4c7ed41 | 4065 | nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */ |
d342894c | 4066 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */ |
8fd78069 | 4067 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL_INHERIT */ |
d342894c | 4068 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */ |
b4d30697 | 4069 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_DF */ |
e7f70af1 | 4070 | nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */ |
d342894c | 4071 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */ |
e4f67add DS |
4072 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */ |
4073 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */ | |
4074 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */ | |
4075 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */ | |
da8b43c0 | 4076 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */ |
d342894c | 4077 | nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */ |
4078 | nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */ | |
05f47d69 | 4079 | nla_total_size(sizeof(struct ifla_vxlan_port_range)) + |
359a0ea9 TH |
4080 | nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */ |
4081 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */ | |
4082 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */ | |
4083 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */ | |
dfd8645e TH |
4084 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */ |
4085 | nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */ | |
d342894c | 4086 | 0; |
4087 | } | |
4088 | ||
4089 | static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev) | |
4090 | { | |
4091 | const struct vxlan_dev *vxlan = netdev_priv(dev); | |
c7995c43 | 4092 | const struct vxlan_rdst *dst = &vxlan->default_dst; |
05f47d69 | 4093 | struct ifla_vxlan_port_range ports = { |
0dfbdf41 TG |
4094 | .low = htons(vxlan->cfg.port_min), |
4095 | .high = htons(vxlan->cfg.port_max), | |
05f47d69 | 4096 | }; |
d342894c | 4097 | |
54bfd872 | 4098 | if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni))) |
d342894c | 4099 | goto nla_put_failure; |
4100 | ||
e4c7ed41 CW |
4101 | if (!vxlan_addr_any(&dst->remote_ip)) { |
4102 | if (dst->remote_ip.sa.sa_family == AF_INET) { | |
930345ea JB |
4103 | if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP, |
4104 | dst->remote_ip.sin.sin_addr.s_addr)) | |
e4c7ed41 CW |
4105 | goto nla_put_failure; |
4106 | #if IS_ENABLED(CONFIG_IPV6) | |
4107 | } else { | |
930345ea JB |
4108 | if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6, |
4109 | &dst->remote_ip.sin6.sin6_addr)) | |
e4c7ed41 CW |
4110 | goto nla_put_failure; |
4111 | #endif | |
4112 | } | |
4113 | } | |
d342894c | 4114 | |
c7995c43 | 4115 | if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex)) |
d342894c | 4116 | goto nla_put_failure; |
4117 | ||
0dfbdf41 TG |
4118 | if (!vxlan_addr_any(&vxlan->cfg.saddr)) { |
4119 | if (vxlan->cfg.saddr.sa.sa_family == AF_INET) { | |
930345ea | 4120 | if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL, |
0dfbdf41 | 4121 | vxlan->cfg.saddr.sin.sin_addr.s_addr)) |
e4c7ed41 CW |
4122 | goto nla_put_failure; |
4123 | #if IS_ENABLED(CONFIG_IPV6) | |
4124 | } else { | |
930345ea | 4125 | if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6, |
0dfbdf41 | 4126 | &vxlan->cfg.saddr.sin6.sin6_addr)) |
e4c7ed41 CW |
4127 | goto nla_put_failure; |
4128 | #endif | |
4129 | } | |
4130 | } | |
d342894c | 4131 | |
0dfbdf41 | 4132 | if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) || |
8fd78069 HL |
4133 | nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT, |
4134 | !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) || | |
0dfbdf41 | 4135 | nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) || |
b4d30697 | 4136 | nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) || |
e7f70af1 | 4137 | nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) || |
e4f67add | 4138 | nla_put_u8(skb, IFLA_VXLAN_LEARNING, |
98c81476 | 4139 | !!(vxlan->cfg.flags & VXLAN_F_LEARN)) || |
e4f67add | 4140 | nla_put_u8(skb, IFLA_VXLAN_PROXY, |
98c81476 | 4141 | !!(vxlan->cfg.flags & VXLAN_F_PROXY)) || |
dc5321d7 MS |
4142 | nla_put_u8(skb, IFLA_VXLAN_RSC, |
4143 | !!(vxlan->cfg.flags & VXLAN_F_RSC)) || | |
e4f67add | 4144 | nla_put_u8(skb, IFLA_VXLAN_L2MISS, |
98c81476 | 4145 | !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) || |
e4f67add | 4146 | nla_put_u8(skb, IFLA_VXLAN_L3MISS, |
98c81476 | 4147 | !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) || |
da8b43c0 | 4148 | nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA, |
dc5321d7 | 4149 | !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) || |
0dfbdf41 TG |
4150 | nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) || |
4151 | nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) || | |
4152 | nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) || | |
359a0ea9 | 4153 | nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM, |
98c81476 | 4154 | !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) || |
359a0ea9 | 4155 | nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX, |
98c81476 | 4156 | !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) || |
359a0ea9 | 4157 | nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, |
98c81476 | 4158 | !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) || |
dfd8645e | 4159 | nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX, |
98c81476 | 4160 | !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) || |
dfd8645e | 4161 | nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX, |
98c81476 | 4162 | !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX))) |
d342894c | 4163 | goto nla_put_failure; |
4164 | ||
05f47d69 | 4165 | if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports)) |
4166 | goto nla_put_failure; | |
4167 | ||
dc5321d7 | 4168 | if (vxlan->cfg.flags & VXLAN_F_GBP && |
3511494c TG |
4169 | nla_put_flag(skb, IFLA_VXLAN_GBP)) |
4170 | goto nla_put_failure; | |
4171 | ||
dc5321d7 | 4172 | if (vxlan->cfg.flags & VXLAN_F_GPE && |
e1e5314d JB |
4173 | nla_put_flag(skb, IFLA_VXLAN_GPE)) |
4174 | goto nla_put_failure; | |
4175 | ||
dc5321d7 | 4176 | if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL && |
0ace2ca8 TH |
4177 | nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL)) |
4178 | goto nla_put_failure; | |
4179 | ||
d342894c | 4180 | return 0; |
4181 | ||
4182 | nla_put_failure: | |
4183 | return -EMSGSIZE; | |
4184 | } | |
4185 | ||
1728d4fa ND |
4186 | static struct net *vxlan_get_link_net(const struct net_device *dev) |
4187 | { | |
4188 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
4189 | ||
4190 | return vxlan->net; | |
4191 | } | |
4192 | ||
d342894c | 4193 | static struct rtnl_link_ops vxlan_link_ops __read_mostly = { |
4194 | .kind = "vxlan", | |
4195 | .maxtype = IFLA_VXLAN_MAX, | |
4196 | .policy = vxlan_policy, | |
4197 | .priv_size = sizeof(struct vxlan_dev), | |
4198 | .setup = vxlan_setup, | |
4199 | .validate = vxlan_validate, | |
4200 | .newlink = vxlan_newlink, | |
8bcdc4f3 | 4201 | .changelink = vxlan_changelink, |
d342894c | 4202 | .dellink = vxlan_dellink, |
4203 | .get_size = vxlan_get_size, | |
4204 | .fill_info = vxlan_fill_info, | |
1728d4fa | 4205 | .get_link_net = vxlan_get_link_net, |
d342894c | 4206 | }; |
4207 | ||
cf5da330 ND |
4208 | struct net_device *vxlan_dev_create(struct net *net, const char *name, |
4209 | u8 name_assign_type, | |
4210 | struct vxlan_config *conf) | |
4211 | { | |
4212 | struct nlattr *tb[IFLA_MAX + 1]; | |
4213 | struct net_device *dev; | |
4214 | int err; | |
4215 | ||
4216 | memset(&tb, 0, sizeof(tb)); | |
4217 | ||
4218 | dev = rtnl_create_link(net, name, name_assign_type, | |
d0522f1c | 4219 | &vxlan_link_ops, tb, NULL); |
cf5da330 ND |
4220 | if (IS_ERR(dev)) |
4221 | return dev; | |
4222 | ||
653ef6a3 | 4223 | err = __vxlan_dev_create(net, dev, conf, NULL); |
cf5da330 ND |
4224 | if (err < 0) { |
4225 | free_netdev(dev); | |
4226 | return ERR_PTR(err); | |
4227 | } | |
4228 | ||
4229 | err = rtnl_configure_link(dev, NULL); | |
4230 | if (err < 0) { | |
4231 | LIST_HEAD(list_kill); | |
4232 | ||
4233 | vxlan_dellink(dev, &list_kill); | |
4234 | unregister_netdevice_many(&list_kill); | |
4235 | return ERR_PTR(err); | |
4236 | } | |
4237 | ||
4238 | return dev; | |
4239 | } | |
4240 | EXPORT_SYMBOL_GPL(vxlan_dev_create); | |
4241 | ||
acaf4e70 DB |
4242 | static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn, |
4243 | struct net_device *dev) | |
4244 | { | |
4245 | struct vxlan_dev *vxlan, *next; | |
4246 | LIST_HEAD(list_kill); | |
4247 | ||
4248 | list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) { | |
4249 | struct vxlan_rdst *dst = &vxlan->default_dst; | |
4250 | ||
4251 | /* In case we created vxlan device with carrier | |
4252 | * and we loose the carrier due to module unload | |
4253 | * we also need to remove vxlan device. In other | |
4254 | * cases, it's not necessary and remote_ifindex | |
4255 | * is 0 here, so no matches. | |
4256 | */ | |
4257 | if (dst->remote_ifindex == dev->ifindex) | |
4258 | vxlan_dellink(vxlan->dev, &list_kill); | |
4259 | } | |
4260 | ||
4261 | unregister_netdevice_many(&list_kill); | |
4262 | } | |
4263 | ||
b7aade15 HFS |
4264 | static int vxlan_netdevice_event(struct notifier_block *unused, |
4265 | unsigned long event, void *ptr) | |
acaf4e70 DB |
4266 | { |
4267 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); | |
783c1463 | 4268 | struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id); |
acaf4e70 | 4269 | |
04584957 SD |
4270 | if (event == NETDEV_UNREGISTER) { |
4271 | vxlan_offload_rx_ports(dev, false); | |
acaf4e70 | 4272 | vxlan_handle_lowerdev_unregister(vn, dev); |
04584957 SD |
4273 | } else if (event == NETDEV_REGISTER) { |
4274 | vxlan_offload_rx_ports(dev, true); | |
4275 | } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO || | |
4276 | event == NETDEV_UDP_TUNNEL_DROP_INFO) { | |
2d2b13fc | 4277 | vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO); |
04584957 | 4278 | } |
acaf4e70 DB |
4279 | |
4280 | return NOTIFY_DONE; | |
4281 | } | |
4282 | ||
4283 | static struct notifier_block vxlan_notifier_block __read_mostly = { | |
b7aade15 | 4284 | .notifier_call = vxlan_netdevice_event, |
acaf4e70 DB |
4285 | }; |
4286 | ||
0efe1173 PM |
4287 | static void |
4288 | vxlan_fdb_offloaded_set(struct net_device *dev, | |
4289 | struct switchdev_notifier_vxlan_fdb_info *fdb_info) | |
4290 | { | |
4291 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
4292 | struct vxlan_rdst *rdst; | |
4293 | struct vxlan_fdb *f; | |
fe1e0713 L |
4294 | u32 hash_index; |
4295 | ||
4296 | hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni); | |
0efe1173 | 4297 | |
fe1e0713 | 4298 | spin_lock_bh(&vxlan->hash_lock[hash_index]); |
0efe1173 PM |
4299 | |
4300 | f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni); | |
4301 | if (!f) | |
4302 | goto out; | |
4303 | ||
4304 | rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip, | |
4305 | fdb_info->remote_port, | |
4306 | fdb_info->remote_vni, | |
4307 | fdb_info->remote_ifindex); | |
4308 | if (!rdst) | |
4309 | goto out; | |
4310 | ||
4311 | rdst->offloaded = fdb_info->offloaded; | |
4312 | ||
4313 | out: | |
fe1e0713 | 4314 | spin_unlock_bh(&vxlan->hash_lock[hash_index]); |
0efe1173 PM |
4315 | } |
4316 | ||
5728ae0d PM |
4317 | static int |
4318 | vxlan_fdb_external_learn_add(struct net_device *dev, | |
4319 | struct switchdev_notifier_vxlan_fdb_info *fdb_info) | |
4320 | { | |
4321 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
4c59b7d1 | 4322 | struct netlink_ext_ack *extack; |
fe1e0713 | 4323 | u32 hash_index; |
5728ae0d PM |
4324 | int err; |
4325 | ||
fe1e0713 | 4326 | hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni); |
4c59b7d1 PM |
4327 | extack = switchdev_notifier_info_to_extack(&fdb_info->info); |
4328 | ||
fe1e0713 | 4329 | spin_lock_bh(&vxlan->hash_lock[hash_index]); |
5728ae0d PM |
4330 | err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip, |
4331 | NUD_REACHABLE, | |
4332 | NLM_F_CREATE | NLM_F_REPLACE, | |
4333 | fdb_info->remote_port, | |
4334 | fdb_info->vni, | |
4335 | fdb_info->remote_vni, | |
4336 | fdb_info->remote_ifindex, | |
4337 | NTF_USE | NTF_SELF | NTF_EXT_LEARNED, | |
4c59b7d1 | 4338 | false, extack); |
fe1e0713 | 4339 | spin_unlock_bh(&vxlan->hash_lock[hash_index]); |
5728ae0d PM |
4340 | |
4341 | return err; | |
4342 | } | |
4343 | ||
4344 | static int | |
4345 | vxlan_fdb_external_learn_del(struct net_device *dev, | |
4346 | struct switchdev_notifier_vxlan_fdb_info *fdb_info) | |
4347 | { | |
4348 | struct vxlan_dev *vxlan = netdev_priv(dev); | |
4349 | struct vxlan_fdb *f; | |
fe1e0713 | 4350 | u32 hash_index; |
5728ae0d PM |
4351 | int err = 0; |
4352 | ||
fe1e0713 L |
4353 | hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni); |
4354 | spin_lock_bh(&vxlan->hash_lock[hash_index]); | |
5728ae0d PM |
4355 | |
4356 | f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni); | |
4357 | if (!f) | |
4358 | err = -ENOENT; | |
4359 | else if (f->flags & NTF_EXT_LEARNED) | |
4360 | err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr, | |
4361 | fdb_info->remote_ip, | |
4362 | fdb_info->remote_port, | |
4363 | fdb_info->vni, | |
4364 | fdb_info->remote_vni, | |
4365 | fdb_info->remote_ifindex, | |
4366 | false); | |
4367 | ||
fe1e0713 | 4368 | spin_unlock_bh(&vxlan->hash_lock[hash_index]); |
5728ae0d PM |
4369 | |
4370 | return err; | |
4371 | } | |
4372 | ||
0efe1173 PM |
4373 | static int vxlan_switchdev_event(struct notifier_block *unused, |
4374 | unsigned long event, void *ptr) | |
4375 | { | |
4376 | struct net_device *dev = switchdev_notifier_info_to_dev(ptr); | |
5728ae0d PM |
4377 | struct switchdev_notifier_vxlan_fdb_info *fdb_info; |
4378 | int err = 0; | |
0efe1173 PM |
4379 | |
4380 | switch (event) { | |
4381 | case SWITCHDEV_VXLAN_FDB_OFFLOADED: | |
4382 | vxlan_fdb_offloaded_set(dev, ptr); | |
4383 | break; | |
5728ae0d PM |
4384 | case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE: |
4385 | fdb_info = ptr; | |
4386 | err = vxlan_fdb_external_learn_add(dev, fdb_info); | |
4387 | if (err) { | |
4388 | err = notifier_from_errno(err); | |
4389 | break; | |
4390 | } | |
4391 | fdb_info->offloaded = true; | |
4392 | vxlan_fdb_offloaded_set(dev, fdb_info); | |
4393 | break; | |
4394 | case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE: | |
4395 | fdb_info = ptr; | |
4396 | err = vxlan_fdb_external_learn_del(dev, fdb_info); | |
4397 | if (err) { | |
4398 | err = notifier_from_errno(err); | |
4399 | break; | |
4400 | } | |
4401 | fdb_info->offloaded = false; | |
4402 | vxlan_fdb_offloaded_set(dev, fdb_info); | |
4403 | break; | |
0efe1173 PM |
4404 | } |
4405 | ||
5728ae0d | 4406 | return err; |
0efe1173 PM |
4407 | } |
4408 | ||
4409 | static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = { | |
4410 | .notifier_call = vxlan_switchdev_event, | |
4411 | }; | |
4412 | ||
d342894c | 4413 | static __net_init int vxlan_init_net(struct net *net) |
4414 | { | |
4415 | struct vxlan_net *vn = net_generic(net, vxlan_net_id); | |
31fec5aa | 4416 | unsigned int h; |
d342894c | 4417 | |
553675fb | 4418 | INIT_LIST_HEAD(&vn->vxlan_list); |
1c51a915 | 4419 | spin_lock_init(&vn->sock_lock); |
d342894c | 4420 | |
553675fb | 4421 | for (h = 0; h < PORT_HASH_SIZE; ++h) |
4422 | INIT_HLIST_HEAD(&vn->sock_list[h]); | |
d342894c | 4423 | |
4424 | return 0; | |
4425 | } | |
4426 | ||
57b61127 | 4427 | static void vxlan_destroy_tunnels(struct net *net, struct list_head *head) |
f01ec1c0 ND |
4428 | { |
4429 | struct vxlan_net *vn = net_generic(net, vxlan_net_id); | |
4430 | struct vxlan_dev *vxlan, *next; | |
4431 | struct net_device *dev, *aux; | |
0e4ec5ac | 4432 | unsigned int h; |
f01ec1c0 | 4433 | |
f01ec1c0 ND |
4434 | for_each_netdev_safe(net, dev, aux) |
4435 | if (dev->rtnl_link_ops == &vxlan_link_ops) | |
57b61127 | 4436 | unregister_netdevice_queue(dev, head); |
f01ec1c0 ND |
4437 | |
4438 | list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) { | |
4439 | /* If vxlan->dev is in the same netns, it has already been added | |
4440 | * to the list by the previous loop. | |
4441 | */ | |
cc4807bb | 4442 | if (!net_eq(dev_net(vxlan->dev), net)) |
57b61127 | 4443 | unregister_netdevice_queue(vxlan->dev, head); |
f01ec1c0 ND |
4444 | } |
4445 | ||
0e4ec5ac VA |
4446 | for (h = 0; h < PORT_HASH_SIZE; ++h) |
4447 | WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h])); | |
f01ec1c0 ND |
4448 | } |
4449 | ||
57b61127 HY |
4450 | static void __net_exit vxlan_exit_batch_net(struct list_head *net_list) |
4451 | { | |
4452 | struct net *net; | |
4453 | LIST_HEAD(list); | |
4454 | ||
4455 | rtnl_lock(); | |
4456 | list_for_each_entry(net, net_list, exit_list) | |
4457 | vxlan_destroy_tunnels(net, &list); | |
4458 | ||
4459 | unregister_netdevice_many(&list); | |
4460 | rtnl_unlock(); | |
4461 | } | |
4462 | ||
d342894c | 4463 | static struct pernet_operations vxlan_net_ops = { |
4464 | .init = vxlan_init_net, | |
57b61127 | 4465 | .exit_batch = vxlan_exit_batch_net, |
d342894c | 4466 | .id = &vxlan_net_id, |
4467 | .size = sizeof(struct vxlan_net), | |
4468 | }; | |
4469 | ||
4470 | static int __init vxlan_init_module(void) | |
4471 | { | |
4472 | int rc; | |
4473 | ||
4474 | get_random_bytes(&vxlan_salt, sizeof(vxlan_salt)); | |
4475 | ||
783c1463 | 4476 | rc = register_pernet_subsys(&vxlan_net_ops); |
d342894c | 4477 | if (rc) |
4478 | goto out1; | |
4479 | ||
acaf4e70 | 4480 | rc = register_netdevice_notifier(&vxlan_notifier_block); |
d342894c | 4481 | if (rc) |
4482 | goto out2; | |
4483 | ||
0efe1173 | 4484 | rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block); |
acaf4e70 DB |
4485 | if (rc) |
4486 | goto out3; | |
d342894c | 4487 | |
0efe1173 PM |
4488 | rc = rtnl_link_register(&vxlan_link_ops); |
4489 | if (rc) | |
4490 | goto out4; | |
4491 | ||
acaf4e70 | 4492 | return 0; |
0efe1173 PM |
4493 | out4: |
4494 | unregister_switchdev_notifier(&vxlan_switchdev_notifier_block); | |
acaf4e70 DB |
4495 | out3: |
4496 | unregister_netdevice_notifier(&vxlan_notifier_block); | |
d342894c | 4497 | out2: |
783c1463 | 4498 | unregister_pernet_subsys(&vxlan_net_ops); |
d342894c | 4499 | out1: |
4500 | return rc; | |
4501 | } | |
7332a13b | 4502 | late_initcall(vxlan_init_module); |
d342894c | 4503 | |
4504 | static void __exit vxlan_cleanup_module(void) | |
4505 | { | |
b7153984 | 4506 | rtnl_link_unregister(&vxlan_link_ops); |
0efe1173 | 4507 | unregister_switchdev_notifier(&vxlan_switchdev_notifier_block); |
acaf4e70 | 4508 | unregister_netdevice_notifier(&vxlan_notifier_block); |
783c1463 DB |
4509 | unregister_pernet_subsys(&vxlan_net_ops); |
4510 | /* rcu_barrier() is called by netns */ | |
d342894c | 4511 | } |
4512 | module_exit(vxlan_cleanup_module); | |
4513 | ||
4514 | MODULE_LICENSE("GPL"); | |
4515 | MODULE_VERSION(VXLAN_VERSION); | |
3b8df3c6 | 4516 | MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>"); |
ead5139a | 4517 | MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic"); |
d342894c | 4518 | MODULE_ALIAS_RTNL_LINK("vxlan"); |