xref: /linux/net/ipv4/ip_tunnel_core.c (revision 132db93572821ec2fdf81e354cc40f558faf7e4f)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2013 Nicira, Inc.
4  */
5 
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 
8 #include <linux/types.h>
9 #include <linux/kernel.h>
10 #include <linux/skbuff.h>
11 #include <linux/netdevice.h>
12 #include <linux/in.h>
13 #include <linux/if_arp.h>
14 #include <linux/init.h>
15 #include <linux/in6.h>
16 #include <linux/inetdevice.h>
17 #include <linux/netfilter_ipv4.h>
18 #include <linux/etherdevice.h>
19 #include <linux/if_ether.h>
20 #include <linux/if_vlan.h>
21 #include <linux/static_key.h>
22 
23 #include <net/ip.h>
24 #include <net/icmp.h>
25 #include <net/protocol.h>
26 #include <net/ip_tunnels.h>
27 #include <net/ip6_tunnel.h>
28 #include <net/arp.h>
29 #include <net/checksum.h>
30 #include <net/dsfield.h>
31 #include <net/inet_ecn.h>
32 #include <net/xfrm.h>
33 #include <net/net_namespace.h>
34 #include <net/netns/generic.h>
35 #include <net/rtnetlink.h>
36 #include <net/dst_metadata.h>
37 #include <net/geneve.h>
38 #include <net/vxlan.h>
39 #include <net/erspan.h>
40 
41 const struct ip_tunnel_encap_ops __rcu *
42 		iptun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
43 EXPORT_SYMBOL(iptun_encaps);
44 
45 const struct ip6_tnl_encap_ops __rcu *
46 		ip6tun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
47 EXPORT_SYMBOL(ip6tun_encaps);
48 
49 void iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb,
50 		   __be32 src, __be32 dst, __u8 proto,
51 		   __u8 tos, __u8 ttl, __be16 df, bool xnet)
52 {
53 	int pkt_len = skb->len - skb_inner_network_offset(skb);
54 	struct net *net = dev_net(rt->dst.dev);
55 	struct net_device *dev = skb->dev;
56 	struct iphdr *iph;
57 	int err;
58 
59 	skb_scrub_packet(skb, xnet);
60 
61 	skb_clear_hash_if_not_l4(skb);
62 	skb_dst_set(skb, &rt->dst);
63 	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
64 
65 	/* Push down and install the IP header. */
66 	skb_push(skb, sizeof(struct iphdr));
67 	skb_reset_network_header(skb);
68 
69 	iph = ip_hdr(skb);
70 
71 	iph->version	=	4;
72 	iph->ihl	=	sizeof(struct iphdr) >> 2;
73 	iph->frag_off	=	ip_mtu_locked(&rt->dst) ? 0 : df;
74 	iph->protocol	=	proto;
75 	iph->tos	=	tos;
76 	iph->daddr	=	dst;
77 	iph->saddr	=	src;
78 	iph->ttl	=	ttl;
79 	__ip_select_ident(net, iph, skb_shinfo(skb)->gso_segs ?: 1);
80 
81 	err = ip_local_out(net, sk, skb);
82 
83 	if (dev) {
84 		if (unlikely(net_xmit_eval(err)))
85 			pkt_len = 0;
86 		iptunnel_xmit_stats(dev, pkt_len);
87 	}
88 }
89 EXPORT_SYMBOL_GPL(iptunnel_xmit);
90 
91 int __iptunnel_pull_header(struct sk_buff *skb, int hdr_len,
92 			   __be16 inner_proto, bool raw_proto, bool xnet)
93 {
94 	if (unlikely(!pskb_may_pull(skb, hdr_len)))
95 		return -ENOMEM;
96 
97 	skb_pull_rcsum(skb, hdr_len);
98 
99 	if (!raw_proto && inner_proto == htons(ETH_P_TEB)) {
100 		struct ethhdr *eh;
101 
102 		if (unlikely(!pskb_may_pull(skb, ETH_HLEN)))
103 			return -ENOMEM;
104 
105 		eh = (struct ethhdr *)skb->data;
106 		if (likely(eth_proto_is_802_3(eh->h_proto)))
107 			skb->protocol = eh->h_proto;
108 		else
109 			skb->protocol = htons(ETH_P_802_2);
110 
111 	} else {
112 		skb->protocol = inner_proto;
113 	}
114 
115 	skb_clear_hash_if_not_l4(skb);
116 	__vlan_hwaccel_clear_tag(skb);
117 	skb_set_queue_mapping(skb, 0);
118 	skb_scrub_packet(skb, xnet);
119 
120 	return iptunnel_pull_offloads(skb);
121 }
122 EXPORT_SYMBOL_GPL(__iptunnel_pull_header);
123 
124 struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md,
125 					     gfp_t flags)
126 {
127 	struct metadata_dst *res;
128 	struct ip_tunnel_info *dst, *src;
129 
130 	if (!md || md->type != METADATA_IP_TUNNEL ||
131 	    md->u.tun_info.mode & IP_TUNNEL_INFO_TX)
132 		return NULL;
133 
134 	src = &md->u.tun_info;
135 	res = metadata_dst_alloc(src->options_len, METADATA_IP_TUNNEL, flags);
136 	if (!res)
137 		return NULL;
138 
139 	dst = &res->u.tun_info;
140 	dst->key.tun_id = src->key.tun_id;
141 	if (src->mode & IP_TUNNEL_INFO_IPV6)
142 		memcpy(&dst->key.u.ipv6.dst, &src->key.u.ipv6.src,
143 		       sizeof(struct in6_addr));
144 	else
145 		dst->key.u.ipv4.dst = src->key.u.ipv4.src;
146 	dst->key.tun_flags = src->key.tun_flags;
147 	dst->mode = src->mode | IP_TUNNEL_INFO_TX;
148 	ip_tunnel_info_opts_set(dst, ip_tunnel_info_opts(src),
149 				src->options_len, 0);
150 
151 	return res;
152 }
153 EXPORT_SYMBOL_GPL(iptunnel_metadata_reply);
154 
155 int iptunnel_handle_offloads(struct sk_buff *skb,
156 			     int gso_type_mask)
157 {
158 	int err;
159 
160 	if (likely(!skb->encapsulation)) {
161 		skb_reset_inner_headers(skb);
162 		skb->encapsulation = 1;
163 	}
164 
165 	if (skb_is_gso(skb)) {
166 		err = skb_header_unclone(skb, GFP_ATOMIC);
167 		if (unlikely(err))
168 			return err;
169 		skb_shinfo(skb)->gso_type |= gso_type_mask;
170 		return 0;
171 	}
172 
173 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
174 		skb->ip_summed = CHECKSUM_NONE;
175 		/* We clear encapsulation here to prevent badly-written
176 		 * drivers potentially deciding to offload an inner checksum
177 		 * if we set CHECKSUM_PARTIAL on the outer header.
178 		 * This should go away when the drivers are all fixed.
179 		 */
180 		skb->encapsulation = 0;
181 	}
182 
183 	return 0;
184 }
185 EXPORT_SYMBOL_GPL(iptunnel_handle_offloads);
186 
187 /* Often modified stats are per cpu, other are shared (netdev->stats) */
188 void ip_tunnel_get_stats64(struct net_device *dev,
189 			   struct rtnl_link_stats64 *tot)
190 {
191 	int i;
192 
193 	netdev_stats_to_stats64(tot, &dev->stats);
194 
195 	for_each_possible_cpu(i) {
196 		const struct pcpu_sw_netstats *tstats =
197 						   per_cpu_ptr(dev->tstats, i);
198 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
199 		unsigned int start;
200 
201 		do {
202 			start = u64_stats_fetch_begin_irq(&tstats->syncp);
203 			rx_packets = tstats->rx_packets;
204 			tx_packets = tstats->tx_packets;
205 			rx_bytes = tstats->rx_bytes;
206 			tx_bytes = tstats->tx_bytes;
207 		} while (u64_stats_fetch_retry_irq(&tstats->syncp, start));
208 
209 		tot->rx_packets += rx_packets;
210 		tot->tx_packets += tx_packets;
211 		tot->rx_bytes   += rx_bytes;
212 		tot->tx_bytes   += tx_bytes;
213 	}
214 }
215 EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64);
216 
217 static const struct nla_policy ip_tun_policy[LWTUNNEL_IP_MAX + 1] = {
218 	[LWTUNNEL_IP_UNSPEC]	= { .strict_start_type = LWTUNNEL_IP_OPTS },
219 	[LWTUNNEL_IP_ID]	= { .type = NLA_U64 },
220 	[LWTUNNEL_IP_DST]	= { .type = NLA_U32 },
221 	[LWTUNNEL_IP_SRC]	= { .type = NLA_U32 },
222 	[LWTUNNEL_IP_TTL]	= { .type = NLA_U8 },
223 	[LWTUNNEL_IP_TOS]	= { .type = NLA_U8 },
224 	[LWTUNNEL_IP_FLAGS]	= { .type = NLA_U16 },
225 	[LWTUNNEL_IP_OPTS]	= { .type = NLA_NESTED },
226 };
227 
228 static const struct nla_policy ip_opts_policy[LWTUNNEL_IP_OPTS_MAX + 1] = {
229 	[LWTUNNEL_IP_OPTS_GENEVE]	= { .type = NLA_NESTED },
230 	[LWTUNNEL_IP_OPTS_VXLAN]	= { .type = NLA_NESTED },
231 	[LWTUNNEL_IP_OPTS_ERSPAN]	= { .type = NLA_NESTED },
232 };
233 
234 static const struct nla_policy
235 geneve_opt_policy[LWTUNNEL_IP_OPT_GENEVE_MAX + 1] = {
236 	[LWTUNNEL_IP_OPT_GENEVE_CLASS]	= { .type = NLA_U16 },
237 	[LWTUNNEL_IP_OPT_GENEVE_TYPE]	= { .type = NLA_U8 },
238 	[LWTUNNEL_IP_OPT_GENEVE_DATA]	= { .type = NLA_BINARY, .len = 128 },
239 };
240 
241 static const struct nla_policy
242 vxlan_opt_policy[LWTUNNEL_IP_OPT_VXLAN_MAX + 1] = {
243 	[LWTUNNEL_IP_OPT_VXLAN_GBP]	= { .type = NLA_U32 },
244 };
245 
246 static const struct nla_policy
247 erspan_opt_policy[LWTUNNEL_IP_OPT_ERSPAN_MAX + 1] = {
248 	[LWTUNNEL_IP_OPT_ERSPAN_VER]	= { .type = NLA_U8 },
249 	[LWTUNNEL_IP_OPT_ERSPAN_INDEX]	= { .type = NLA_U32 },
250 	[LWTUNNEL_IP_OPT_ERSPAN_DIR]	= { .type = NLA_U8 },
251 	[LWTUNNEL_IP_OPT_ERSPAN_HWID]	= { .type = NLA_U8 },
252 };
253 
254 static int ip_tun_parse_opts_geneve(struct nlattr *attr,
255 				    struct ip_tunnel_info *info, int opts_len,
256 				    struct netlink_ext_ack *extack)
257 {
258 	struct nlattr *tb[LWTUNNEL_IP_OPT_GENEVE_MAX + 1];
259 	int data_len, err;
260 
261 	err = nla_parse_nested(tb, LWTUNNEL_IP_OPT_GENEVE_MAX, attr,
262 			       geneve_opt_policy, extack);
263 	if (err)
264 		return err;
265 
266 	if (!tb[LWTUNNEL_IP_OPT_GENEVE_CLASS] ||
267 	    !tb[LWTUNNEL_IP_OPT_GENEVE_TYPE] ||
268 	    !tb[LWTUNNEL_IP_OPT_GENEVE_DATA])
269 		return -EINVAL;
270 
271 	attr = tb[LWTUNNEL_IP_OPT_GENEVE_DATA];
272 	data_len = nla_len(attr);
273 	if (data_len % 4)
274 		return -EINVAL;
275 
276 	if (info) {
277 		struct geneve_opt *opt = ip_tunnel_info_opts(info) + opts_len;
278 
279 		memcpy(opt->opt_data, nla_data(attr), data_len);
280 		opt->length = data_len / 4;
281 		attr = tb[LWTUNNEL_IP_OPT_GENEVE_CLASS];
282 		opt->opt_class = nla_get_be16(attr);
283 		attr = tb[LWTUNNEL_IP_OPT_GENEVE_TYPE];
284 		opt->type = nla_get_u8(attr);
285 		info->key.tun_flags |= TUNNEL_GENEVE_OPT;
286 	}
287 
288 	return sizeof(struct geneve_opt) + data_len;
289 }
290 
291 static int ip_tun_parse_opts_vxlan(struct nlattr *attr,
292 				   struct ip_tunnel_info *info, int opts_len,
293 				   struct netlink_ext_ack *extack)
294 {
295 	struct nlattr *tb[LWTUNNEL_IP_OPT_VXLAN_MAX + 1];
296 	int err;
297 
298 	err = nla_parse_nested(tb, LWTUNNEL_IP_OPT_VXLAN_MAX, attr,
299 			       vxlan_opt_policy, extack);
300 	if (err)
301 		return err;
302 
303 	if (!tb[LWTUNNEL_IP_OPT_VXLAN_GBP])
304 		return -EINVAL;
305 
306 	if (info) {
307 		struct vxlan_metadata *md =
308 			ip_tunnel_info_opts(info) + opts_len;
309 
310 		attr = tb[LWTUNNEL_IP_OPT_VXLAN_GBP];
311 		md->gbp = nla_get_u32(attr);
312 		info->key.tun_flags |= TUNNEL_VXLAN_OPT;
313 	}
314 
315 	return sizeof(struct vxlan_metadata);
316 }
317 
318 static int ip_tun_parse_opts_erspan(struct nlattr *attr,
319 				    struct ip_tunnel_info *info, int opts_len,
320 				    struct netlink_ext_ack *extack)
321 {
322 	struct nlattr *tb[LWTUNNEL_IP_OPT_ERSPAN_MAX + 1];
323 	int err;
324 	u8 ver;
325 
326 	err = nla_parse_nested(tb, LWTUNNEL_IP_OPT_ERSPAN_MAX, attr,
327 			       erspan_opt_policy, extack);
328 	if (err)
329 		return err;
330 
331 	if (!tb[LWTUNNEL_IP_OPT_ERSPAN_VER])
332 		return -EINVAL;
333 
334 	ver = nla_get_u8(tb[LWTUNNEL_IP_OPT_ERSPAN_VER]);
335 	if (ver == 1) {
336 		if (!tb[LWTUNNEL_IP_OPT_ERSPAN_INDEX])
337 			return -EINVAL;
338 	} else if (ver == 2) {
339 		if (!tb[LWTUNNEL_IP_OPT_ERSPAN_DIR] ||
340 		    !tb[LWTUNNEL_IP_OPT_ERSPAN_HWID])
341 			return -EINVAL;
342 	} else {
343 		return -EINVAL;
344 	}
345 
346 	if (info) {
347 		struct erspan_metadata *md =
348 			ip_tunnel_info_opts(info) + opts_len;
349 
350 		md->version = ver;
351 		if (ver == 1) {
352 			attr = tb[LWTUNNEL_IP_OPT_ERSPAN_INDEX];
353 			md->u.index = nla_get_be32(attr);
354 		} else {
355 			attr = tb[LWTUNNEL_IP_OPT_ERSPAN_DIR];
356 			md->u.md2.dir = nla_get_u8(attr);
357 			attr = tb[LWTUNNEL_IP_OPT_ERSPAN_HWID];
358 			set_hwid(&md->u.md2, nla_get_u8(attr));
359 		}
360 
361 		info->key.tun_flags |= TUNNEL_ERSPAN_OPT;
362 	}
363 
364 	return sizeof(struct erspan_metadata);
365 }
366 
367 static int ip_tun_parse_opts(struct nlattr *attr, struct ip_tunnel_info *info,
368 			     struct netlink_ext_ack *extack)
369 {
370 	int err, rem, opt_len, opts_len = 0, type = 0;
371 	struct nlattr *nla;
372 
373 	if (!attr)
374 		return 0;
375 
376 	err = nla_validate(nla_data(attr), nla_len(attr), LWTUNNEL_IP_OPTS_MAX,
377 			   ip_opts_policy, extack);
378 	if (err)
379 		return err;
380 
381 	nla_for_each_attr(nla, nla_data(attr), nla_len(attr), rem) {
382 		switch (nla_type(nla)) {
383 		case LWTUNNEL_IP_OPTS_GENEVE:
384 			if (type && type != TUNNEL_GENEVE_OPT)
385 				return -EINVAL;
386 			opt_len = ip_tun_parse_opts_geneve(nla, info, opts_len,
387 							   extack);
388 			if (opt_len < 0)
389 				return opt_len;
390 			opts_len += opt_len;
391 			if (opts_len > IP_TUNNEL_OPTS_MAX)
392 				return -EINVAL;
393 			type = TUNNEL_GENEVE_OPT;
394 			break;
395 		case LWTUNNEL_IP_OPTS_VXLAN:
396 			if (type)
397 				return -EINVAL;
398 			opt_len = ip_tun_parse_opts_vxlan(nla, info, opts_len,
399 							  extack);
400 			if (opt_len < 0)
401 				return opt_len;
402 			opts_len += opt_len;
403 			type = TUNNEL_VXLAN_OPT;
404 			break;
405 		case LWTUNNEL_IP_OPTS_ERSPAN:
406 			if (type)
407 				return -EINVAL;
408 			opt_len = ip_tun_parse_opts_erspan(nla, info, opts_len,
409 							   extack);
410 			if (opt_len < 0)
411 				return opt_len;
412 			opts_len += opt_len;
413 			type = TUNNEL_ERSPAN_OPT;
414 			break;
415 		default:
416 			return -EINVAL;
417 		}
418 	}
419 
420 	return opts_len;
421 }
422 
423 static int ip_tun_get_optlen(struct nlattr *attr,
424 			     struct netlink_ext_ack *extack)
425 {
426 	return ip_tun_parse_opts(attr, NULL, extack);
427 }
428 
429 static int ip_tun_set_opts(struct nlattr *attr, struct ip_tunnel_info *info,
430 			   struct netlink_ext_ack *extack)
431 {
432 	return ip_tun_parse_opts(attr, info, extack);
433 }
434 
435 static int ip_tun_build_state(struct net *net, struct nlattr *attr,
436 			      unsigned int family, const void *cfg,
437 			      struct lwtunnel_state **ts,
438 			      struct netlink_ext_ack *extack)
439 {
440 	struct nlattr *tb[LWTUNNEL_IP_MAX + 1];
441 	struct lwtunnel_state *new_state;
442 	struct ip_tunnel_info *tun_info;
443 	int err, opt_len;
444 
445 	err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP_MAX, attr,
446 					  ip_tun_policy, extack);
447 	if (err < 0)
448 		return err;
449 
450 	opt_len = ip_tun_get_optlen(tb[LWTUNNEL_IP_OPTS], extack);
451 	if (opt_len < 0)
452 		return opt_len;
453 
454 	new_state = lwtunnel_state_alloc(sizeof(*tun_info) + opt_len);
455 	if (!new_state)
456 		return -ENOMEM;
457 
458 	new_state->type = LWTUNNEL_ENCAP_IP;
459 
460 	tun_info = lwt_tun_info(new_state);
461 
462 	err = ip_tun_set_opts(tb[LWTUNNEL_IP_OPTS], tun_info, extack);
463 	if (err < 0) {
464 		lwtstate_free(new_state);
465 		return err;
466 	}
467 
468 #ifdef CONFIG_DST_CACHE
469 	err = dst_cache_init(&tun_info->dst_cache, GFP_KERNEL);
470 	if (err) {
471 		lwtstate_free(new_state);
472 		return err;
473 	}
474 #endif
475 
476 	if (tb[LWTUNNEL_IP_ID])
477 		tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]);
478 
479 	if (tb[LWTUNNEL_IP_DST])
480 		tun_info->key.u.ipv4.dst = nla_get_in_addr(tb[LWTUNNEL_IP_DST]);
481 
482 	if (tb[LWTUNNEL_IP_SRC])
483 		tun_info->key.u.ipv4.src = nla_get_in_addr(tb[LWTUNNEL_IP_SRC]);
484 
485 	if (tb[LWTUNNEL_IP_TTL])
486 		tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]);
487 
488 	if (tb[LWTUNNEL_IP_TOS])
489 		tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]);
490 
491 	if (tb[LWTUNNEL_IP_FLAGS])
492 		tun_info->key.tun_flags |=
493 				(nla_get_be16(tb[LWTUNNEL_IP_FLAGS]) &
494 				 ~TUNNEL_OPTIONS_PRESENT);
495 
496 	tun_info->mode = IP_TUNNEL_INFO_TX;
497 	tun_info->options_len = opt_len;
498 
499 	*ts = new_state;
500 
501 	return 0;
502 }
503 
504 static void ip_tun_destroy_state(struct lwtunnel_state *lwtstate)
505 {
506 #ifdef CONFIG_DST_CACHE
507 	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
508 
509 	dst_cache_destroy(&tun_info->dst_cache);
510 #endif
511 }
512 
513 static int ip_tun_fill_encap_opts_geneve(struct sk_buff *skb,
514 					 struct ip_tunnel_info *tun_info)
515 {
516 	struct geneve_opt *opt;
517 	struct nlattr *nest;
518 	int offset = 0;
519 
520 	nest = nla_nest_start_noflag(skb, LWTUNNEL_IP_OPTS_GENEVE);
521 	if (!nest)
522 		return -ENOMEM;
523 
524 	while (tun_info->options_len > offset) {
525 		opt = ip_tunnel_info_opts(tun_info) + offset;
526 		if (nla_put_be16(skb, LWTUNNEL_IP_OPT_GENEVE_CLASS,
527 				 opt->opt_class) ||
528 		    nla_put_u8(skb, LWTUNNEL_IP_OPT_GENEVE_TYPE, opt->type) ||
529 		    nla_put(skb, LWTUNNEL_IP_OPT_GENEVE_DATA, opt->length * 4,
530 			    opt->opt_data)) {
531 			nla_nest_cancel(skb, nest);
532 			return -ENOMEM;
533 		}
534 		offset += sizeof(*opt) + opt->length * 4;
535 	}
536 
537 	nla_nest_end(skb, nest);
538 	return 0;
539 }
540 
541 static int ip_tun_fill_encap_opts_vxlan(struct sk_buff *skb,
542 					struct ip_tunnel_info *tun_info)
543 {
544 	struct vxlan_metadata *md;
545 	struct nlattr *nest;
546 
547 	nest = nla_nest_start_noflag(skb, LWTUNNEL_IP_OPTS_VXLAN);
548 	if (!nest)
549 		return -ENOMEM;
550 
551 	md = ip_tunnel_info_opts(tun_info);
552 	if (nla_put_u32(skb, LWTUNNEL_IP_OPT_VXLAN_GBP, md->gbp)) {
553 		nla_nest_cancel(skb, nest);
554 		return -ENOMEM;
555 	}
556 
557 	nla_nest_end(skb, nest);
558 	return 0;
559 }
560 
561 static int ip_tun_fill_encap_opts_erspan(struct sk_buff *skb,
562 					 struct ip_tunnel_info *tun_info)
563 {
564 	struct erspan_metadata *md;
565 	struct nlattr *nest;
566 
567 	nest = nla_nest_start_noflag(skb, LWTUNNEL_IP_OPTS_ERSPAN);
568 	if (!nest)
569 		return -ENOMEM;
570 
571 	md = ip_tunnel_info_opts(tun_info);
572 	if (nla_put_u8(skb, LWTUNNEL_IP_OPT_ERSPAN_VER, md->version))
573 		goto err;
574 
575 	if (md->version == 1 &&
576 	    nla_put_be32(skb, LWTUNNEL_IP_OPT_ERSPAN_INDEX, md->u.index))
577 		goto err;
578 
579 	if (md->version == 2 &&
580 	    (nla_put_u8(skb, LWTUNNEL_IP_OPT_ERSPAN_DIR, md->u.md2.dir) ||
581 	     nla_put_u8(skb, LWTUNNEL_IP_OPT_ERSPAN_HWID,
582 			get_hwid(&md->u.md2))))
583 		goto err;
584 
585 	nla_nest_end(skb, nest);
586 	return 0;
587 err:
588 	nla_nest_cancel(skb, nest);
589 	return -ENOMEM;
590 }
591 
592 static int ip_tun_fill_encap_opts(struct sk_buff *skb, int type,
593 				  struct ip_tunnel_info *tun_info)
594 {
595 	struct nlattr *nest;
596 	int err = 0;
597 
598 	if (!(tun_info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))
599 		return 0;
600 
601 	nest = nla_nest_start_noflag(skb, type);
602 	if (!nest)
603 		return -ENOMEM;
604 
605 	if (tun_info->key.tun_flags & TUNNEL_GENEVE_OPT)
606 		err = ip_tun_fill_encap_opts_geneve(skb, tun_info);
607 	else if (tun_info->key.tun_flags & TUNNEL_VXLAN_OPT)
608 		err = ip_tun_fill_encap_opts_vxlan(skb, tun_info);
609 	else if (tun_info->key.tun_flags & TUNNEL_ERSPAN_OPT)
610 		err = ip_tun_fill_encap_opts_erspan(skb, tun_info);
611 
612 	if (err) {
613 		nla_nest_cancel(skb, nest);
614 		return err;
615 	}
616 
617 	nla_nest_end(skb, nest);
618 	return 0;
619 }
620 
621 static int ip_tun_fill_encap_info(struct sk_buff *skb,
622 				  struct lwtunnel_state *lwtstate)
623 {
624 	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
625 
626 	if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id,
627 			 LWTUNNEL_IP_PAD) ||
628 	    nla_put_in_addr(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) ||
629 	    nla_put_in_addr(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) ||
630 	    nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) ||
631 	    nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) ||
632 	    nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags) ||
633 	    ip_tun_fill_encap_opts(skb, LWTUNNEL_IP_OPTS, tun_info))
634 		return -ENOMEM;
635 
636 	return 0;
637 }
638 
639 static int ip_tun_opts_nlsize(struct ip_tunnel_info *info)
640 {
641 	int opt_len;
642 
643 	if (!(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))
644 		return 0;
645 
646 	opt_len = nla_total_size(0);		/* LWTUNNEL_IP_OPTS */
647 	if (info->key.tun_flags & TUNNEL_GENEVE_OPT) {
648 		struct geneve_opt *opt;
649 		int offset = 0;
650 
651 		opt_len += nla_total_size(0);	/* LWTUNNEL_IP_OPTS_GENEVE */
652 		while (info->options_len > offset) {
653 			opt = ip_tunnel_info_opts(info) + offset;
654 			opt_len += nla_total_size(2)	/* OPT_GENEVE_CLASS */
655 				   + nla_total_size(1)	/* OPT_GENEVE_TYPE */
656 				   + nla_total_size(opt->length * 4);
657 							/* OPT_GENEVE_DATA */
658 			offset += sizeof(*opt) + opt->length * 4;
659 		}
660 	} else if (info->key.tun_flags & TUNNEL_VXLAN_OPT) {
661 		opt_len += nla_total_size(0)	/* LWTUNNEL_IP_OPTS_VXLAN */
662 			   + nla_total_size(4);	/* OPT_VXLAN_GBP */
663 	} else if (info->key.tun_flags & TUNNEL_ERSPAN_OPT) {
664 		struct erspan_metadata *md = ip_tunnel_info_opts(info);
665 
666 		opt_len += nla_total_size(0)	/* LWTUNNEL_IP_OPTS_ERSPAN */
667 			   + nla_total_size(1)	/* OPT_ERSPAN_VER */
668 			   + (md->version == 1 ? nla_total_size(4)
669 						/* OPT_ERSPAN_INDEX (v1) */
670 					       : nla_total_size(1) +
671 						 nla_total_size(1));
672 						/* OPT_ERSPAN_DIR + HWID (v2) */
673 	}
674 
675 	return opt_len;
676 }
677 
678 static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
679 {
680 	return nla_total_size_64bit(8)	/* LWTUNNEL_IP_ID */
681 		+ nla_total_size(4)	/* LWTUNNEL_IP_DST */
682 		+ nla_total_size(4)	/* LWTUNNEL_IP_SRC */
683 		+ nla_total_size(1)	/* LWTUNNEL_IP_TOS */
684 		+ nla_total_size(1)	/* LWTUNNEL_IP_TTL */
685 		+ nla_total_size(2)	/* LWTUNNEL_IP_FLAGS */
686 		+ ip_tun_opts_nlsize(lwt_tun_info(lwtstate));
687 					/* LWTUNNEL_IP_OPTS */
688 }
689 
690 static int ip_tun_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
691 {
692 	struct ip_tunnel_info *info_a = lwt_tun_info(a);
693 	struct ip_tunnel_info *info_b = lwt_tun_info(b);
694 
695 	return memcmp(info_a, info_b, sizeof(info_a->key)) ||
696 	       info_a->mode != info_b->mode ||
697 	       info_a->options_len != info_b->options_len ||
698 	       memcmp(ip_tunnel_info_opts(info_a),
699 		      ip_tunnel_info_opts(info_b), info_a->options_len);
700 }
701 
702 static const struct lwtunnel_encap_ops ip_tun_lwt_ops = {
703 	.build_state = ip_tun_build_state,
704 	.destroy_state = ip_tun_destroy_state,
705 	.fill_encap = ip_tun_fill_encap_info,
706 	.get_encap_size = ip_tun_encap_nlsize,
707 	.cmp_encap = ip_tun_cmp_encap,
708 	.owner = THIS_MODULE,
709 };
710 
711 static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = {
712 	[LWTUNNEL_IP6_UNSPEC]	= { .strict_start_type = LWTUNNEL_IP6_OPTS },
713 	[LWTUNNEL_IP6_ID]		= { .type = NLA_U64 },
714 	[LWTUNNEL_IP6_DST]		= { .len = sizeof(struct in6_addr) },
715 	[LWTUNNEL_IP6_SRC]		= { .len = sizeof(struct in6_addr) },
716 	[LWTUNNEL_IP6_HOPLIMIT]		= { .type = NLA_U8 },
717 	[LWTUNNEL_IP6_TC]		= { .type = NLA_U8 },
718 	[LWTUNNEL_IP6_FLAGS]		= { .type = NLA_U16 },
719 	[LWTUNNEL_IP6_OPTS]		= { .type = NLA_NESTED },
720 };
721 
722 static int ip6_tun_build_state(struct net *net, struct nlattr *attr,
723 			       unsigned int family, const void *cfg,
724 			       struct lwtunnel_state **ts,
725 			       struct netlink_ext_ack *extack)
726 {
727 	struct nlattr *tb[LWTUNNEL_IP6_MAX + 1];
728 	struct lwtunnel_state *new_state;
729 	struct ip_tunnel_info *tun_info;
730 	int err, opt_len;
731 
732 	err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP6_MAX, attr,
733 					  ip6_tun_policy, extack);
734 	if (err < 0)
735 		return err;
736 
737 	opt_len = ip_tun_get_optlen(tb[LWTUNNEL_IP6_OPTS], extack);
738 	if (opt_len < 0)
739 		return opt_len;
740 
741 	new_state = lwtunnel_state_alloc(sizeof(*tun_info) + opt_len);
742 	if (!new_state)
743 		return -ENOMEM;
744 
745 	new_state->type = LWTUNNEL_ENCAP_IP6;
746 
747 	tun_info = lwt_tun_info(new_state);
748 
749 	err = ip_tun_set_opts(tb[LWTUNNEL_IP6_OPTS], tun_info, extack);
750 	if (err < 0) {
751 		lwtstate_free(new_state);
752 		return err;
753 	}
754 
755 	if (tb[LWTUNNEL_IP6_ID])
756 		tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]);
757 
758 	if (tb[LWTUNNEL_IP6_DST])
759 		tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]);
760 
761 	if (tb[LWTUNNEL_IP6_SRC])
762 		tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]);
763 
764 	if (tb[LWTUNNEL_IP6_HOPLIMIT])
765 		tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]);
766 
767 	if (tb[LWTUNNEL_IP6_TC])
768 		tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]);
769 
770 	if (tb[LWTUNNEL_IP6_FLAGS])
771 		tun_info->key.tun_flags |=
772 				(nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]) &
773 				 ~TUNNEL_OPTIONS_PRESENT);
774 
775 	tun_info->mode = IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6;
776 	tun_info->options_len = opt_len;
777 
778 	*ts = new_state;
779 
780 	return 0;
781 }
782 
783 static int ip6_tun_fill_encap_info(struct sk_buff *skb,
784 				   struct lwtunnel_state *lwtstate)
785 {
786 	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
787 
788 	if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id,
789 			 LWTUNNEL_IP6_PAD) ||
790 	    nla_put_in6_addr(skb, LWTUNNEL_IP6_DST, &tun_info->key.u.ipv6.dst) ||
791 	    nla_put_in6_addr(skb, LWTUNNEL_IP6_SRC, &tun_info->key.u.ipv6.src) ||
792 	    nla_put_u8(skb, LWTUNNEL_IP6_TC, tun_info->key.tos) ||
793 	    nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, tun_info->key.ttl) ||
794 	    nla_put_be16(skb, LWTUNNEL_IP6_FLAGS, tun_info->key.tun_flags) ||
795 	    ip_tun_fill_encap_opts(skb, LWTUNNEL_IP6_OPTS, tun_info))
796 		return -ENOMEM;
797 
798 	return 0;
799 }
800 
801 static int ip6_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
802 {
803 	return nla_total_size_64bit(8)	/* LWTUNNEL_IP6_ID */
804 		+ nla_total_size(16)	/* LWTUNNEL_IP6_DST */
805 		+ nla_total_size(16)	/* LWTUNNEL_IP6_SRC */
806 		+ nla_total_size(1)	/* LWTUNNEL_IP6_HOPLIMIT */
807 		+ nla_total_size(1)	/* LWTUNNEL_IP6_TC */
808 		+ nla_total_size(2)	/* LWTUNNEL_IP6_FLAGS */
809 		+ ip_tun_opts_nlsize(lwt_tun_info(lwtstate));
810 					/* LWTUNNEL_IP6_OPTS */
811 }
812 
813 static const struct lwtunnel_encap_ops ip6_tun_lwt_ops = {
814 	.build_state = ip6_tun_build_state,
815 	.fill_encap = ip6_tun_fill_encap_info,
816 	.get_encap_size = ip6_tun_encap_nlsize,
817 	.cmp_encap = ip_tun_cmp_encap,
818 	.owner = THIS_MODULE,
819 };
820 
821 void __init ip_tunnel_core_init(void)
822 {
823 	/* If you land here, make sure whether increasing ip_tunnel_info's
824 	 * options_len is a reasonable choice with its usage in front ends
825 	 * (f.e., it's part of flow keys, etc).
826 	 */
827 	BUILD_BUG_ON(IP_TUNNEL_OPTS_MAX != 255);
828 
829 	lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP);
830 	lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6);
831 }
832 
833 DEFINE_STATIC_KEY_FALSE(ip_tunnel_metadata_cnt);
834 EXPORT_SYMBOL(ip_tunnel_metadata_cnt);
835 
836 void ip_tunnel_need_metadata(void)
837 {
838 	static_branch_inc(&ip_tunnel_metadata_cnt);
839 }
840 EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata);
841 
842 void ip_tunnel_unneed_metadata(void)
843 {
844 	static_branch_dec(&ip_tunnel_metadata_cnt);
845 }
846 EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata);
847