xref: /openbmc/linux/net/ipv4/ip_tunnel_core.c (revision ebd09753)
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->type != METADATA_IP_TUNNEL ||
138 	    md->u.tun_info.mode & IP_TUNNEL_INFO_TX)
139 
140 		return NULL;
141 
142 	res = metadata_dst_alloc(0, METADATA_IP_TUNNEL, flags);
143 	if (!res)
144 		return NULL;
145 
146 	dst = &res->u.tun_info;
147 	src = &md->u.tun_info;
148 	dst->key.tun_id = src->key.tun_id;
149 	if (src->mode & IP_TUNNEL_INFO_IPV6)
150 		memcpy(&dst->key.u.ipv6.dst, &src->key.u.ipv6.src,
151 		       sizeof(struct in6_addr));
152 	else
153 		dst->key.u.ipv4.dst = src->key.u.ipv4.src;
154 	dst->mode = src->mode | IP_TUNNEL_INFO_TX;
155 
156 	return res;
157 }
158 EXPORT_SYMBOL_GPL(iptunnel_metadata_reply);
159 
160 int iptunnel_handle_offloads(struct sk_buff *skb,
161 			     int gso_type_mask)
162 {
163 	int err;
164 
165 	if (likely(!skb->encapsulation)) {
166 		skb_reset_inner_headers(skb);
167 		skb->encapsulation = 1;
168 	}
169 
170 	if (skb_is_gso(skb)) {
171 		err = skb_header_unclone(skb, GFP_ATOMIC);
172 		if (unlikely(err))
173 			return err;
174 		skb_shinfo(skb)->gso_type |= gso_type_mask;
175 		return 0;
176 	}
177 
178 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
179 		skb->ip_summed = CHECKSUM_NONE;
180 		/* We clear encapsulation here to prevent badly-written
181 		 * drivers potentially deciding to offload an inner checksum
182 		 * if we set CHECKSUM_PARTIAL on the outer header.
183 		 * This should go away when the drivers are all fixed.
184 		 */
185 		skb->encapsulation = 0;
186 	}
187 
188 	return 0;
189 }
190 EXPORT_SYMBOL_GPL(iptunnel_handle_offloads);
191 
192 /* Often modified stats are per cpu, other are shared (netdev->stats) */
193 void ip_tunnel_get_stats64(struct net_device *dev,
194 			   struct rtnl_link_stats64 *tot)
195 {
196 	int i;
197 
198 	netdev_stats_to_stats64(tot, &dev->stats);
199 
200 	for_each_possible_cpu(i) {
201 		const struct pcpu_sw_netstats *tstats =
202 						   per_cpu_ptr(dev->tstats, i);
203 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
204 		unsigned int start;
205 
206 		do {
207 			start = u64_stats_fetch_begin_irq(&tstats->syncp);
208 			rx_packets = tstats->rx_packets;
209 			tx_packets = tstats->tx_packets;
210 			rx_bytes = tstats->rx_bytes;
211 			tx_bytes = tstats->tx_bytes;
212 		} while (u64_stats_fetch_retry_irq(&tstats->syncp, start));
213 
214 		tot->rx_packets += rx_packets;
215 		tot->tx_packets += tx_packets;
216 		tot->rx_bytes   += rx_bytes;
217 		tot->tx_bytes   += tx_bytes;
218 	}
219 }
220 EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64);
221 
222 static const struct nla_policy ip_tun_policy[LWTUNNEL_IP_MAX + 1] = {
223 	[LWTUNNEL_IP_ID]	= { .type = NLA_U64 },
224 	[LWTUNNEL_IP_DST]	= { .type = NLA_U32 },
225 	[LWTUNNEL_IP_SRC]	= { .type = NLA_U32 },
226 	[LWTUNNEL_IP_TTL]	= { .type = NLA_U8 },
227 	[LWTUNNEL_IP_TOS]	= { .type = NLA_U8 },
228 	[LWTUNNEL_IP_FLAGS]	= { .type = NLA_U16 },
229 };
230 
231 static int ip_tun_build_state(struct nlattr *attr,
232 			      unsigned int family, const void *cfg,
233 			      struct lwtunnel_state **ts,
234 			      struct netlink_ext_ack *extack)
235 {
236 	struct ip_tunnel_info *tun_info;
237 	struct lwtunnel_state *new_state;
238 	struct nlattr *tb[LWTUNNEL_IP_MAX + 1];
239 	int err;
240 
241 	err = nla_parse_nested(tb, LWTUNNEL_IP_MAX, attr, ip_tun_policy,
242 			       extack);
243 	if (err < 0)
244 		return err;
245 
246 	new_state = lwtunnel_state_alloc(sizeof(*tun_info));
247 	if (!new_state)
248 		return -ENOMEM;
249 
250 	new_state->type = LWTUNNEL_ENCAP_IP;
251 
252 	tun_info = lwt_tun_info(new_state);
253 
254 	if (tb[LWTUNNEL_IP_ID])
255 		tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]);
256 
257 	if (tb[LWTUNNEL_IP_DST])
258 		tun_info->key.u.ipv4.dst = nla_get_in_addr(tb[LWTUNNEL_IP_DST]);
259 
260 	if (tb[LWTUNNEL_IP_SRC])
261 		tun_info->key.u.ipv4.src = nla_get_in_addr(tb[LWTUNNEL_IP_SRC]);
262 
263 	if (tb[LWTUNNEL_IP_TTL])
264 		tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]);
265 
266 	if (tb[LWTUNNEL_IP_TOS])
267 		tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]);
268 
269 	if (tb[LWTUNNEL_IP_FLAGS])
270 		tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP_FLAGS]);
271 
272 	tun_info->mode = IP_TUNNEL_INFO_TX;
273 	tun_info->options_len = 0;
274 
275 	*ts = new_state;
276 
277 	return 0;
278 }
279 
280 static int ip_tun_fill_encap_info(struct sk_buff *skb,
281 				  struct lwtunnel_state *lwtstate)
282 {
283 	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
284 
285 	if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id,
286 			 LWTUNNEL_IP_PAD) ||
287 	    nla_put_in_addr(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) ||
288 	    nla_put_in_addr(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) ||
289 	    nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) ||
290 	    nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) ||
291 	    nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags))
292 		return -ENOMEM;
293 
294 	return 0;
295 }
296 
297 static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
298 {
299 	return nla_total_size_64bit(8)	/* LWTUNNEL_IP_ID */
300 		+ nla_total_size(4)	/* LWTUNNEL_IP_DST */
301 		+ nla_total_size(4)	/* LWTUNNEL_IP_SRC */
302 		+ nla_total_size(1)	/* LWTUNNEL_IP_TOS */
303 		+ nla_total_size(1)	/* LWTUNNEL_IP_TTL */
304 		+ nla_total_size(2);	/* LWTUNNEL_IP_FLAGS */
305 }
306 
307 static int ip_tun_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
308 {
309 	return memcmp(lwt_tun_info(a), lwt_tun_info(b),
310 		      sizeof(struct ip_tunnel_info));
311 }
312 
313 static const struct lwtunnel_encap_ops ip_tun_lwt_ops = {
314 	.build_state = ip_tun_build_state,
315 	.fill_encap = ip_tun_fill_encap_info,
316 	.get_encap_size = ip_tun_encap_nlsize,
317 	.cmp_encap = ip_tun_cmp_encap,
318 	.owner = THIS_MODULE,
319 };
320 
321 static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = {
322 	[LWTUNNEL_IP6_ID]		= { .type = NLA_U64 },
323 	[LWTUNNEL_IP6_DST]		= { .len = sizeof(struct in6_addr) },
324 	[LWTUNNEL_IP6_SRC]		= { .len = sizeof(struct in6_addr) },
325 	[LWTUNNEL_IP6_HOPLIMIT]		= { .type = NLA_U8 },
326 	[LWTUNNEL_IP6_TC]		= { .type = NLA_U8 },
327 	[LWTUNNEL_IP6_FLAGS]		= { .type = NLA_U16 },
328 };
329 
330 static int ip6_tun_build_state(struct nlattr *attr,
331 			       unsigned int family, const void *cfg,
332 			       struct lwtunnel_state **ts,
333 			       struct netlink_ext_ack *extack)
334 {
335 	struct ip_tunnel_info *tun_info;
336 	struct lwtunnel_state *new_state;
337 	struct nlattr *tb[LWTUNNEL_IP6_MAX + 1];
338 	int err;
339 
340 	err = nla_parse_nested(tb, LWTUNNEL_IP6_MAX, attr, ip6_tun_policy,
341 			       extack);
342 	if (err < 0)
343 		return err;
344 
345 	new_state = lwtunnel_state_alloc(sizeof(*tun_info));
346 	if (!new_state)
347 		return -ENOMEM;
348 
349 	new_state->type = LWTUNNEL_ENCAP_IP6;
350 
351 	tun_info = lwt_tun_info(new_state);
352 
353 	if (tb[LWTUNNEL_IP6_ID])
354 		tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]);
355 
356 	if (tb[LWTUNNEL_IP6_DST])
357 		tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]);
358 
359 	if (tb[LWTUNNEL_IP6_SRC])
360 		tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]);
361 
362 	if (tb[LWTUNNEL_IP6_HOPLIMIT])
363 		tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]);
364 
365 	if (tb[LWTUNNEL_IP6_TC])
366 		tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]);
367 
368 	if (tb[LWTUNNEL_IP6_FLAGS])
369 		tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]);
370 
371 	tun_info->mode = IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6;
372 	tun_info->options_len = 0;
373 
374 	*ts = new_state;
375 
376 	return 0;
377 }
378 
379 static int ip6_tun_fill_encap_info(struct sk_buff *skb,
380 				   struct lwtunnel_state *lwtstate)
381 {
382 	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
383 
384 	if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id,
385 			 LWTUNNEL_IP6_PAD) ||
386 	    nla_put_in6_addr(skb, LWTUNNEL_IP6_DST, &tun_info->key.u.ipv6.dst) ||
387 	    nla_put_in6_addr(skb, LWTUNNEL_IP6_SRC, &tun_info->key.u.ipv6.src) ||
388 	    nla_put_u8(skb, LWTUNNEL_IP6_TC, tun_info->key.tos) ||
389 	    nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, tun_info->key.ttl) ||
390 	    nla_put_be16(skb, LWTUNNEL_IP6_FLAGS, tun_info->key.tun_flags))
391 		return -ENOMEM;
392 
393 	return 0;
394 }
395 
396 static int ip6_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
397 {
398 	return nla_total_size_64bit(8)	/* LWTUNNEL_IP6_ID */
399 		+ nla_total_size(16)	/* LWTUNNEL_IP6_DST */
400 		+ nla_total_size(16)	/* LWTUNNEL_IP6_SRC */
401 		+ nla_total_size(1)	/* LWTUNNEL_IP6_HOPLIMIT */
402 		+ nla_total_size(1)	/* LWTUNNEL_IP6_TC */
403 		+ nla_total_size(2);	/* LWTUNNEL_IP6_FLAGS */
404 }
405 
406 static const struct lwtunnel_encap_ops ip6_tun_lwt_ops = {
407 	.build_state = ip6_tun_build_state,
408 	.fill_encap = ip6_tun_fill_encap_info,
409 	.get_encap_size = ip6_tun_encap_nlsize,
410 	.cmp_encap = ip_tun_cmp_encap,
411 	.owner = THIS_MODULE,
412 };
413 
414 void __init ip_tunnel_core_init(void)
415 {
416 	/* If you land here, make sure whether increasing ip_tunnel_info's
417 	 * options_len is a reasonable choice with its usage in front ends
418 	 * (f.e., it's part of flow keys, etc).
419 	 */
420 	BUILD_BUG_ON(IP_TUNNEL_OPTS_MAX != 255);
421 
422 	lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP);
423 	lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6);
424 }
425 
426 DEFINE_STATIC_KEY_FALSE(ip_tunnel_metadata_cnt);
427 EXPORT_SYMBOL(ip_tunnel_metadata_cnt);
428 
429 void ip_tunnel_need_metadata(void)
430 {
431 	static_branch_inc(&ip_tunnel_metadata_cnt);
432 }
433 EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata);
434 
435 void ip_tunnel_unneed_metadata(void)
436 {
437 	static_branch_dec(&ip_tunnel_metadata_cnt);
438 }
439 EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata);
440