xref: /linux/net/ipv4/ip_tunnel_core.c (revision e58e871becec2d3b04ed91c0c16fe8deac9c9dfa)
1 /*
2  * Copyright (c) 2013 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16  * 02110-1301, USA
17  */
18 
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 
21 #include <linux/types.h>
22 #include <linux/kernel.h>
23 #include <linux/skbuff.h>
24 #include <linux/netdevice.h>
25 #include <linux/in.h>
26 #include <linux/if_arp.h>
27 #include <linux/init.h>
28 #include <linux/in6.h>
29 #include <linux/inetdevice.h>
30 #include <linux/netfilter_ipv4.h>
31 #include <linux/etherdevice.h>
32 #include <linux/if_ether.h>
33 #include <linux/if_vlan.h>
34 #include <linux/static_key.h>
35 
36 #include <net/ip.h>
37 #include <net/icmp.h>
38 #include <net/protocol.h>
39 #include <net/ip_tunnels.h>
40 #include <net/ip6_tunnel.h>
41 #include <net/arp.h>
42 #include <net/checksum.h>
43 #include <net/dsfield.h>
44 #include <net/inet_ecn.h>
45 #include <net/xfrm.h>
46 #include <net/net_namespace.h>
47 #include <net/netns/generic.h>
48 #include <net/rtnetlink.h>
49 #include <net/dst_metadata.h>
50 
51 const struct ip_tunnel_encap_ops __rcu *
52 		iptun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
53 EXPORT_SYMBOL(iptun_encaps);
54 
55 const struct ip6_tnl_encap_ops __rcu *
56 		ip6tun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
57 EXPORT_SYMBOL(ip6tun_encaps);
58 
59 void iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb,
60 		   __be32 src, __be32 dst, __u8 proto,
61 		   __u8 tos, __u8 ttl, __be16 df, bool xnet)
62 {
63 	int pkt_len = skb->len - skb_inner_network_offset(skb);
64 	struct net *net = dev_net(rt->dst.dev);
65 	struct net_device *dev = skb->dev;
66 	struct iphdr *iph;
67 	int err;
68 
69 	skb_scrub_packet(skb, xnet);
70 
71 	skb_clear_hash_if_not_l4(skb);
72 	skb_dst_set(skb, &rt->dst);
73 	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
74 
75 	/* Push down and install the IP header. */
76 	skb_push(skb, sizeof(struct iphdr));
77 	skb_reset_network_header(skb);
78 
79 	iph = ip_hdr(skb);
80 
81 	iph->version	=	4;
82 	iph->ihl	=	sizeof(struct iphdr) >> 2;
83 	iph->frag_off	=	df;
84 	iph->protocol	=	proto;
85 	iph->tos	=	tos;
86 	iph->daddr	=	dst;
87 	iph->saddr	=	src;
88 	iph->ttl	=	ttl;
89 	__ip_select_ident(net, iph, skb_shinfo(skb)->gso_segs ?: 1);
90 
91 	err = ip_local_out(net, sk, skb);
92 	if (unlikely(net_xmit_eval(err)))
93 		pkt_len = 0;
94 	iptunnel_xmit_stats(dev, pkt_len);
95 }
96 EXPORT_SYMBOL_GPL(iptunnel_xmit);
97 
98 int __iptunnel_pull_header(struct sk_buff *skb, int hdr_len,
99 			   __be16 inner_proto, bool raw_proto, bool xnet)
100 {
101 	if (unlikely(!pskb_may_pull(skb, hdr_len)))
102 		return -ENOMEM;
103 
104 	skb_pull_rcsum(skb, hdr_len);
105 
106 	if (!raw_proto && inner_proto == htons(ETH_P_TEB)) {
107 		struct ethhdr *eh;
108 
109 		if (unlikely(!pskb_may_pull(skb, ETH_HLEN)))
110 			return -ENOMEM;
111 
112 		eh = (struct ethhdr *)skb->data;
113 		if (likely(eth_proto_is_802_3(eh->h_proto)))
114 			skb->protocol = eh->h_proto;
115 		else
116 			skb->protocol = htons(ETH_P_802_2);
117 
118 	} else {
119 		skb->protocol = inner_proto;
120 	}
121 
122 	skb_clear_hash_if_not_l4(skb);
123 	skb->vlan_tci = 0;
124 	skb_set_queue_mapping(skb, 0);
125 	skb_scrub_packet(skb, xnet);
126 
127 	return iptunnel_pull_offloads(skb);
128 }
129 EXPORT_SYMBOL_GPL(__iptunnel_pull_header);
130 
131 struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md,
132 					     gfp_t flags)
133 {
134 	struct metadata_dst *res;
135 	struct ip_tunnel_info *dst, *src;
136 
137 	if (!md || md->u.tun_info.mode & IP_TUNNEL_INFO_TX)
138 		return NULL;
139 
140 	res = metadata_dst_alloc(0, flags);
141 	if (!res)
142 		return NULL;
143 
144 	dst = &res->u.tun_info;
145 	src = &md->u.tun_info;
146 	dst->key.tun_id = src->key.tun_id;
147 	if (src->mode & IP_TUNNEL_INFO_IPV6)
148 		memcpy(&dst->key.u.ipv6.dst, &src->key.u.ipv6.src,
149 		       sizeof(struct in6_addr));
150 	else
151 		dst->key.u.ipv4.dst = src->key.u.ipv4.src;
152 	dst->mode = src->mode | IP_TUNNEL_INFO_TX;
153 
154 	return res;
155 }
156 EXPORT_SYMBOL_GPL(iptunnel_metadata_reply);
157 
158 int iptunnel_handle_offloads(struct sk_buff *skb,
159 			     int gso_type_mask)
160 {
161 	int err;
162 
163 	if (likely(!skb->encapsulation)) {
164 		skb_reset_inner_headers(skb);
165 		skb->encapsulation = 1;
166 	}
167 
168 	if (skb_is_gso(skb)) {
169 		err = skb_header_unclone(skb, GFP_ATOMIC);
170 		if (unlikely(err))
171 			return err;
172 		skb_shinfo(skb)->gso_type |= gso_type_mask;
173 		return 0;
174 	}
175 
176 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
177 		skb->ip_summed = CHECKSUM_NONE;
178 		/* We clear encapsulation here to prevent badly-written
179 		 * drivers potentially deciding to offload an inner checksum
180 		 * if we set CHECKSUM_PARTIAL on the outer header.
181 		 * This should go away when the drivers are all fixed.
182 		 */
183 		skb->encapsulation = 0;
184 	}
185 
186 	return 0;
187 }
188 EXPORT_SYMBOL_GPL(iptunnel_handle_offloads);
189 
190 /* Often modified stats are per cpu, other are shared (netdev->stats) */
191 void ip_tunnel_get_stats64(struct net_device *dev,
192 			   struct rtnl_link_stats64 *tot)
193 {
194 	int i;
195 
196 	netdev_stats_to_stats64(tot, &dev->stats);
197 
198 	for_each_possible_cpu(i) {
199 		const struct pcpu_sw_netstats *tstats =
200 						   per_cpu_ptr(dev->tstats, i);
201 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
202 		unsigned int start;
203 
204 		do {
205 			start = u64_stats_fetch_begin_irq(&tstats->syncp);
206 			rx_packets = tstats->rx_packets;
207 			tx_packets = tstats->tx_packets;
208 			rx_bytes = tstats->rx_bytes;
209 			tx_bytes = tstats->tx_bytes;
210 		} while (u64_stats_fetch_retry_irq(&tstats->syncp, start));
211 
212 		tot->rx_packets += rx_packets;
213 		tot->tx_packets += tx_packets;
214 		tot->rx_bytes   += rx_bytes;
215 		tot->tx_bytes   += tx_bytes;
216 	}
217 }
218 EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64);
219 
220 static const struct nla_policy ip_tun_policy[LWTUNNEL_IP_MAX + 1] = {
221 	[LWTUNNEL_IP_ID]	= { .type = NLA_U64 },
222 	[LWTUNNEL_IP_DST]	= { .type = NLA_U32 },
223 	[LWTUNNEL_IP_SRC]	= { .type = NLA_U32 },
224 	[LWTUNNEL_IP_TTL]	= { .type = NLA_U8 },
225 	[LWTUNNEL_IP_TOS]	= { .type = NLA_U8 },
226 	[LWTUNNEL_IP_FLAGS]	= { .type = NLA_U16 },
227 };
228 
229 static int ip_tun_build_state(struct nlattr *attr,
230 			      unsigned int family, const void *cfg,
231 			      struct lwtunnel_state **ts)
232 {
233 	struct ip_tunnel_info *tun_info;
234 	struct lwtunnel_state *new_state;
235 	struct nlattr *tb[LWTUNNEL_IP_MAX + 1];
236 	int err;
237 
238 	err = nla_parse_nested(tb, LWTUNNEL_IP_MAX, attr, ip_tun_policy, NULL);
239 	if (err < 0)
240 		return err;
241 
242 	new_state = lwtunnel_state_alloc(sizeof(*tun_info));
243 	if (!new_state)
244 		return -ENOMEM;
245 
246 	new_state->type = LWTUNNEL_ENCAP_IP;
247 
248 	tun_info = lwt_tun_info(new_state);
249 
250 	if (tb[LWTUNNEL_IP_ID])
251 		tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]);
252 
253 	if (tb[LWTUNNEL_IP_DST])
254 		tun_info->key.u.ipv4.dst = nla_get_in_addr(tb[LWTUNNEL_IP_DST]);
255 
256 	if (tb[LWTUNNEL_IP_SRC])
257 		tun_info->key.u.ipv4.src = nla_get_in_addr(tb[LWTUNNEL_IP_SRC]);
258 
259 	if (tb[LWTUNNEL_IP_TTL])
260 		tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]);
261 
262 	if (tb[LWTUNNEL_IP_TOS])
263 		tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]);
264 
265 	if (tb[LWTUNNEL_IP_FLAGS])
266 		tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP_FLAGS]);
267 
268 	tun_info->mode = IP_TUNNEL_INFO_TX;
269 	tun_info->options_len = 0;
270 
271 	*ts = new_state;
272 
273 	return 0;
274 }
275 
276 static int ip_tun_fill_encap_info(struct sk_buff *skb,
277 				  struct lwtunnel_state *lwtstate)
278 {
279 	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
280 
281 	if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id,
282 			 LWTUNNEL_IP_PAD) ||
283 	    nla_put_in_addr(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) ||
284 	    nla_put_in_addr(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) ||
285 	    nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) ||
286 	    nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) ||
287 	    nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags))
288 		return -ENOMEM;
289 
290 	return 0;
291 }
292 
293 static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
294 {
295 	return nla_total_size_64bit(8)	/* LWTUNNEL_IP_ID */
296 		+ nla_total_size(4)	/* LWTUNNEL_IP_DST */
297 		+ nla_total_size(4)	/* LWTUNNEL_IP_SRC */
298 		+ nla_total_size(1)	/* LWTUNNEL_IP_TOS */
299 		+ nla_total_size(1)	/* LWTUNNEL_IP_TTL */
300 		+ nla_total_size(2);	/* LWTUNNEL_IP_FLAGS */
301 }
302 
303 static int ip_tun_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
304 {
305 	return memcmp(lwt_tun_info(a), lwt_tun_info(b),
306 		      sizeof(struct ip_tunnel_info));
307 }
308 
309 static const struct lwtunnel_encap_ops ip_tun_lwt_ops = {
310 	.build_state = ip_tun_build_state,
311 	.fill_encap = ip_tun_fill_encap_info,
312 	.get_encap_size = ip_tun_encap_nlsize,
313 	.cmp_encap = ip_tun_cmp_encap,
314 	.owner = THIS_MODULE,
315 };
316 
317 static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = {
318 	[LWTUNNEL_IP6_ID]		= { .type = NLA_U64 },
319 	[LWTUNNEL_IP6_DST]		= { .len = sizeof(struct in6_addr) },
320 	[LWTUNNEL_IP6_SRC]		= { .len = sizeof(struct in6_addr) },
321 	[LWTUNNEL_IP6_HOPLIMIT]		= { .type = NLA_U8 },
322 	[LWTUNNEL_IP6_TC]		= { .type = NLA_U8 },
323 	[LWTUNNEL_IP6_FLAGS]		= { .type = NLA_U16 },
324 };
325 
326 static int ip6_tun_build_state(struct nlattr *attr,
327 			       unsigned int family, const void *cfg,
328 			       struct lwtunnel_state **ts)
329 {
330 	struct ip_tunnel_info *tun_info;
331 	struct lwtunnel_state *new_state;
332 	struct nlattr *tb[LWTUNNEL_IP6_MAX + 1];
333 	int err;
334 
335 	err = nla_parse_nested(tb, LWTUNNEL_IP6_MAX, attr, ip6_tun_policy,
336 			       NULL);
337 	if (err < 0)
338 		return err;
339 
340 	new_state = lwtunnel_state_alloc(sizeof(*tun_info));
341 	if (!new_state)
342 		return -ENOMEM;
343 
344 	new_state->type = LWTUNNEL_ENCAP_IP6;
345 
346 	tun_info = lwt_tun_info(new_state);
347 
348 	if (tb[LWTUNNEL_IP6_ID])
349 		tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]);
350 
351 	if (tb[LWTUNNEL_IP6_DST])
352 		tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]);
353 
354 	if (tb[LWTUNNEL_IP6_SRC])
355 		tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]);
356 
357 	if (tb[LWTUNNEL_IP6_HOPLIMIT])
358 		tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]);
359 
360 	if (tb[LWTUNNEL_IP6_TC])
361 		tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]);
362 
363 	if (tb[LWTUNNEL_IP6_FLAGS])
364 		tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]);
365 
366 	tun_info->mode = IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6;
367 	tun_info->options_len = 0;
368 
369 	*ts = new_state;
370 
371 	return 0;
372 }
373 
374 static int ip6_tun_fill_encap_info(struct sk_buff *skb,
375 				   struct lwtunnel_state *lwtstate)
376 {
377 	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
378 
379 	if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id,
380 			 LWTUNNEL_IP6_PAD) ||
381 	    nla_put_in6_addr(skb, LWTUNNEL_IP6_DST, &tun_info->key.u.ipv6.dst) ||
382 	    nla_put_in6_addr(skb, LWTUNNEL_IP6_SRC, &tun_info->key.u.ipv6.src) ||
383 	    nla_put_u8(skb, LWTUNNEL_IP6_TC, tun_info->key.tos) ||
384 	    nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, tun_info->key.ttl) ||
385 	    nla_put_be16(skb, LWTUNNEL_IP6_FLAGS, tun_info->key.tun_flags))
386 		return -ENOMEM;
387 
388 	return 0;
389 }
390 
391 static int ip6_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
392 {
393 	return nla_total_size_64bit(8)	/* LWTUNNEL_IP6_ID */
394 		+ nla_total_size(16)	/* LWTUNNEL_IP6_DST */
395 		+ nla_total_size(16)	/* LWTUNNEL_IP6_SRC */
396 		+ nla_total_size(1)	/* LWTUNNEL_IP6_HOPLIMIT */
397 		+ nla_total_size(1)	/* LWTUNNEL_IP6_TC */
398 		+ nla_total_size(2);	/* LWTUNNEL_IP6_FLAGS */
399 }
400 
401 static const struct lwtunnel_encap_ops ip6_tun_lwt_ops = {
402 	.build_state = ip6_tun_build_state,
403 	.fill_encap = ip6_tun_fill_encap_info,
404 	.get_encap_size = ip6_tun_encap_nlsize,
405 	.cmp_encap = ip_tun_cmp_encap,
406 	.owner = THIS_MODULE,
407 };
408 
409 void __init ip_tunnel_core_init(void)
410 {
411 	/* If you land here, make sure whether increasing ip_tunnel_info's
412 	 * options_len is a reasonable choice with its usage in front ends
413 	 * (f.e., it's part of flow keys, etc).
414 	 */
415 	BUILD_BUG_ON(IP_TUNNEL_OPTS_MAX != 255);
416 
417 	lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP);
418 	lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6);
419 }
420 
421 struct static_key ip_tunnel_metadata_cnt = STATIC_KEY_INIT_FALSE;
422 EXPORT_SYMBOL(ip_tunnel_metadata_cnt);
423 
424 void ip_tunnel_need_metadata(void)
425 {
426 	static_key_slow_inc(&ip_tunnel_metadata_cnt);
427 }
428 EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata);
429 
430 void ip_tunnel_unneed_metadata(void)
431 {
432 	static_key_slow_dec(&ip_tunnel_metadata_cnt);
433 }
434 EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata);
435