1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2013 Nicira, Inc.
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 #include <linux/types.h>
9 #include <linux/kernel.h>
10 #include <linux/skbuff.h>
11 #include <linux/netdevice.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>
25 #include <net/protocol.h>
26 #include <net/ip_tunnels.h>
27 #include <net/ip6_tunnel.h>
29 #include <net/checksum.h>
30 #include <net/dsfield.h>
31 #include <net/inet_ecn.h>
33 #include <net/net_namespace.h>
34 #include <net/netns/generic.h>
35 #include <net/rtnetlink.h>
36 #include <net/dst_metadata.h>
38 const struct ip_tunnel_encap_ops __rcu *
39 iptun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
40 EXPORT_SYMBOL(iptun_encaps);
42 const struct ip6_tnl_encap_ops __rcu *
43 ip6tun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
44 EXPORT_SYMBOL(ip6tun_encaps);
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)
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;
56 skb_scrub_packet(skb, xnet);
58 skb_clear_hash_if_not_l4(skb);
59 skb_dst_set(skb, &rt->dst);
60 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
62 /* Push down and install the IP header. */
63 skb_push(skb, sizeof(struct iphdr));
64 skb_reset_network_header(skb);
69 iph->ihl = sizeof(struct iphdr) >> 2;
70 iph->frag_off = ip_mtu_locked(&rt->dst) ? 0 : df;
71 iph->protocol = proto;
76 __ip_select_ident(net, iph, skb_shinfo(skb)->gso_segs ?: 1);
78 err = ip_local_out(net, sk, skb);
79 if (unlikely(net_xmit_eval(err)))
81 iptunnel_xmit_stats(dev, pkt_len);
83 EXPORT_SYMBOL_GPL(iptunnel_xmit);
85 int __iptunnel_pull_header(struct sk_buff *skb, int hdr_len,
86 __be16 inner_proto, bool raw_proto, bool xnet)
88 if (unlikely(!pskb_may_pull(skb, hdr_len)))
91 skb_pull_rcsum(skb, hdr_len);
93 if (!raw_proto && inner_proto == htons(ETH_P_TEB)) {
96 if (unlikely(!pskb_may_pull(skb, ETH_HLEN)))
99 eh = (struct ethhdr *)skb->data;
100 if (likely(eth_proto_is_802_3(eh->h_proto)))
101 skb->protocol = eh->h_proto;
103 skb->protocol = htons(ETH_P_802_2);
106 skb->protocol = inner_proto;
109 skb_clear_hash_if_not_l4(skb);
110 __vlan_hwaccel_clear_tag(skb);
111 skb_set_queue_mapping(skb, 0);
112 skb_scrub_packet(skb, xnet);
114 return iptunnel_pull_offloads(skb);
116 EXPORT_SYMBOL_GPL(__iptunnel_pull_header);
118 struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md,
121 struct metadata_dst *res;
122 struct ip_tunnel_info *dst, *src;
124 if (!md || md->type != METADATA_IP_TUNNEL ||
125 md->u.tun_info.mode & IP_TUNNEL_INFO_TX)
129 res = metadata_dst_alloc(0, METADATA_IP_TUNNEL, flags);
133 dst = &res->u.tun_info;
134 src = &md->u.tun_info;
135 dst->key.tun_id = src->key.tun_id;
136 if (src->mode & IP_TUNNEL_INFO_IPV6)
137 memcpy(&dst->key.u.ipv6.dst, &src->key.u.ipv6.src,
138 sizeof(struct in6_addr));
140 dst->key.u.ipv4.dst = src->key.u.ipv4.src;
141 dst->key.tun_flags = src->key.tun_flags;
142 dst->mode = src->mode | IP_TUNNEL_INFO_TX;
146 EXPORT_SYMBOL_GPL(iptunnel_metadata_reply);
148 int iptunnel_handle_offloads(struct sk_buff *skb,
153 if (likely(!skb->encapsulation)) {
154 skb_reset_inner_headers(skb);
155 skb->encapsulation = 1;
158 if (skb_is_gso(skb)) {
159 err = skb_header_unclone(skb, GFP_ATOMIC);
162 skb_shinfo(skb)->gso_type |= gso_type_mask;
166 if (skb->ip_summed != CHECKSUM_PARTIAL) {
167 skb->ip_summed = CHECKSUM_NONE;
168 /* We clear encapsulation here to prevent badly-written
169 * drivers potentially deciding to offload an inner checksum
170 * if we set CHECKSUM_PARTIAL on the outer header.
171 * This should go away when the drivers are all fixed.
173 skb->encapsulation = 0;
178 EXPORT_SYMBOL_GPL(iptunnel_handle_offloads);
180 /* Often modified stats are per cpu, other are shared (netdev->stats) */
181 void ip_tunnel_get_stats64(struct net_device *dev,
182 struct rtnl_link_stats64 *tot)
186 netdev_stats_to_stats64(tot, &dev->stats);
188 for_each_possible_cpu(i) {
189 const struct pcpu_sw_netstats *tstats =
190 per_cpu_ptr(dev->tstats, i);
191 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
195 start = u64_stats_fetch_begin_irq(&tstats->syncp);
196 rx_packets = tstats->rx_packets;
197 tx_packets = tstats->tx_packets;
198 rx_bytes = tstats->rx_bytes;
199 tx_bytes = tstats->tx_bytes;
200 } while (u64_stats_fetch_retry_irq(&tstats->syncp, start));
202 tot->rx_packets += rx_packets;
203 tot->tx_packets += tx_packets;
204 tot->rx_bytes += rx_bytes;
205 tot->tx_bytes += tx_bytes;
208 EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64);
210 static const struct nla_policy ip_tun_policy[LWTUNNEL_IP_MAX + 1] = {
211 [LWTUNNEL_IP_ID] = { .type = NLA_U64 },
212 [LWTUNNEL_IP_DST] = { .type = NLA_U32 },
213 [LWTUNNEL_IP_SRC] = { .type = NLA_U32 },
214 [LWTUNNEL_IP_TTL] = { .type = NLA_U8 },
215 [LWTUNNEL_IP_TOS] = { .type = NLA_U8 },
216 [LWTUNNEL_IP_FLAGS] = { .type = NLA_U16 },
219 static int ip_tun_build_state(struct nlattr *attr,
220 unsigned int family, const void *cfg,
221 struct lwtunnel_state **ts,
222 struct netlink_ext_ack *extack)
224 struct ip_tunnel_info *tun_info;
225 struct lwtunnel_state *new_state;
226 struct nlattr *tb[LWTUNNEL_IP_MAX + 1];
229 err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP_MAX, attr,
230 ip_tun_policy, extack);
234 new_state = lwtunnel_state_alloc(sizeof(*tun_info));
238 new_state->type = LWTUNNEL_ENCAP_IP;
240 tun_info = lwt_tun_info(new_state);
242 #ifdef CONFIG_DST_CACHE
243 err = dst_cache_init(&tun_info->dst_cache, GFP_KERNEL);
245 lwtstate_free(new_state);
250 if (tb[LWTUNNEL_IP_ID])
251 tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]);
253 if (tb[LWTUNNEL_IP_DST])
254 tun_info->key.u.ipv4.dst = nla_get_in_addr(tb[LWTUNNEL_IP_DST]);
256 if (tb[LWTUNNEL_IP_SRC])
257 tun_info->key.u.ipv4.src = nla_get_in_addr(tb[LWTUNNEL_IP_SRC]);
259 if (tb[LWTUNNEL_IP_TTL])
260 tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]);
262 if (tb[LWTUNNEL_IP_TOS])
263 tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]);
265 if (tb[LWTUNNEL_IP_FLAGS])
266 tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP_FLAGS]);
268 tun_info->mode = IP_TUNNEL_INFO_TX;
269 tun_info->options_len = 0;
276 static void ip_tun_destroy_state(struct lwtunnel_state *lwtstate)
278 #ifdef CONFIG_DST_CACHE
279 struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
281 dst_cache_destroy(&tun_info->dst_cache);
285 static int ip_tun_fill_encap_info(struct sk_buff *skb,
286 struct lwtunnel_state *lwtstate)
288 struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
290 if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id,
292 nla_put_in_addr(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) ||
293 nla_put_in_addr(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) ||
294 nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) ||
295 nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) ||
296 nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags))
302 static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
304 return nla_total_size_64bit(8) /* LWTUNNEL_IP_ID */
305 + nla_total_size(4) /* LWTUNNEL_IP_DST */
306 + nla_total_size(4) /* LWTUNNEL_IP_SRC */
307 + nla_total_size(1) /* LWTUNNEL_IP_TOS */
308 + nla_total_size(1) /* LWTUNNEL_IP_TTL */
309 + nla_total_size(2); /* LWTUNNEL_IP_FLAGS */
312 static int ip_tun_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
314 return memcmp(lwt_tun_info(a), lwt_tun_info(b),
315 sizeof(struct ip_tunnel_info));
318 static const struct lwtunnel_encap_ops ip_tun_lwt_ops = {
319 .build_state = ip_tun_build_state,
320 .destroy_state = ip_tun_destroy_state,
321 .fill_encap = ip_tun_fill_encap_info,
322 .get_encap_size = ip_tun_encap_nlsize,
323 .cmp_encap = ip_tun_cmp_encap,
324 .owner = THIS_MODULE,
327 static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = {
328 [LWTUNNEL_IP6_ID] = { .type = NLA_U64 },
329 [LWTUNNEL_IP6_DST] = { .len = sizeof(struct in6_addr) },
330 [LWTUNNEL_IP6_SRC] = { .len = sizeof(struct in6_addr) },
331 [LWTUNNEL_IP6_HOPLIMIT] = { .type = NLA_U8 },
332 [LWTUNNEL_IP6_TC] = { .type = NLA_U8 },
333 [LWTUNNEL_IP6_FLAGS] = { .type = NLA_U16 },
336 static int ip6_tun_build_state(struct nlattr *attr,
337 unsigned int family, const void *cfg,
338 struct lwtunnel_state **ts,
339 struct netlink_ext_ack *extack)
341 struct ip_tunnel_info *tun_info;
342 struct lwtunnel_state *new_state;
343 struct nlattr *tb[LWTUNNEL_IP6_MAX + 1];
346 err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP6_MAX, attr,
347 ip6_tun_policy, extack);
351 new_state = lwtunnel_state_alloc(sizeof(*tun_info));
355 new_state->type = LWTUNNEL_ENCAP_IP6;
357 tun_info = lwt_tun_info(new_state);
359 if (tb[LWTUNNEL_IP6_ID])
360 tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]);
362 if (tb[LWTUNNEL_IP6_DST])
363 tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]);
365 if (tb[LWTUNNEL_IP6_SRC])
366 tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]);
368 if (tb[LWTUNNEL_IP6_HOPLIMIT])
369 tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]);
371 if (tb[LWTUNNEL_IP6_TC])
372 tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]);
374 if (tb[LWTUNNEL_IP6_FLAGS])
375 tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]);
377 tun_info->mode = IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6;
378 tun_info->options_len = 0;
385 static int ip6_tun_fill_encap_info(struct sk_buff *skb,
386 struct lwtunnel_state *lwtstate)
388 struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
390 if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id,
392 nla_put_in6_addr(skb, LWTUNNEL_IP6_DST, &tun_info->key.u.ipv6.dst) ||
393 nla_put_in6_addr(skb, LWTUNNEL_IP6_SRC, &tun_info->key.u.ipv6.src) ||
394 nla_put_u8(skb, LWTUNNEL_IP6_TC, tun_info->key.tos) ||
395 nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, tun_info->key.ttl) ||
396 nla_put_be16(skb, LWTUNNEL_IP6_FLAGS, tun_info->key.tun_flags))
402 static int ip6_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
404 return nla_total_size_64bit(8) /* LWTUNNEL_IP6_ID */
405 + nla_total_size(16) /* LWTUNNEL_IP6_DST */
406 + nla_total_size(16) /* LWTUNNEL_IP6_SRC */
407 + nla_total_size(1) /* LWTUNNEL_IP6_HOPLIMIT */
408 + nla_total_size(1) /* LWTUNNEL_IP6_TC */
409 + nla_total_size(2); /* LWTUNNEL_IP6_FLAGS */
412 static const struct lwtunnel_encap_ops ip6_tun_lwt_ops = {
413 .build_state = ip6_tun_build_state,
414 .fill_encap = ip6_tun_fill_encap_info,
415 .get_encap_size = ip6_tun_encap_nlsize,
416 .cmp_encap = ip_tun_cmp_encap,
417 .owner = THIS_MODULE,
420 void __init ip_tunnel_core_init(void)
422 /* If you land here, make sure whether increasing ip_tunnel_info's
423 * options_len is a reasonable choice with its usage in front ends
424 * (f.e., it's part of flow keys, etc).
426 BUILD_BUG_ON(IP_TUNNEL_OPTS_MAX != 255);
428 lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP);
429 lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6);
432 DEFINE_STATIC_KEY_FALSE(ip_tunnel_metadata_cnt);
433 EXPORT_SYMBOL(ip_tunnel_metadata_cnt);
435 void ip_tunnel_need_metadata(void)
437 static_branch_inc(&ip_tunnel_metadata_cnt);
439 EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata);
441 void ip_tunnel_unneed_metadata(void)
443 static_branch_dec(&ip_tunnel_metadata_cnt);
445 EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata);