xref: /openbmc/linux/net/ipv4/ip_tunnel_core.c (revision a20eefae)
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 
38 const struct ip_tunnel_encap_ops __rcu *
39 		iptun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
40 EXPORT_SYMBOL(iptun_encaps);
41 
42 const struct ip6_tnl_encap_ops __rcu *
43 		ip6tun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
44 EXPORT_SYMBOL(ip6tun_encaps);
45 
46 void iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb,
47 		   __be32 src, __be32 dst, __u8 proto,
48 		   __u8 tos, __u8 ttl, __be16 df, bool xnet)
49 {
50 	int pkt_len = skb->len - skb_inner_network_offset(skb);
51 	struct net *net = dev_net(rt->dst.dev);
52 	struct net_device *dev = skb->dev;
53 	struct iphdr *iph;
54 	int err;
55 
56 	skb_scrub_packet(skb, xnet);
57 
58 	skb_clear_hash_if_not_l4(skb);
59 	skb_dst_set(skb, &rt->dst);
60 	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
61 
62 	/* Push down and install the IP header. */
63 	skb_push(skb, sizeof(struct iphdr));
64 	skb_reset_network_header(skb);
65 
66 	iph = ip_hdr(skb);
67 
68 	iph->version	=	4;
69 	iph->ihl	=	sizeof(struct iphdr) >> 2;
70 	iph->frag_off	=	ip_mtu_locked(&rt->dst) ? 0 : df;
71 	iph->protocol	=	proto;
72 	iph->tos	=	tos;
73 	iph->daddr	=	dst;
74 	iph->saddr	=	src;
75 	iph->ttl	=	ttl;
76 	__ip_select_ident(net, iph, skb_shinfo(skb)->gso_segs ?: 1);
77 
78 	err = ip_local_out(net, sk, skb);
79 
80 	if (dev) {
81 		if (unlikely(net_xmit_eval(err)))
82 			pkt_len = 0;
83 		iptunnel_xmit_stats(dev, pkt_len);
84 	}
85 }
86 EXPORT_SYMBOL_GPL(iptunnel_xmit);
87 
88 int __iptunnel_pull_header(struct sk_buff *skb, int hdr_len,
89 			   __be16 inner_proto, bool raw_proto, bool xnet)
90 {
91 	if (unlikely(!pskb_may_pull(skb, hdr_len)))
92 		return -ENOMEM;
93 
94 	skb_pull_rcsum(skb, hdr_len);
95 
96 	if (!raw_proto && inner_proto == htons(ETH_P_TEB)) {
97 		struct ethhdr *eh;
98 
99 		if (unlikely(!pskb_may_pull(skb, ETH_HLEN)))
100 			return -ENOMEM;
101 
102 		eh = (struct ethhdr *)skb->data;
103 		if (likely(eth_proto_is_802_3(eh->h_proto)))
104 			skb->protocol = eh->h_proto;
105 		else
106 			skb->protocol = htons(ETH_P_802_2);
107 
108 	} else {
109 		skb->protocol = inner_proto;
110 	}
111 
112 	skb_clear_hash_if_not_l4(skb);
113 	__vlan_hwaccel_clear_tag(skb);
114 	skb_set_queue_mapping(skb, 0);
115 	skb_scrub_packet(skb, xnet);
116 
117 	return iptunnel_pull_offloads(skb);
118 }
119 EXPORT_SYMBOL_GPL(__iptunnel_pull_header);
120 
121 struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md,
122 					     gfp_t flags)
123 {
124 	struct metadata_dst *res;
125 	struct ip_tunnel_info *dst, *src;
126 
127 	if (!md || md->type != METADATA_IP_TUNNEL ||
128 	    md->u.tun_info.mode & IP_TUNNEL_INFO_TX)
129 
130 		return NULL;
131 
132 	res = metadata_dst_alloc(0, METADATA_IP_TUNNEL, flags);
133 	if (!res)
134 		return NULL;
135 
136 	dst = &res->u.tun_info;
137 	src = &md->u.tun_info;
138 	dst->key.tun_id = src->key.tun_id;
139 	if (src->mode & IP_TUNNEL_INFO_IPV6)
140 		memcpy(&dst->key.u.ipv6.dst, &src->key.u.ipv6.src,
141 		       sizeof(struct in6_addr));
142 	else
143 		dst->key.u.ipv4.dst = src->key.u.ipv4.src;
144 	dst->key.tun_flags = src->key.tun_flags;
145 	dst->mode = src->mode | IP_TUNNEL_INFO_TX;
146 
147 	return res;
148 }
149 EXPORT_SYMBOL_GPL(iptunnel_metadata_reply);
150 
151 int iptunnel_handle_offloads(struct sk_buff *skb,
152 			     int gso_type_mask)
153 {
154 	int err;
155 
156 	if (likely(!skb->encapsulation)) {
157 		skb_reset_inner_headers(skb);
158 		skb->encapsulation = 1;
159 	}
160 
161 	if (skb_is_gso(skb)) {
162 		err = skb_header_unclone(skb, GFP_ATOMIC);
163 		if (unlikely(err))
164 			return err;
165 		skb_shinfo(skb)->gso_type |= gso_type_mask;
166 		return 0;
167 	}
168 
169 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
170 		skb->ip_summed = CHECKSUM_NONE;
171 		/* We clear encapsulation here to prevent badly-written
172 		 * drivers potentially deciding to offload an inner checksum
173 		 * if we set CHECKSUM_PARTIAL on the outer header.
174 		 * This should go away when the drivers are all fixed.
175 		 */
176 		skb->encapsulation = 0;
177 	}
178 
179 	return 0;
180 }
181 EXPORT_SYMBOL_GPL(iptunnel_handle_offloads);
182 
183 /* Often modified stats are per cpu, other are shared (netdev->stats) */
184 void ip_tunnel_get_stats64(struct net_device *dev,
185 			   struct rtnl_link_stats64 *tot)
186 {
187 	int i;
188 
189 	netdev_stats_to_stats64(tot, &dev->stats);
190 
191 	for_each_possible_cpu(i) {
192 		const struct pcpu_sw_netstats *tstats =
193 						   per_cpu_ptr(dev->tstats, i);
194 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
195 		unsigned int start;
196 
197 		do {
198 			start = u64_stats_fetch_begin_irq(&tstats->syncp);
199 			rx_packets = tstats->rx_packets;
200 			tx_packets = tstats->tx_packets;
201 			rx_bytes = tstats->rx_bytes;
202 			tx_bytes = tstats->tx_bytes;
203 		} while (u64_stats_fetch_retry_irq(&tstats->syncp, start));
204 
205 		tot->rx_packets += rx_packets;
206 		tot->tx_packets += tx_packets;
207 		tot->rx_bytes   += rx_bytes;
208 		tot->tx_bytes   += tx_bytes;
209 	}
210 }
211 EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64);
212 
213 static const struct nla_policy ip_tun_policy[LWTUNNEL_IP_MAX + 1] = {
214 	[LWTUNNEL_IP_ID]	= { .type = NLA_U64 },
215 	[LWTUNNEL_IP_DST]	= { .type = NLA_U32 },
216 	[LWTUNNEL_IP_SRC]	= { .type = NLA_U32 },
217 	[LWTUNNEL_IP_TTL]	= { .type = NLA_U8 },
218 	[LWTUNNEL_IP_TOS]	= { .type = NLA_U8 },
219 	[LWTUNNEL_IP_FLAGS]	= { .type = NLA_U16 },
220 };
221 
222 static int ip_tun_build_state(struct nlattr *attr,
223 			      unsigned int family, const void *cfg,
224 			      struct lwtunnel_state **ts,
225 			      struct netlink_ext_ack *extack)
226 {
227 	struct ip_tunnel_info *tun_info;
228 	struct lwtunnel_state *new_state;
229 	struct nlattr *tb[LWTUNNEL_IP_MAX + 1];
230 	int err;
231 
232 	err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP_MAX, attr,
233 					  ip_tun_policy, extack);
234 	if (err < 0)
235 		return err;
236 
237 	new_state = lwtunnel_state_alloc(sizeof(*tun_info));
238 	if (!new_state)
239 		return -ENOMEM;
240 
241 	new_state->type = LWTUNNEL_ENCAP_IP;
242 
243 	tun_info = lwt_tun_info(new_state);
244 
245 #ifdef CONFIG_DST_CACHE
246 	err = dst_cache_init(&tun_info->dst_cache, GFP_KERNEL);
247 	if (err) {
248 		lwtstate_free(new_state);
249 		return err;
250 	}
251 #endif
252 
253 	if (tb[LWTUNNEL_IP_ID])
254 		tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]);
255 
256 	if (tb[LWTUNNEL_IP_DST])
257 		tun_info->key.u.ipv4.dst = nla_get_in_addr(tb[LWTUNNEL_IP_DST]);
258 
259 	if (tb[LWTUNNEL_IP_SRC])
260 		tun_info->key.u.ipv4.src = nla_get_in_addr(tb[LWTUNNEL_IP_SRC]);
261 
262 	if (tb[LWTUNNEL_IP_TTL])
263 		tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]);
264 
265 	if (tb[LWTUNNEL_IP_TOS])
266 		tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]);
267 
268 	if (tb[LWTUNNEL_IP_FLAGS])
269 		tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP_FLAGS]);
270 
271 	tun_info->mode = IP_TUNNEL_INFO_TX;
272 	tun_info->options_len = 0;
273 
274 	*ts = new_state;
275 
276 	return 0;
277 }
278 
279 static void ip_tun_destroy_state(struct lwtunnel_state *lwtstate)
280 {
281 #ifdef CONFIG_DST_CACHE
282 	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
283 
284 	dst_cache_destroy(&tun_info->dst_cache);
285 #endif
286 }
287 
288 static int ip_tun_fill_encap_info(struct sk_buff *skb,
289 				  struct lwtunnel_state *lwtstate)
290 {
291 	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
292 
293 	if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id,
294 			 LWTUNNEL_IP_PAD) ||
295 	    nla_put_in_addr(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) ||
296 	    nla_put_in_addr(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) ||
297 	    nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) ||
298 	    nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) ||
299 	    nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags))
300 		return -ENOMEM;
301 
302 	return 0;
303 }
304 
305 static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
306 {
307 	return nla_total_size_64bit(8)	/* LWTUNNEL_IP_ID */
308 		+ nla_total_size(4)	/* LWTUNNEL_IP_DST */
309 		+ nla_total_size(4)	/* LWTUNNEL_IP_SRC */
310 		+ nla_total_size(1)	/* LWTUNNEL_IP_TOS */
311 		+ nla_total_size(1)	/* LWTUNNEL_IP_TTL */
312 		+ nla_total_size(2);	/* LWTUNNEL_IP_FLAGS */
313 }
314 
315 static int ip_tun_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
316 {
317 	return memcmp(lwt_tun_info(a), lwt_tun_info(b),
318 		      sizeof(struct ip_tunnel_info));
319 }
320 
321 static const struct lwtunnel_encap_ops ip_tun_lwt_ops = {
322 	.build_state = ip_tun_build_state,
323 	.destroy_state = ip_tun_destroy_state,
324 	.fill_encap = ip_tun_fill_encap_info,
325 	.get_encap_size = ip_tun_encap_nlsize,
326 	.cmp_encap = ip_tun_cmp_encap,
327 	.owner = THIS_MODULE,
328 };
329 
330 static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = {
331 	[LWTUNNEL_IP6_ID]		= { .type = NLA_U64 },
332 	[LWTUNNEL_IP6_DST]		= { .len = sizeof(struct in6_addr) },
333 	[LWTUNNEL_IP6_SRC]		= { .len = sizeof(struct in6_addr) },
334 	[LWTUNNEL_IP6_HOPLIMIT]		= { .type = NLA_U8 },
335 	[LWTUNNEL_IP6_TC]		= { .type = NLA_U8 },
336 	[LWTUNNEL_IP6_FLAGS]		= { .type = NLA_U16 },
337 };
338 
339 static int ip6_tun_build_state(struct nlattr *attr,
340 			       unsigned int family, const void *cfg,
341 			       struct lwtunnel_state **ts,
342 			       struct netlink_ext_ack *extack)
343 {
344 	struct ip_tunnel_info *tun_info;
345 	struct lwtunnel_state *new_state;
346 	struct nlattr *tb[LWTUNNEL_IP6_MAX + 1];
347 	int err;
348 
349 	err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP6_MAX, attr,
350 					  ip6_tun_policy, extack);
351 	if (err < 0)
352 		return err;
353 
354 	new_state = lwtunnel_state_alloc(sizeof(*tun_info));
355 	if (!new_state)
356 		return -ENOMEM;
357 
358 	new_state->type = LWTUNNEL_ENCAP_IP6;
359 
360 	tun_info = lwt_tun_info(new_state);
361 
362 	if (tb[LWTUNNEL_IP6_ID])
363 		tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]);
364 
365 	if (tb[LWTUNNEL_IP6_DST])
366 		tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]);
367 
368 	if (tb[LWTUNNEL_IP6_SRC])
369 		tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]);
370 
371 	if (tb[LWTUNNEL_IP6_HOPLIMIT])
372 		tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]);
373 
374 	if (tb[LWTUNNEL_IP6_TC])
375 		tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]);
376 
377 	if (tb[LWTUNNEL_IP6_FLAGS])
378 		tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]);
379 
380 	tun_info->mode = IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6;
381 	tun_info->options_len = 0;
382 
383 	*ts = new_state;
384 
385 	return 0;
386 }
387 
388 static int ip6_tun_fill_encap_info(struct sk_buff *skb,
389 				   struct lwtunnel_state *lwtstate)
390 {
391 	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
392 
393 	if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id,
394 			 LWTUNNEL_IP6_PAD) ||
395 	    nla_put_in6_addr(skb, LWTUNNEL_IP6_DST, &tun_info->key.u.ipv6.dst) ||
396 	    nla_put_in6_addr(skb, LWTUNNEL_IP6_SRC, &tun_info->key.u.ipv6.src) ||
397 	    nla_put_u8(skb, LWTUNNEL_IP6_TC, tun_info->key.tos) ||
398 	    nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, tun_info->key.ttl) ||
399 	    nla_put_be16(skb, LWTUNNEL_IP6_FLAGS, tun_info->key.tun_flags))
400 		return -ENOMEM;
401 
402 	return 0;
403 }
404 
405 static int ip6_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
406 {
407 	return nla_total_size_64bit(8)	/* LWTUNNEL_IP6_ID */
408 		+ nla_total_size(16)	/* LWTUNNEL_IP6_DST */
409 		+ nla_total_size(16)	/* LWTUNNEL_IP6_SRC */
410 		+ nla_total_size(1)	/* LWTUNNEL_IP6_HOPLIMIT */
411 		+ nla_total_size(1)	/* LWTUNNEL_IP6_TC */
412 		+ nla_total_size(2);	/* LWTUNNEL_IP6_FLAGS */
413 }
414 
415 static const struct lwtunnel_encap_ops ip6_tun_lwt_ops = {
416 	.build_state = ip6_tun_build_state,
417 	.fill_encap = ip6_tun_fill_encap_info,
418 	.get_encap_size = ip6_tun_encap_nlsize,
419 	.cmp_encap = ip_tun_cmp_encap,
420 	.owner = THIS_MODULE,
421 };
422 
423 void __init ip_tunnel_core_init(void)
424 {
425 	/* If you land here, make sure whether increasing ip_tunnel_info's
426 	 * options_len is a reasonable choice with its usage in front ends
427 	 * (f.e., it's part of flow keys, etc).
428 	 */
429 	BUILD_BUG_ON(IP_TUNNEL_OPTS_MAX != 255);
430 
431 	lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP);
432 	lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6);
433 }
434 
435 DEFINE_STATIC_KEY_FALSE(ip_tunnel_metadata_cnt);
436 EXPORT_SYMBOL(ip_tunnel_metadata_cnt);
437 
438 void ip_tunnel_need_metadata(void)
439 {
440 	static_branch_inc(&ip_tunnel_metadata_cnt);
441 }
442 EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata);
443 
444 void ip_tunnel_unneed_metadata(void)
445 {
446 	static_branch_dec(&ip_tunnel_metadata_cnt);
447 }
448 EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata);
449