Loading...
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/capability.h>
22#include <linux/module.h>
23#include <linux/types.h>
24#include <linux/kernel.h>
25#include <linux/slab.h>
26#include <linux/uaccess.h>
27#include <linux/skbuff.h>
28#include <linux/netdevice.h>
29#include <linux/in.h>
30#include <linux/tcp.h>
31#include <linux/udp.h>
32#include <linux/if_arp.h>
33#include <linux/init.h>
34#include <linux/in6.h>
35#include <linux/inetdevice.h>
36#include <linux/igmp.h>
37#include <linux/netfilter_ipv4.h>
38#include <linux/etherdevice.h>
39#include <linux/if_ether.h>
40#include <linux/if_vlan.h>
41#include <linux/rculist.h>
42#include <linux/err.h>
43
44#include <net/sock.h>
45#include <net/ip.h>
46#include <net/icmp.h>
47#include <net/protocol.h>
48#include <net/ip_tunnels.h>
49#include <net/arp.h>
50#include <net/checksum.h>
51#include <net/dsfield.h>
52#include <net/inet_ecn.h>
53#include <net/xfrm.h>
54#include <net/net_namespace.h>
55#include <net/netns/generic.h>
56#include <net/rtnetlink.h>
57#include <net/udp.h>
58#include <net/dst_metadata.h>
59
60#if IS_ENABLED(CONFIG_IPV6)
61#include <net/ipv6.h>
62#include <net/ip6_fib.h>
63#include <net/ip6_route.h>
64#endif
65
66static unsigned int ip_tunnel_hash(__be32 key, __be32 remote)
67{
68 return hash_32((__force u32)key ^ (__force u32)remote,
69 IP_TNL_HASH_BITS);
70}
71
72static bool ip_tunnel_key_match(const struct ip_tunnel_parm *p,
73 __be16 flags, __be32 key)
74{
75 if (p->i_flags & TUNNEL_KEY) {
76 if (flags & TUNNEL_KEY)
77 return key == p->i_key;
78 else
79 /* key expected, none present */
80 return false;
81 } else
82 return !(flags & TUNNEL_KEY);
83}
84
85/* Fallback tunnel: no source, no destination, no key, no options
86
87 Tunnel hash table:
88 We require exact key match i.e. if a key is present in packet
89 it will match only tunnel with the same key; if it is not present,
90 it will match only keyless tunnel.
91
92 All keysless packets, if not matched configured keyless tunnels
93 will match fallback tunnel.
94 Given src, dst and key, find appropriate for input tunnel.
95*/
96struct ip_tunnel *ip_tunnel_lookup(struct ip_tunnel_net *itn,
97 int link, __be16 flags,
98 __be32 remote, __be32 local,
99 __be32 key)
100{
101 unsigned int hash;
102 struct ip_tunnel *t, *cand = NULL;
103 struct hlist_head *head;
104
105 hash = ip_tunnel_hash(key, remote);
106 head = &itn->tunnels[hash];
107
108 hlist_for_each_entry_rcu(t, head, hash_node) {
109 if (local != t->parms.iph.saddr ||
110 remote != t->parms.iph.daddr ||
111 !(t->dev->flags & IFF_UP))
112 continue;
113
114 if (!ip_tunnel_key_match(&t->parms, flags, key))
115 continue;
116
117 if (t->parms.link == link)
118 return t;
119 else
120 cand = t;
121 }
122
123 hlist_for_each_entry_rcu(t, head, hash_node) {
124 if (remote != t->parms.iph.daddr ||
125 t->parms.iph.saddr != 0 ||
126 !(t->dev->flags & IFF_UP))
127 continue;
128
129 if (!ip_tunnel_key_match(&t->parms, flags, key))
130 continue;
131
132 if (t->parms.link == link)
133 return t;
134 else if (!cand)
135 cand = t;
136 }
137
138 hash = ip_tunnel_hash(key, 0);
139 head = &itn->tunnels[hash];
140
141 hlist_for_each_entry_rcu(t, head, hash_node) {
142 if ((local != t->parms.iph.saddr || t->parms.iph.daddr != 0) &&
143 (local != t->parms.iph.daddr || !ipv4_is_multicast(local)))
144 continue;
145
146 if (!(t->dev->flags & IFF_UP))
147 continue;
148
149 if (!ip_tunnel_key_match(&t->parms, flags, key))
150 continue;
151
152 if (t->parms.link == link)
153 return t;
154 else if (!cand)
155 cand = t;
156 }
157
158 if (flags & TUNNEL_NO_KEY)
159 goto skip_key_lookup;
160
161 hlist_for_each_entry_rcu(t, head, hash_node) {
162 if (t->parms.i_key != key ||
163 t->parms.iph.saddr != 0 ||
164 t->parms.iph.daddr != 0 ||
165 !(t->dev->flags & IFF_UP))
166 continue;
167
168 if (t->parms.link == link)
169 return t;
170 else if (!cand)
171 cand = t;
172 }
173
174skip_key_lookup:
175 if (cand)
176 return cand;
177
178 t = rcu_dereference(itn->collect_md_tun);
179 if (t && t->dev->flags & IFF_UP)
180 return t;
181
182 if (itn->fb_tunnel_dev && itn->fb_tunnel_dev->flags & IFF_UP)
183 return netdev_priv(itn->fb_tunnel_dev);
184
185 return NULL;
186}
187EXPORT_SYMBOL_GPL(ip_tunnel_lookup);
188
189static struct hlist_head *ip_bucket(struct ip_tunnel_net *itn,
190 struct ip_tunnel_parm *parms)
191{
192 unsigned int h;
193 __be32 remote;
194 __be32 i_key = parms->i_key;
195
196 if (parms->iph.daddr && !ipv4_is_multicast(parms->iph.daddr))
197 remote = parms->iph.daddr;
198 else
199 remote = 0;
200
201 if (!(parms->i_flags & TUNNEL_KEY) && (parms->i_flags & VTI_ISVTI))
202 i_key = 0;
203
204 h = ip_tunnel_hash(i_key, remote);
205 return &itn->tunnels[h];
206}
207
208static void ip_tunnel_add(struct ip_tunnel_net *itn, struct ip_tunnel *t)
209{
210 struct hlist_head *head = ip_bucket(itn, &t->parms);
211
212 if (t->collect_md)
213 rcu_assign_pointer(itn->collect_md_tun, t);
214 hlist_add_head_rcu(&t->hash_node, head);
215}
216
217static void ip_tunnel_del(struct ip_tunnel_net *itn, struct ip_tunnel *t)
218{
219 if (t->collect_md)
220 rcu_assign_pointer(itn->collect_md_tun, NULL);
221 hlist_del_init_rcu(&t->hash_node);
222}
223
224static struct ip_tunnel *ip_tunnel_find(struct ip_tunnel_net *itn,
225 struct ip_tunnel_parm *parms,
226 int type)
227{
228 __be32 remote = parms->iph.daddr;
229 __be32 local = parms->iph.saddr;
230 __be32 key = parms->i_key;
231 __be16 flags = parms->i_flags;
232 int link = parms->link;
233 struct ip_tunnel *t = NULL;
234 struct hlist_head *head = ip_bucket(itn, parms);
235
236 hlist_for_each_entry_rcu(t, head, hash_node) {
237 if (local == t->parms.iph.saddr &&
238 remote == t->parms.iph.daddr &&
239 link == t->parms.link &&
240 type == t->dev->type &&
241 ip_tunnel_key_match(&t->parms, flags, key))
242 break;
243 }
244 return t;
245}
246
247static struct net_device *__ip_tunnel_create(struct net *net,
248 const struct rtnl_link_ops *ops,
249 struct ip_tunnel_parm *parms)
250{
251 int err;
252 struct ip_tunnel *tunnel;
253 struct net_device *dev;
254 char name[IFNAMSIZ];
255
256 err = -E2BIG;
257 if (parms->name[0]) {
258 if (!dev_valid_name(parms->name))
259 goto failed;
260 strlcpy(name, parms->name, IFNAMSIZ);
261 } else {
262 if (strlen(ops->kind) > (IFNAMSIZ - 3))
263 goto failed;
264 strlcpy(name, ops->kind, IFNAMSIZ);
265 strncat(name, "%d", 2);
266 }
267
268 ASSERT_RTNL();
269 dev = alloc_netdev(ops->priv_size, name, NET_NAME_UNKNOWN, ops->setup);
270 if (!dev) {
271 err = -ENOMEM;
272 goto failed;
273 }
274 dev_net_set(dev, net);
275
276 dev->rtnl_link_ops = ops;
277
278 tunnel = netdev_priv(dev);
279 tunnel->parms = *parms;
280 tunnel->net = net;
281
282 err = register_netdevice(dev);
283 if (err)
284 goto failed_free;
285
286 return dev;
287
288failed_free:
289 free_netdev(dev);
290failed:
291 return ERR_PTR(err);
292}
293
294static int ip_tunnel_bind_dev(struct net_device *dev)
295{
296 struct net_device *tdev = NULL;
297 struct ip_tunnel *tunnel = netdev_priv(dev);
298 const struct iphdr *iph;
299 int hlen = LL_MAX_HEADER;
300 int mtu = ETH_DATA_LEN;
301 int t_hlen = tunnel->hlen + sizeof(struct iphdr);
302
303 iph = &tunnel->parms.iph;
304
305 /* Guess output device to choose reasonable mtu and needed_headroom */
306 if (iph->daddr) {
307 struct flowi4 fl4;
308 struct rtable *rt;
309
310 ip_tunnel_init_flow(&fl4, iph->protocol, iph->daddr,
311 iph->saddr, tunnel->parms.o_key,
312 RT_TOS(iph->tos), tunnel->parms.link,
313 tunnel->fwmark);
314 rt = ip_route_output_key(tunnel->net, &fl4);
315
316 if (!IS_ERR(rt)) {
317 tdev = rt->dst.dev;
318 ip_rt_put(rt);
319 }
320 if (dev->type != ARPHRD_ETHER)
321 dev->flags |= IFF_POINTOPOINT;
322
323 dst_cache_reset(&tunnel->dst_cache);
324 }
325
326 if (!tdev && tunnel->parms.link)
327 tdev = __dev_get_by_index(tunnel->net, tunnel->parms.link);
328
329 if (tdev) {
330 hlen = tdev->hard_header_len + tdev->needed_headroom;
331 mtu = min(tdev->mtu, IP_MAX_MTU);
332 }
333
334 dev->needed_headroom = t_hlen + hlen;
335 mtu -= (dev->hard_header_len + t_hlen);
336
337 if (mtu < IPV4_MIN_MTU)
338 mtu = IPV4_MIN_MTU;
339
340 return mtu;
341}
342
343static struct ip_tunnel *ip_tunnel_create(struct net *net,
344 struct ip_tunnel_net *itn,
345 struct ip_tunnel_parm *parms)
346{
347 struct ip_tunnel *nt;
348 struct net_device *dev;
349 int t_hlen;
350 int mtu;
351 int err;
352
353 dev = __ip_tunnel_create(net, itn->rtnl_link_ops, parms);
354 if (IS_ERR(dev))
355 return ERR_CAST(dev);
356
357 mtu = ip_tunnel_bind_dev(dev);
358 err = dev_set_mtu(dev, mtu);
359 if (err)
360 goto err_dev_set_mtu;
361
362 nt = netdev_priv(dev);
363 t_hlen = nt->hlen + sizeof(struct iphdr);
364 dev->min_mtu = ETH_MIN_MTU;
365 dev->max_mtu = IP_MAX_MTU - dev->hard_header_len - t_hlen;
366 ip_tunnel_add(itn, nt);
367 return nt;
368
369err_dev_set_mtu:
370 unregister_netdevice(dev);
371 return ERR_PTR(err);
372}
373
374int ip_tunnel_rcv(struct ip_tunnel *tunnel, struct sk_buff *skb,
375 const struct tnl_ptk_info *tpi, struct metadata_dst *tun_dst,
376 bool log_ecn_error)
377{
378 struct pcpu_sw_netstats *tstats;
379 const struct iphdr *iph = ip_hdr(skb);
380 int err;
381
382#ifdef CONFIG_NET_IPGRE_BROADCAST
383 if (ipv4_is_multicast(iph->daddr)) {
384 tunnel->dev->stats.multicast++;
385 skb->pkt_type = PACKET_BROADCAST;
386 }
387#endif
388
389 if ((!(tpi->flags&TUNNEL_CSUM) && (tunnel->parms.i_flags&TUNNEL_CSUM)) ||
390 ((tpi->flags&TUNNEL_CSUM) && !(tunnel->parms.i_flags&TUNNEL_CSUM))) {
391 tunnel->dev->stats.rx_crc_errors++;
392 tunnel->dev->stats.rx_errors++;
393 goto drop;
394 }
395
396 if (tunnel->parms.i_flags&TUNNEL_SEQ) {
397 if (!(tpi->flags&TUNNEL_SEQ) ||
398 (tunnel->i_seqno && (s32)(ntohl(tpi->seq) - tunnel->i_seqno) < 0)) {
399 tunnel->dev->stats.rx_fifo_errors++;
400 tunnel->dev->stats.rx_errors++;
401 goto drop;
402 }
403 tunnel->i_seqno = ntohl(tpi->seq) + 1;
404 }
405
406 skb_reset_network_header(skb);
407
408 err = IP_ECN_decapsulate(iph, skb);
409 if (unlikely(err)) {
410 if (log_ecn_error)
411 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
412 &iph->saddr, iph->tos);
413 if (err > 1) {
414 ++tunnel->dev->stats.rx_frame_errors;
415 ++tunnel->dev->stats.rx_errors;
416 goto drop;
417 }
418 }
419
420 tstats = this_cpu_ptr(tunnel->dev->tstats);
421 u64_stats_update_begin(&tstats->syncp);
422 tstats->rx_packets++;
423 tstats->rx_bytes += skb->len;
424 u64_stats_update_end(&tstats->syncp);
425
426 skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(tunnel->dev)));
427
428 if (tunnel->dev->type == ARPHRD_ETHER) {
429 skb->protocol = eth_type_trans(skb, tunnel->dev);
430 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
431 } else {
432 skb->dev = tunnel->dev;
433 }
434
435 if (tun_dst)
436 skb_dst_set(skb, (struct dst_entry *)tun_dst);
437
438 gro_cells_receive(&tunnel->gro_cells, skb);
439 return 0;
440
441drop:
442 if (tun_dst)
443 dst_release((struct dst_entry *)tun_dst);
444 kfree_skb(skb);
445 return 0;
446}
447EXPORT_SYMBOL_GPL(ip_tunnel_rcv);
448
449int ip_tunnel_encap_add_ops(const struct ip_tunnel_encap_ops *ops,
450 unsigned int num)
451{
452 if (num >= MAX_IPTUN_ENCAP_OPS)
453 return -ERANGE;
454
455 return !cmpxchg((const struct ip_tunnel_encap_ops **)
456 &iptun_encaps[num],
457 NULL, ops) ? 0 : -1;
458}
459EXPORT_SYMBOL(ip_tunnel_encap_add_ops);
460
461int ip_tunnel_encap_del_ops(const struct ip_tunnel_encap_ops *ops,
462 unsigned int num)
463{
464 int ret;
465
466 if (num >= MAX_IPTUN_ENCAP_OPS)
467 return -ERANGE;
468
469 ret = (cmpxchg((const struct ip_tunnel_encap_ops **)
470 &iptun_encaps[num],
471 ops, NULL) == ops) ? 0 : -1;
472
473 synchronize_net();
474
475 return ret;
476}
477EXPORT_SYMBOL(ip_tunnel_encap_del_ops);
478
479int ip_tunnel_encap_setup(struct ip_tunnel *t,
480 struct ip_tunnel_encap *ipencap)
481{
482 int hlen;
483
484 memset(&t->encap, 0, sizeof(t->encap));
485
486 hlen = ip_encap_hlen(ipencap);
487 if (hlen < 0)
488 return hlen;
489
490 t->encap.type = ipencap->type;
491 t->encap.sport = ipencap->sport;
492 t->encap.dport = ipencap->dport;
493 t->encap.flags = ipencap->flags;
494
495 t->encap_hlen = hlen;
496 t->hlen = t->encap_hlen + t->tun_hlen;
497
498 return 0;
499}
500EXPORT_SYMBOL_GPL(ip_tunnel_encap_setup);
501
502static int tnl_update_pmtu(struct net_device *dev, struct sk_buff *skb,
503 struct rtable *rt, __be16 df,
504 const struct iphdr *inner_iph)
505{
506 struct ip_tunnel *tunnel = netdev_priv(dev);
507 int pkt_size = skb->len - tunnel->hlen - dev->hard_header_len;
508 int mtu;
509
510 if (df)
511 mtu = dst_mtu(&rt->dst) - dev->hard_header_len
512 - sizeof(struct iphdr) - tunnel->hlen;
513 else
514 mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
515
516 skb_dst_update_pmtu(skb, mtu);
517
518 if (skb->protocol == htons(ETH_P_IP)) {
519 if (!skb_is_gso(skb) &&
520 (inner_iph->frag_off & htons(IP_DF)) &&
521 mtu < pkt_size) {
522 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
523 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
524 return -E2BIG;
525 }
526 }
527#if IS_ENABLED(CONFIG_IPV6)
528 else if (skb->protocol == htons(ETH_P_IPV6)) {
529 struct rt6_info *rt6 = (struct rt6_info *)skb_dst(skb);
530
531 if (rt6 && mtu < dst_mtu(skb_dst(skb)) &&
532 mtu >= IPV6_MIN_MTU) {
533 if ((tunnel->parms.iph.daddr &&
534 !ipv4_is_multicast(tunnel->parms.iph.daddr)) ||
535 rt6->rt6i_dst.plen == 128) {
536 rt6->rt6i_flags |= RTF_MODIFIED;
537 dst_metric_set(skb_dst(skb), RTAX_MTU, mtu);
538 }
539 }
540
541 if (!skb_is_gso(skb) && mtu >= IPV6_MIN_MTU &&
542 mtu < pkt_size) {
543 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
544 return -E2BIG;
545 }
546 }
547#endif
548 return 0;
549}
550
551void ip_md_tunnel_xmit(struct sk_buff *skb, struct net_device *dev, u8 proto)
552{
553 struct ip_tunnel *tunnel = netdev_priv(dev);
554 u32 headroom = sizeof(struct iphdr);
555 struct ip_tunnel_info *tun_info;
556 const struct ip_tunnel_key *key;
557 const struct iphdr *inner_iph;
558 struct rtable *rt;
559 struct flowi4 fl4;
560 __be16 df = 0;
561 u8 tos, ttl;
562
563 tun_info = skb_tunnel_info(skb);
564 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
565 ip_tunnel_info_af(tun_info) != AF_INET))
566 goto tx_error;
567 key = &tun_info->key;
568 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
569 inner_iph = (const struct iphdr *)skb_inner_network_header(skb);
570 tos = key->tos;
571 if (tos == 1) {
572 if (skb->protocol == htons(ETH_P_IP))
573 tos = inner_iph->tos;
574 else if (skb->protocol == htons(ETH_P_IPV6))
575 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph);
576 }
577 ip_tunnel_init_flow(&fl4, proto, key->u.ipv4.dst, key->u.ipv4.src, 0,
578 RT_TOS(tos), tunnel->parms.link, tunnel->fwmark);
579 if (tunnel->encap.type != TUNNEL_ENCAP_NONE)
580 goto tx_error;
581 rt = ip_route_output_key(tunnel->net, &fl4);
582 if (IS_ERR(rt)) {
583 dev->stats.tx_carrier_errors++;
584 goto tx_error;
585 }
586 if (rt->dst.dev == dev) {
587 ip_rt_put(rt);
588 dev->stats.collisions++;
589 goto tx_error;
590 }
591 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb);
592 ttl = key->ttl;
593 if (ttl == 0) {
594 if (skb->protocol == htons(ETH_P_IP))
595 ttl = inner_iph->ttl;
596 else if (skb->protocol == htons(ETH_P_IPV6))
597 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit;
598 else
599 ttl = ip4_dst_hoplimit(&rt->dst);
600 }
601 if (key->tun_flags & TUNNEL_DONT_FRAGMENT)
602 df = htons(IP_DF);
603 else if (skb->protocol == htons(ETH_P_IP))
604 df = inner_iph->frag_off & htons(IP_DF);
605 headroom += LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len;
606 if (headroom > dev->needed_headroom)
607 dev->needed_headroom = headroom;
608
609 if (skb_cow_head(skb, dev->needed_headroom)) {
610 ip_rt_put(rt);
611 goto tx_dropped;
612 }
613 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, proto, tos, ttl,
614 df, !net_eq(tunnel->net, dev_net(dev)));
615 return;
616tx_error:
617 dev->stats.tx_errors++;
618 goto kfree;
619tx_dropped:
620 dev->stats.tx_dropped++;
621kfree:
622 kfree_skb(skb);
623}
624EXPORT_SYMBOL_GPL(ip_md_tunnel_xmit);
625
626void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev,
627 const struct iphdr *tnl_params, u8 protocol)
628{
629 struct ip_tunnel *tunnel = netdev_priv(dev);
630 const struct iphdr *inner_iph;
631 struct flowi4 fl4;
632 u8 tos, ttl;
633 __be16 df;
634 struct rtable *rt; /* Route to the other host */
635 unsigned int max_headroom; /* The extra header space needed */
636 __be32 dst;
637 bool connected;
638
639 inner_iph = (const struct iphdr *)skb_inner_network_header(skb);
640 connected = (tunnel->parms.iph.daddr != 0);
641
642 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
643
644 dst = tnl_params->daddr;
645 if (dst == 0) {
646 /* NBMA tunnel */
647
648 if (!skb_dst(skb)) {
649 dev->stats.tx_fifo_errors++;
650 goto tx_error;
651 }
652
653 if (skb->protocol == htons(ETH_P_IP)) {
654 rt = skb_rtable(skb);
655 dst = rt_nexthop(rt, inner_iph->daddr);
656 }
657#if IS_ENABLED(CONFIG_IPV6)
658 else if (skb->protocol == htons(ETH_P_IPV6)) {
659 const struct in6_addr *addr6;
660 struct neighbour *neigh;
661 bool do_tx_error_icmp;
662 int addr_type;
663
664 neigh = dst_neigh_lookup(skb_dst(skb),
665 &ipv6_hdr(skb)->daddr);
666 if (!neigh)
667 goto tx_error;
668
669 addr6 = (const struct in6_addr *)&neigh->primary_key;
670 addr_type = ipv6_addr_type(addr6);
671
672 if (addr_type == IPV6_ADDR_ANY) {
673 addr6 = &ipv6_hdr(skb)->daddr;
674 addr_type = ipv6_addr_type(addr6);
675 }
676
677 if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
678 do_tx_error_icmp = true;
679 else {
680 do_tx_error_icmp = false;
681 dst = addr6->s6_addr32[3];
682 }
683 neigh_release(neigh);
684 if (do_tx_error_icmp)
685 goto tx_error_icmp;
686 }
687#endif
688 else
689 goto tx_error;
690
691 connected = false;
692 }
693
694 tos = tnl_params->tos;
695 if (tos & 0x1) {
696 tos &= ~0x1;
697 if (skb->protocol == htons(ETH_P_IP)) {
698 tos = inner_iph->tos;
699 connected = false;
700 } else if (skb->protocol == htons(ETH_P_IPV6)) {
701 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph);
702 connected = false;
703 }
704 }
705
706 ip_tunnel_init_flow(&fl4, protocol, dst, tnl_params->saddr,
707 tunnel->parms.o_key, RT_TOS(tos), tunnel->parms.link,
708 tunnel->fwmark);
709
710 if (ip_tunnel_encap(skb, tunnel, &protocol, &fl4) < 0)
711 goto tx_error;
712
713 rt = connected ? dst_cache_get_ip4(&tunnel->dst_cache, &fl4.saddr) :
714 NULL;
715
716 if (!rt) {
717 rt = ip_route_output_key(tunnel->net, &fl4);
718
719 if (IS_ERR(rt)) {
720 dev->stats.tx_carrier_errors++;
721 goto tx_error;
722 }
723 if (connected)
724 dst_cache_set_ip4(&tunnel->dst_cache, &rt->dst,
725 fl4.saddr);
726 }
727
728 if (rt->dst.dev == dev) {
729 ip_rt_put(rt);
730 dev->stats.collisions++;
731 goto tx_error;
732 }
733
734 if (tnl_update_pmtu(dev, skb, rt, tnl_params->frag_off, inner_iph)) {
735 ip_rt_put(rt);
736 goto tx_error;
737 }
738
739 if (tunnel->err_count > 0) {
740 if (time_before(jiffies,
741 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
742 tunnel->err_count--;
743
744 dst_link_failure(skb);
745 } else
746 tunnel->err_count = 0;
747 }
748
749 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb);
750 ttl = tnl_params->ttl;
751 if (ttl == 0) {
752 if (skb->protocol == htons(ETH_P_IP))
753 ttl = inner_iph->ttl;
754#if IS_ENABLED(CONFIG_IPV6)
755 else if (skb->protocol == htons(ETH_P_IPV6))
756 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit;
757#endif
758 else
759 ttl = ip4_dst_hoplimit(&rt->dst);
760 }
761
762 df = tnl_params->frag_off;
763 if (skb->protocol == htons(ETH_P_IP) && !tunnel->ignore_df)
764 df |= (inner_iph->frag_off&htons(IP_DF));
765
766 max_headroom = LL_RESERVED_SPACE(rt->dst.dev) + sizeof(struct iphdr)
767 + rt->dst.header_len + ip_encap_hlen(&tunnel->encap);
768 if (max_headroom > dev->needed_headroom)
769 dev->needed_headroom = max_headroom;
770
771 if (skb_cow_head(skb, dev->needed_headroom)) {
772 ip_rt_put(rt);
773 dev->stats.tx_dropped++;
774 kfree_skb(skb);
775 return;
776 }
777
778 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, protocol, tos, ttl,
779 df, !net_eq(tunnel->net, dev_net(dev)));
780 return;
781
782#if IS_ENABLED(CONFIG_IPV6)
783tx_error_icmp:
784 dst_link_failure(skb);
785#endif
786tx_error:
787 dev->stats.tx_errors++;
788 kfree_skb(skb);
789}
790EXPORT_SYMBOL_GPL(ip_tunnel_xmit);
791
792static void ip_tunnel_update(struct ip_tunnel_net *itn,
793 struct ip_tunnel *t,
794 struct net_device *dev,
795 struct ip_tunnel_parm *p,
796 bool set_mtu,
797 __u32 fwmark)
798{
799 ip_tunnel_del(itn, t);
800 t->parms.iph.saddr = p->iph.saddr;
801 t->parms.iph.daddr = p->iph.daddr;
802 t->parms.i_key = p->i_key;
803 t->parms.o_key = p->o_key;
804 if (dev->type != ARPHRD_ETHER) {
805 memcpy(dev->dev_addr, &p->iph.saddr, 4);
806 memcpy(dev->broadcast, &p->iph.daddr, 4);
807 }
808 ip_tunnel_add(itn, t);
809
810 t->parms.iph.ttl = p->iph.ttl;
811 t->parms.iph.tos = p->iph.tos;
812 t->parms.iph.frag_off = p->iph.frag_off;
813
814 if (t->parms.link != p->link || t->fwmark != fwmark) {
815 int mtu;
816
817 t->parms.link = p->link;
818 t->fwmark = fwmark;
819 mtu = ip_tunnel_bind_dev(dev);
820 if (set_mtu)
821 dev->mtu = mtu;
822 }
823 dst_cache_reset(&t->dst_cache);
824 netdev_state_change(dev);
825}
826
827int ip_tunnel_ioctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd)
828{
829 int err = 0;
830 struct ip_tunnel *t = netdev_priv(dev);
831 struct net *net = t->net;
832 struct ip_tunnel_net *itn = net_generic(net, t->ip_tnl_net_id);
833
834 switch (cmd) {
835 case SIOCGETTUNNEL:
836 if (dev == itn->fb_tunnel_dev) {
837 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type);
838 if (!t)
839 t = netdev_priv(dev);
840 }
841 memcpy(p, &t->parms, sizeof(*p));
842 break;
843
844 case SIOCADDTUNNEL:
845 case SIOCCHGTUNNEL:
846 err = -EPERM;
847 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
848 goto done;
849 if (p->iph.ttl)
850 p->iph.frag_off |= htons(IP_DF);
851 if (!(p->i_flags & VTI_ISVTI)) {
852 if (!(p->i_flags & TUNNEL_KEY))
853 p->i_key = 0;
854 if (!(p->o_flags & TUNNEL_KEY))
855 p->o_key = 0;
856 }
857
858 t = ip_tunnel_find(itn, p, itn->type);
859
860 if (cmd == SIOCADDTUNNEL) {
861 if (!t) {
862 t = ip_tunnel_create(net, itn, p);
863 err = PTR_ERR_OR_ZERO(t);
864 break;
865 }
866
867 err = -EEXIST;
868 break;
869 }
870 if (dev != itn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
871 if (t) {
872 if (t->dev != dev) {
873 err = -EEXIST;
874 break;
875 }
876 } else {
877 unsigned int nflags = 0;
878
879 if (ipv4_is_multicast(p->iph.daddr))
880 nflags = IFF_BROADCAST;
881 else if (p->iph.daddr)
882 nflags = IFF_POINTOPOINT;
883
884 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
885 err = -EINVAL;
886 break;
887 }
888
889 t = netdev_priv(dev);
890 }
891 }
892
893 if (t) {
894 err = 0;
895 ip_tunnel_update(itn, t, dev, p, true, 0);
896 } else {
897 err = -ENOENT;
898 }
899 break;
900
901 case SIOCDELTUNNEL:
902 err = -EPERM;
903 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
904 goto done;
905
906 if (dev == itn->fb_tunnel_dev) {
907 err = -ENOENT;
908 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type);
909 if (!t)
910 goto done;
911 err = -EPERM;
912 if (t == netdev_priv(itn->fb_tunnel_dev))
913 goto done;
914 dev = t->dev;
915 }
916 unregister_netdevice(dev);
917 err = 0;
918 break;
919
920 default:
921 err = -EINVAL;
922 }
923
924done:
925 return err;
926}
927EXPORT_SYMBOL_GPL(ip_tunnel_ioctl);
928
929int __ip_tunnel_change_mtu(struct net_device *dev, int new_mtu, bool strict)
930{
931 struct ip_tunnel *tunnel = netdev_priv(dev);
932 int t_hlen = tunnel->hlen + sizeof(struct iphdr);
933 int max_mtu = IP_MAX_MTU - dev->hard_header_len - t_hlen;
934
935 if (new_mtu < ETH_MIN_MTU)
936 return -EINVAL;
937
938 if (new_mtu > max_mtu) {
939 if (strict)
940 return -EINVAL;
941
942 new_mtu = max_mtu;
943 }
944
945 dev->mtu = new_mtu;
946 return 0;
947}
948EXPORT_SYMBOL_GPL(__ip_tunnel_change_mtu);
949
950int ip_tunnel_change_mtu(struct net_device *dev, int new_mtu)
951{
952 return __ip_tunnel_change_mtu(dev, new_mtu, true);
953}
954EXPORT_SYMBOL_GPL(ip_tunnel_change_mtu);
955
956static void ip_tunnel_dev_free(struct net_device *dev)
957{
958 struct ip_tunnel *tunnel = netdev_priv(dev);
959
960 gro_cells_destroy(&tunnel->gro_cells);
961 dst_cache_destroy(&tunnel->dst_cache);
962 free_percpu(dev->tstats);
963}
964
965void ip_tunnel_dellink(struct net_device *dev, struct list_head *head)
966{
967 struct ip_tunnel *tunnel = netdev_priv(dev);
968 struct ip_tunnel_net *itn;
969
970 itn = net_generic(tunnel->net, tunnel->ip_tnl_net_id);
971
972 if (itn->fb_tunnel_dev != dev) {
973 ip_tunnel_del(itn, netdev_priv(dev));
974 unregister_netdevice_queue(dev, head);
975 }
976}
977EXPORT_SYMBOL_GPL(ip_tunnel_dellink);
978
979struct net *ip_tunnel_get_link_net(const struct net_device *dev)
980{
981 struct ip_tunnel *tunnel = netdev_priv(dev);
982
983 return tunnel->net;
984}
985EXPORT_SYMBOL(ip_tunnel_get_link_net);
986
987int ip_tunnel_get_iflink(const struct net_device *dev)
988{
989 struct ip_tunnel *tunnel = netdev_priv(dev);
990
991 return tunnel->parms.link;
992}
993EXPORT_SYMBOL(ip_tunnel_get_iflink);
994
995int ip_tunnel_init_net(struct net *net, unsigned int ip_tnl_net_id,
996 struct rtnl_link_ops *ops, char *devname)
997{
998 struct ip_tunnel_net *itn = net_generic(net, ip_tnl_net_id);
999 struct ip_tunnel_parm parms;
1000 unsigned int i;
1001
1002 itn->rtnl_link_ops = ops;
1003 for (i = 0; i < IP_TNL_HASH_SIZE; i++)
1004 INIT_HLIST_HEAD(&itn->tunnels[i]);
1005
1006 if (!ops || !net_has_fallback_tunnels(net)) {
1007 struct ip_tunnel_net *it_init_net;
1008
1009 it_init_net = net_generic(&init_net, ip_tnl_net_id);
1010 itn->type = it_init_net->type;
1011 itn->fb_tunnel_dev = NULL;
1012 return 0;
1013 }
1014
1015 memset(&parms, 0, sizeof(parms));
1016 if (devname)
1017 strlcpy(parms.name, devname, IFNAMSIZ);
1018
1019 rtnl_lock();
1020 itn->fb_tunnel_dev = __ip_tunnel_create(net, ops, &parms);
1021 /* FB netdevice is special: we have one, and only one per netns.
1022 * Allowing to move it to another netns is clearly unsafe.
1023 */
1024 if (!IS_ERR(itn->fb_tunnel_dev)) {
1025 itn->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1026 itn->fb_tunnel_dev->mtu = ip_tunnel_bind_dev(itn->fb_tunnel_dev);
1027 ip_tunnel_add(itn, netdev_priv(itn->fb_tunnel_dev));
1028 itn->type = itn->fb_tunnel_dev->type;
1029 }
1030 rtnl_unlock();
1031
1032 return PTR_ERR_OR_ZERO(itn->fb_tunnel_dev);
1033}
1034EXPORT_SYMBOL_GPL(ip_tunnel_init_net);
1035
1036static void ip_tunnel_destroy(struct net *net, struct ip_tunnel_net *itn,
1037 struct list_head *head,
1038 struct rtnl_link_ops *ops)
1039{
1040 struct net_device *dev, *aux;
1041 int h;
1042
1043 for_each_netdev_safe(net, dev, aux)
1044 if (dev->rtnl_link_ops == ops)
1045 unregister_netdevice_queue(dev, head);
1046
1047 for (h = 0; h < IP_TNL_HASH_SIZE; h++) {
1048 struct ip_tunnel *t;
1049 struct hlist_node *n;
1050 struct hlist_head *thead = &itn->tunnels[h];
1051
1052 hlist_for_each_entry_safe(t, n, thead, hash_node)
1053 /* If dev is in the same netns, it has already
1054 * been added to the list by the previous loop.
1055 */
1056 if (!net_eq(dev_net(t->dev), net))
1057 unregister_netdevice_queue(t->dev, head);
1058 }
1059}
1060
1061void ip_tunnel_delete_nets(struct list_head *net_list, unsigned int id,
1062 struct rtnl_link_ops *ops)
1063{
1064 struct ip_tunnel_net *itn;
1065 struct net *net;
1066 LIST_HEAD(list);
1067
1068 rtnl_lock();
1069 list_for_each_entry(net, net_list, exit_list) {
1070 itn = net_generic(net, id);
1071 ip_tunnel_destroy(net, itn, &list, ops);
1072 }
1073 unregister_netdevice_many(&list);
1074 rtnl_unlock();
1075}
1076EXPORT_SYMBOL_GPL(ip_tunnel_delete_nets);
1077
1078int ip_tunnel_newlink(struct net_device *dev, struct nlattr *tb[],
1079 struct ip_tunnel_parm *p, __u32 fwmark)
1080{
1081 struct ip_tunnel *nt;
1082 struct net *net = dev_net(dev);
1083 struct ip_tunnel_net *itn;
1084 int mtu;
1085 int err;
1086
1087 nt = netdev_priv(dev);
1088 itn = net_generic(net, nt->ip_tnl_net_id);
1089
1090 if (nt->collect_md) {
1091 if (rtnl_dereference(itn->collect_md_tun))
1092 return -EEXIST;
1093 } else {
1094 if (ip_tunnel_find(itn, p, dev->type))
1095 return -EEXIST;
1096 }
1097
1098 nt->net = net;
1099 nt->parms = *p;
1100 nt->fwmark = fwmark;
1101 err = register_netdevice(dev);
1102 if (err)
1103 goto err_register_netdevice;
1104
1105 if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1106 eth_hw_addr_random(dev);
1107
1108 mtu = ip_tunnel_bind_dev(dev);
1109 if (tb[IFLA_MTU]) {
1110 unsigned int max = IP_MAX_MTU - dev->hard_header_len - nt->hlen;
1111
1112 mtu = clamp(dev->mtu, (unsigned int)ETH_MIN_MTU,
1113 (unsigned int)(max - sizeof(struct iphdr)));
1114 }
1115
1116 err = dev_set_mtu(dev, mtu);
1117 if (err)
1118 goto err_dev_set_mtu;
1119
1120 ip_tunnel_add(itn, nt);
1121 return 0;
1122
1123err_dev_set_mtu:
1124 unregister_netdevice(dev);
1125err_register_netdevice:
1126 return err;
1127}
1128EXPORT_SYMBOL_GPL(ip_tunnel_newlink);
1129
1130int ip_tunnel_changelink(struct net_device *dev, struct nlattr *tb[],
1131 struct ip_tunnel_parm *p, __u32 fwmark)
1132{
1133 struct ip_tunnel *t;
1134 struct ip_tunnel *tunnel = netdev_priv(dev);
1135 struct net *net = tunnel->net;
1136 struct ip_tunnel_net *itn = net_generic(net, tunnel->ip_tnl_net_id);
1137
1138 if (dev == itn->fb_tunnel_dev)
1139 return -EINVAL;
1140
1141 t = ip_tunnel_find(itn, p, dev->type);
1142
1143 if (t) {
1144 if (t->dev != dev)
1145 return -EEXIST;
1146 } else {
1147 t = tunnel;
1148
1149 if (dev->type != ARPHRD_ETHER) {
1150 unsigned int nflags = 0;
1151
1152 if (ipv4_is_multicast(p->iph.daddr))
1153 nflags = IFF_BROADCAST;
1154 else if (p->iph.daddr)
1155 nflags = IFF_POINTOPOINT;
1156
1157 if ((dev->flags ^ nflags) &
1158 (IFF_POINTOPOINT | IFF_BROADCAST))
1159 return -EINVAL;
1160 }
1161 }
1162
1163 ip_tunnel_update(itn, t, dev, p, !tb[IFLA_MTU], fwmark);
1164 return 0;
1165}
1166EXPORT_SYMBOL_GPL(ip_tunnel_changelink);
1167
1168int ip_tunnel_init(struct net_device *dev)
1169{
1170 struct ip_tunnel *tunnel = netdev_priv(dev);
1171 struct iphdr *iph = &tunnel->parms.iph;
1172 int err;
1173
1174 dev->needs_free_netdev = true;
1175 dev->priv_destructor = ip_tunnel_dev_free;
1176 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1177 if (!dev->tstats)
1178 return -ENOMEM;
1179
1180 err = dst_cache_init(&tunnel->dst_cache, GFP_KERNEL);
1181 if (err) {
1182 free_percpu(dev->tstats);
1183 return err;
1184 }
1185
1186 err = gro_cells_init(&tunnel->gro_cells, dev);
1187 if (err) {
1188 dst_cache_destroy(&tunnel->dst_cache);
1189 free_percpu(dev->tstats);
1190 return err;
1191 }
1192
1193 tunnel->dev = dev;
1194 tunnel->net = dev_net(dev);
1195 strcpy(tunnel->parms.name, dev->name);
1196 iph->version = 4;
1197 iph->ihl = 5;
1198
1199 if (tunnel->collect_md) {
1200 dev->features |= NETIF_F_NETNS_LOCAL;
1201 netif_keep_dst(dev);
1202 }
1203 return 0;
1204}
1205EXPORT_SYMBOL_GPL(ip_tunnel_init);
1206
1207void ip_tunnel_uninit(struct net_device *dev)
1208{
1209 struct ip_tunnel *tunnel = netdev_priv(dev);
1210 struct net *net = tunnel->net;
1211 struct ip_tunnel_net *itn;
1212
1213 itn = net_generic(net, tunnel->ip_tnl_net_id);
1214 /* fb_tunnel_dev will be unregisted in net-exit call. */
1215 if (itn->fb_tunnel_dev != dev)
1216 ip_tunnel_del(itn, netdev_priv(dev));
1217
1218 dst_cache_reset(&tunnel->dst_cache);
1219}
1220EXPORT_SYMBOL_GPL(ip_tunnel_uninit);
1221
1222/* Do least required initialization, rest of init is done in tunnel_init call */
1223void ip_tunnel_setup(struct net_device *dev, unsigned int net_id)
1224{
1225 struct ip_tunnel *tunnel = netdev_priv(dev);
1226 tunnel->ip_tnl_net_id = net_id;
1227}
1228EXPORT_SYMBOL_GPL(ip_tunnel_setup);
1229
1230MODULE_LICENSE("GPL");
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/capability.h>
9#include <linux/module.h>
10#include <linux/types.h>
11#include <linux/kernel.h>
12#include <linux/slab.h>
13#include <linux/uaccess.h>
14#include <linux/skbuff.h>
15#include <linux/netdevice.h>
16#include <linux/in.h>
17#include <linux/tcp.h>
18#include <linux/udp.h>
19#include <linux/if_arp.h>
20#include <linux/init.h>
21#include <linux/in6.h>
22#include <linux/inetdevice.h>
23#include <linux/igmp.h>
24#include <linux/netfilter_ipv4.h>
25#include <linux/etherdevice.h>
26#include <linux/if_ether.h>
27#include <linux/if_vlan.h>
28#include <linux/rculist.h>
29#include <linux/err.h>
30
31#include <net/sock.h>
32#include <net/ip.h>
33#include <net/icmp.h>
34#include <net/protocol.h>
35#include <net/ip_tunnels.h>
36#include <net/arp.h>
37#include <net/checksum.h>
38#include <net/dsfield.h>
39#include <net/inet_ecn.h>
40#include <net/xfrm.h>
41#include <net/net_namespace.h>
42#include <net/netns/generic.h>
43#include <net/rtnetlink.h>
44#include <net/udp.h>
45#include <net/dst_metadata.h>
46#include <net/inet_dscp.h>
47
48#if IS_ENABLED(CONFIG_IPV6)
49#include <net/ipv6.h>
50#include <net/ip6_fib.h>
51#include <net/ip6_route.h>
52#endif
53
54static unsigned int ip_tunnel_hash(__be32 key, __be32 remote)
55{
56 return hash_32((__force u32)key ^ (__force u32)remote,
57 IP_TNL_HASH_BITS);
58}
59
60static bool ip_tunnel_key_match(const struct ip_tunnel_parm_kern *p,
61 const unsigned long *flags, __be32 key)
62{
63 if (!test_bit(IP_TUNNEL_KEY_BIT, flags))
64 return !test_bit(IP_TUNNEL_KEY_BIT, p->i_flags);
65
66 return test_bit(IP_TUNNEL_KEY_BIT, p->i_flags) && p->i_key == key;
67}
68
69/* Fallback tunnel: no source, no destination, no key, no options
70
71 Tunnel hash table:
72 We require exact key match i.e. if a key is present in packet
73 it will match only tunnel with the same key; if it is not present,
74 it will match only keyless tunnel.
75
76 All keysless packets, if not matched configured keyless tunnels
77 will match fallback tunnel.
78 Given src, dst and key, find appropriate for input tunnel.
79*/
80struct ip_tunnel *ip_tunnel_lookup(struct ip_tunnel_net *itn,
81 int link, const unsigned long *flags,
82 __be32 remote, __be32 local,
83 __be32 key)
84{
85 struct ip_tunnel *t, *cand = NULL;
86 struct hlist_head *head;
87 struct net_device *ndev;
88 unsigned int hash;
89
90 hash = ip_tunnel_hash(key, remote);
91 head = &itn->tunnels[hash];
92
93 hlist_for_each_entry_rcu(t, head, hash_node) {
94 if (local != t->parms.iph.saddr ||
95 remote != t->parms.iph.daddr ||
96 !(t->dev->flags & IFF_UP))
97 continue;
98
99 if (!ip_tunnel_key_match(&t->parms, flags, key))
100 continue;
101
102 if (READ_ONCE(t->parms.link) == link)
103 return t;
104 cand = t;
105 }
106
107 hlist_for_each_entry_rcu(t, head, hash_node) {
108 if (remote != t->parms.iph.daddr ||
109 t->parms.iph.saddr != 0 ||
110 !(t->dev->flags & IFF_UP))
111 continue;
112
113 if (!ip_tunnel_key_match(&t->parms, flags, key))
114 continue;
115
116 if (READ_ONCE(t->parms.link) == link)
117 return t;
118 if (!cand)
119 cand = t;
120 }
121
122 hash = ip_tunnel_hash(key, 0);
123 head = &itn->tunnels[hash];
124
125 hlist_for_each_entry_rcu(t, head, hash_node) {
126 if ((local != t->parms.iph.saddr || t->parms.iph.daddr != 0) &&
127 (local != t->parms.iph.daddr || !ipv4_is_multicast(local)))
128 continue;
129
130 if (!(t->dev->flags & IFF_UP))
131 continue;
132
133 if (!ip_tunnel_key_match(&t->parms, flags, key))
134 continue;
135
136 if (READ_ONCE(t->parms.link) == link)
137 return t;
138 if (!cand)
139 cand = t;
140 }
141
142 hlist_for_each_entry_rcu(t, head, hash_node) {
143 if ((!test_bit(IP_TUNNEL_NO_KEY_BIT, flags) &&
144 t->parms.i_key != key) ||
145 t->parms.iph.saddr != 0 ||
146 t->parms.iph.daddr != 0 ||
147 !(t->dev->flags & IFF_UP))
148 continue;
149
150 if (READ_ONCE(t->parms.link) == link)
151 return t;
152 if (!cand)
153 cand = t;
154 }
155
156 if (cand)
157 return cand;
158
159 t = rcu_dereference(itn->collect_md_tun);
160 if (t && t->dev->flags & IFF_UP)
161 return t;
162
163 ndev = READ_ONCE(itn->fb_tunnel_dev);
164 if (ndev && ndev->flags & IFF_UP)
165 return netdev_priv(ndev);
166
167 return NULL;
168}
169EXPORT_SYMBOL_GPL(ip_tunnel_lookup);
170
171static struct hlist_head *ip_bucket(struct ip_tunnel_net *itn,
172 struct ip_tunnel_parm_kern *parms)
173{
174 unsigned int h;
175 __be32 remote;
176 __be32 i_key = parms->i_key;
177
178 if (parms->iph.daddr && !ipv4_is_multicast(parms->iph.daddr))
179 remote = parms->iph.daddr;
180 else
181 remote = 0;
182
183 if (!test_bit(IP_TUNNEL_KEY_BIT, parms->i_flags) &&
184 test_bit(IP_TUNNEL_VTI_BIT, parms->i_flags))
185 i_key = 0;
186
187 h = ip_tunnel_hash(i_key, remote);
188 return &itn->tunnels[h];
189}
190
191static void ip_tunnel_add(struct ip_tunnel_net *itn, struct ip_tunnel *t)
192{
193 struct hlist_head *head = ip_bucket(itn, &t->parms);
194
195 if (t->collect_md)
196 rcu_assign_pointer(itn->collect_md_tun, t);
197 hlist_add_head_rcu(&t->hash_node, head);
198}
199
200static void ip_tunnel_del(struct ip_tunnel_net *itn, struct ip_tunnel *t)
201{
202 if (t->collect_md)
203 rcu_assign_pointer(itn->collect_md_tun, NULL);
204 hlist_del_init_rcu(&t->hash_node);
205}
206
207static struct ip_tunnel *ip_tunnel_find(struct ip_tunnel_net *itn,
208 struct ip_tunnel_parm_kern *parms,
209 int type)
210{
211 __be32 remote = parms->iph.daddr;
212 __be32 local = parms->iph.saddr;
213 IP_TUNNEL_DECLARE_FLAGS(flags);
214 __be32 key = parms->i_key;
215 int link = parms->link;
216 struct ip_tunnel *t = NULL;
217 struct hlist_head *head = ip_bucket(itn, parms);
218
219 ip_tunnel_flags_copy(flags, parms->i_flags);
220
221 hlist_for_each_entry_rcu(t, head, hash_node, lockdep_rtnl_is_held()) {
222 if (local == t->parms.iph.saddr &&
223 remote == t->parms.iph.daddr &&
224 link == READ_ONCE(t->parms.link) &&
225 type == t->dev->type &&
226 ip_tunnel_key_match(&t->parms, flags, key))
227 break;
228 }
229 return t;
230}
231
232static struct net_device *__ip_tunnel_create(struct net *net,
233 const struct rtnl_link_ops *ops,
234 struct ip_tunnel_parm_kern *parms)
235{
236 int err;
237 struct ip_tunnel *tunnel;
238 struct net_device *dev;
239 char name[IFNAMSIZ];
240
241 err = -E2BIG;
242 if (parms->name[0]) {
243 if (!dev_valid_name(parms->name))
244 goto failed;
245 strscpy(name, parms->name, IFNAMSIZ);
246 } else {
247 if (strlen(ops->kind) > (IFNAMSIZ - 3))
248 goto failed;
249 strcpy(name, ops->kind);
250 strcat(name, "%d");
251 }
252
253 ASSERT_RTNL();
254 dev = alloc_netdev(ops->priv_size, name, NET_NAME_UNKNOWN, ops->setup);
255 if (!dev) {
256 err = -ENOMEM;
257 goto failed;
258 }
259 dev_net_set(dev, net);
260
261 dev->rtnl_link_ops = ops;
262
263 tunnel = netdev_priv(dev);
264 tunnel->parms = *parms;
265 tunnel->net = net;
266
267 err = register_netdevice(dev);
268 if (err)
269 goto failed_free;
270
271 return dev;
272
273failed_free:
274 free_netdev(dev);
275failed:
276 return ERR_PTR(err);
277}
278
279static int ip_tunnel_bind_dev(struct net_device *dev)
280{
281 struct net_device *tdev = NULL;
282 struct ip_tunnel *tunnel = netdev_priv(dev);
283 const struct iphdr *iph;
284 int hlen = LL_MAX_HEADER;
285 int mtu = ETH_DATA_LEN;
286 int t_hlen = tunnel->hlen + sizeof(struct iphdr);
287
288 iph = &tunnel->parms.iph;
289
290 /* Guess output device to choose reasonable mtu and needed_headroom */
291 if (iph->daddr) {
292 struct flowi4 fl4;
293 struct rtable *rt;
294
295 ip_tunnel_init_flow(&fl4, iph->protocol, iph->daddr,
296 iph->saddr, tunnel->parms.o_key,
297 iph->tos & INET_DSCP_MASK, tunnel->net,
298 tunnel->parms.link, tunnel->fwmark, 0, 0);
299 rt = ip_route_output_key(tunnel->net, &fl4);
300
301 if (!IS_ERR(rt)) {
302 tdev = rt->dst.dev;
303 ip_rt_put(rt);
304 }
305 if (dev->type != ARPHRD_ETHER)
306 dev->flags |= IFF_POINTOPOINT;
307
308 dst_cache_reset(&tunnel->dst_cache);
309 }
310
311 if (!tdev && tunnel->parms.link)
312 tdev = __dev_get_by_index(tunnel->net, tunnel->parms.link);
313
314 if (tdev) {
315 hlen = tdev->hard_header_len + tdev->needed_headroom;
316 mtu = min(tdev->mtu, IP_MAX_MTU);
317 }
318
319 dev->needed_headroom = t_hlen + hlen;
320 mtu -= t_hlen + (dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0);
321
322 if (mtu < IPV4_MIN_MTU)
323 mtu = IPV4_MIN_MTU;
324
325 return mtu;
326}
327
328static struct ip_tunnel *ip_tunnel_create(struct net *net,
329 struct ip_tunnel_net *itn,
330 struct ip_tunnel_parm_kern *parms)
331{
332 struct ip_tunnel *nt;
333 struct net_device *dev;
334 int t_hlen;
335 int mtu;
336 int err;
337
338 dev = __ip_tunnel_create(net, itn->rtnl_link_ops, parms);
339 if (IS_ERR(dev))
340 return ERR_CAST(dev);
341
342 mtu = ip_tunnel_bind_dev(dev);
343 err = dev_set_mtu(dev, mtu);
344 if (err)
345 goto err_dev_set_mtu;
346
347 nt = netdev_priv(dev);
348 t_hlen = nt->hlen + sizeof(struct iphdr);
349 dev->min_mtu = ETH_MIN_MTU;
350 dev->max_mtu = IP_MAX_MTU - t_hlen;
351 if (dev->type == ARPHRD_ETHER)
352 dev->max_mtu -= dev->hard_header_len;
353
354 ip_tunnel_add(itn, nt);
355 return nt;
356
357err_dev_set_mtu:
358 unregister_netdevice(dev);
359 return ERR_PTR(err);
360}
361
362void ip_tunnel_md_udp_encap(struct sk_buff *skb, struct ip_tunnel_info *info)
363{
364 const struct iphdr *iph = ip_hdr(skb);
365 const struct udphdr *udph;
366
367 if (iph->protocol != IPPROTO_UDP)
368 return;
369
370 udph = (struct udphdr *)((__u8 *)iph + (iph->ihl << 2));
371 info->encap.sport = udph->source;
372 info->encap.dport = udph->dest;
373}
374EXPORT_SYMBOL(ip_tunnel_md_udp_encap);
375
376int ip_tunnel_rcv(struct ip_tunnel *tunnel, struct sk_buff *skb,
377 const struct tnl_ptk_info *tpi, struct metadata_dst *tun_dst,
378 bool log_ecn_error)
379{
380 const struct iphdr *iph = ip_hdr(skb);
381 int nh, err;
382
383#ifdef CONFIG_NET_IPGRE_BROADCAST
384 if (ipv4_is_multicast(iph->daddr)) {
385 DEV_STATS_INC(tunnel->dev, multicast);
386 skb->pkt_type = PACKET_BROADCAST;
387 }
388#endif
389
390 if (test_bit(IP_TUNNEL_CSUM_BIT, tunnel->parms.i_flags) !=
391 test_bit(IP_TUNNEL_CSUM_BIT, tpi->flags)) {
392 DEV_STATS_INC(tunnel->dev, rx_crc_errors);
393 DEV_STATS_INC(tunnel->dev, rx_errors);
394 goto drop;
395 }
396
397 if (test_bit(IP_TUNNEL_SEQ_BIT, tunnel->parms.i_flags)) {
398 if (!test_bit(IP_TUNNEL_SEQ_BIT, tpi->flags) ||
399 (tunnel->i_seqno && (s32)(ntohl(tpi->seq) - tunnel->i_seqno) < 0)) {
400 DEV_STATS_INC(tunnel->dev, rx_fifo_errors);
401 DEV_STATS_INC(tunnel->dev, rx_errors);
402 goto drop;
403 }
404 tunnel->i_seqno = ntohl(tpi->seq) + 1;
405 }
406
407 /* Save offset of outer header relative to skb->head,
408 * because we are going to reset the network header to the inner header
409 * and might change skb->head.
410 */
411 nh = skb_network_header(skb) - skb->head;
412
413 skb_set_network_header(skb, (tunnel->dev->type == ARPHRD_ETHER) ? ETH_HLEN : 0);
414
415 if (!pskb_inet_may_pull(skb)) {
416 DEV_STATS_INC(tunnel->dev, rx_length_errors);
417 DEV_STATS_INC(tunnel->dev, rx_errors);
418 goto drop;
419 }
420 iph = (struct iphdr *)(skb->head + nh);
421
422 err = IP_ECN_decapsulate(iph, skb);
423 if (unlikely(err)) {
424 if (log_ecn_error)
425 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
426 &iph->saddr, iph->tos);
427 if (err > 1) {
428 DEV_STATS_INC(tunnel->dev, rx_frame_errors);
429 DEV_STATS_INC(tunnel->dev, rx_errors);
430 goto drop;
431 }
432 }
433
434 dev_sw_netstats_rx_add(tunnel->dev, skb->len);
435 skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(tunnel->dev)));
436
437 if (tunnel->dev->type == ARPHRD_ETHER) {
438 skb->protocol = eth_type_trans(skb, tunnel->dev);
439 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
440 } else {
441 skb->dev = tunnel->dev;
442 }
443
444 if (tun_dst)
445 skb_dst_set(skb, (struct dst_entry *)tun_dst);
446
447 gro_cells_receive(&tunnel->gro_cells, skb);
448 return 0;
449
450drop:
451 if (tun_dst)
452 dst_release((struct dst_entry *)tun_dst);
453 kfree_skb(skb);
454 return 0;
455}
456EXPORT_SYMBOL_GPL(ip_tunnel_rcv);
457
458int ip_tunnel_encap_add_ops(const struct ip_tunnel_encap_ops *ops,
459 unsigned int num)
460{
461 if (num >= MAX_IPTUN_ENCAP_OPS)
462 return -ERANGE;
463
464 return !cmpxchg((const struct ip_tunnel_encap_ops **)
465 &iptun_encaps[num],
466 NULL, ops) ? 0 : -1;
467}
468EXPORT_SYMBOL(ip_tunnel_encap_add_ops);
469
470int ip_tunnel_encap_del_ops(const struct ip_tunnel_encap_ops *ops,
471 unsigned int num)
472{
473 int ret;
474
475 if (num >= MAX_IPTUN_ENCAP_OPS)
476 return -ERANGE;
477
478 ret = (cmpxchg((const struct ip_tunnel_encap_ops **)
479 &iptun_encaps[num],
480 ops, NULL) == ops) ? 0 : -1;
481
482 synchronize_net();
483
484 return ret;
485}
486EXPORT_SYMBOL(ip_tunnel_encap_del_ops);
487
488int ip_tunnel_encap_setup(struct ip_tunnel *t,
489 struct ip_tunnel_encap *ipencap)
490{
491 int hlen;
492
493 memset(&t->encap, 0, sizeof(t->encap));
494
495 hlen = ip_encap_hlen(ipencap);
496 if (hlen < 0)
497 return hlen;
498
499 t->encap.type = ipencap->type;
500 t->encap.sport = ipencap->sport;
501 t->encap.dport = ipencap->dport;
502 t->encap.flags = ipencap->flags;
503
504 t->encap_hlen = hlen;
505 t->hlen = t->encap_hlen + t->tun_hlen;
506
507 return 0;
508}
509EXPORT_SYMBOL_GPL(ip_tunnel_encap_setup);
510
511static int tnl_update_pmtu(struct net_device *dev, struct sk_buff *skb,
512 struct rtable *rt, __be16 df,
513 const struct iphdr *inner_iph,
514 int tunnel_hlen, __be32 dst, bool md)
515{
516 struct ip_tunnel *tunnel = netdev_priv(dev);
517 int pkt_size;
518 int mtu;
519
520 tunnel_hlen = md ? tunnel_hlen : tunnel->hlen;
521 pkt_size = skb->len - tunnel_hlen;
522 pkt_size -= dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0;
523
524 if (df) {
525 mtu = dst_mtu(&rt->dst) - (sizeof(struct iphdr) + tunnel_hlen);
526 mtu -= dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0;
527 } else {
528 mtu = skb_valid_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
529 }
530
531 if (skb_valid_dst(skb))
532 skb_dst_update_pmtu_no_confirm(skb, mtu);
533
534 if (skb->protocol == htons(ETH_P_IP)) {
535 if (!skb_is_gso(skb) &&
536 (inner_iph->frag_off & htons(IP_DF)) &&
537 mtu < pkt_size) {
538 icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
539 return -E2BIG;
540 }
541 }
542#if IS_ENABLED(CONFIG_IPV6)
543 else if (skb->protocol == htons(ETH_P_IPV6)) {
544 struct rt6_info *rt6;
545 __be32 daddr;
546
547 rt6 = skb_valid_dst(skb) ? dst_rt6_info(skb_dst(skb)) :
548 NULL;
549 daddr = md ? dst : tunnel->parms.iph.daddr;
550
551 if (rt6 && mtu < dst_mtu(skb_dst(skb)) &&
552 mtu >= IPV6_MIN_MTU) {
553 if ((daddr && !ipv4_is_multicast(daddr)) ||
554 rt6->rt6i_dst.plen == 128) {
555 rt6->rt6i_flags |= RTF_MODIFIED;
556 dst_metric_set(skb_dst(skb), RTAX_MTU, mtu);
557 }
558 }
559
560 if (!skb_is_gso(skb) && mtu >= IPV6_MIN_MTU &&
561 mtu < pkt_size) {
562 icmpv6_ndo_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
563 return -E2BIG;
564 }
565 }
566#endif
567 return 0;
568}
569
570static void ip_tunnel_adj_headroom(struct net_device *dev, unsigned int headroom)
571{
572 /* we must cap headroom to some upperlimit, else pskb_expand_head
573 * will overflow header offsets in skb_headers_offset_update().
574 */
575 static const unsigned int max_allowed = 512;
576
577 if (headroom > max_allowed)
578 headroom = max_allowed;
579
580 if (headroom > READ_ONCE(dev->needed_headroom))
581 WRITE_ONCE(dev->needed_headroom, headroom);
582}
583
584void ip_md_tunnel_xmit(struct sk_buff *skb, struct net_device *dev,
585 u8 proto, int tunnel_hlen)
586{
587 struct ip_tunnel *tunnel = netdev_priv(dev);
588 u32 headroom = sizeof(struct iphdr);
589 struct ip_tunnel_info *tun_info;
590 const struct ip_tunnel_key *key;
591 const struct iphdr *inner_iph;
592 struct rtable *rt = NULL;
593 struct flowi4 fl4;
594 __be16 df = 0;
595 u8 tos, ttl;
596 bool use_cache;
597
598 tun_info = skb_tunnel_info(skb);
599 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
600 ip_tunnel_info_af(tun_info) != AF_INET))
601 goto tx_error;
602 key = &tun_info->key;
603 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
604 inner_iph = (const struct iphdr *)skb_inner_network_header(skb);
605 tos = key->tos;
606 if (tos == 1) {
607 if (skb->protocol == htons(ETH_P_IP))
608 tos = inner_iph->tos;
609 else if (skb->protocol == htons(ETH_P_IPV6))
610 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph);
611 }
612 ip_tunnel_init_flow(&fl4, proto, key->u.ipv4.dst, key->u.ipv4.src,
613 tunnel_id_to_key32(key->tun_id),
614 tos & INET_DSCP_MASK, tunnel->net, 0, skb->mark,
615 skb_get_hash(skb), key->flow_flags);
616
617 if (!tunnel_hlen)
618 tunnel_hlen = ip_encap_hlen(&tun_info->encap);
619
620 if (ip_tunnel_encap(skb, &tun_info->encap, &proto, &fl4) < 0)
621 goto tx_error;
622
623 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info);
624 if (use_cache)
625 rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl4.saddr);
626 if (!rt) {
627 rt = ip_route_output_key(tunnel->net, &fl4);
628 if (IS_ERR(rt)) {
629 DEV_STATS_INC(dev, tx_carrier_errors);
630 goto tx_error;
631 }
632 if (use_cache)
633 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst,
634 fl4.saddr);
635 }
636 if (rt->dst.dev == dev) {
637 ip_rt_put(rt);
638 DEV_STATS_INC(dev, collisions);
639 goto tx_error;
640 }
641
642 if (test_bit(IP_TUNNEL_DONT_FRAGMENT_BIT, key->tun_flags))
643 df = htons(IP_DF);
644 if (tnl_update_pmtu(dev, skb, rt, df, inner_iph, tunnel_hlen,
645 key->u.ipv4.dst, true)) {
646 ip_rt_put(rt);
647 goto tx_error;
648 }
649
650 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb);
651 ttl = key->ttl;
652 if (ttl == 0) {
653 if (skb->protocol == htons(ETH_P_IP))
654 ttl = inner_iph->ttl;
655 else if (skb->protocol == htons(ETH_P_IPV6))
656 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit;
657 else
658 ttl = ip4_dst_hoplimit(&rt->dst);
659 }
660
661 headroom += LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len;
662 if (skb_cow_head(skb, headroom)) {
663 ip_rt_put(rt);
664 goto tx_dropped;
665 }
666
667 ip_tunnel_adj_headroom(dev, headroom);
668
669 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, proto, tos, ttl,
670 df, !net_eq(tunnel->net, dev_net(dev)));
671 return;
672tx_error:
673 DEV_STATS_INC(dev, tx_errors);
674 goto kfree;
675tx_dropped:
676 DEV_STATS_INC(dev, tx_dropped);
677kfree:
678 kfree_skb(skb);
679}
680EXPORT_SYMBOL_GPL(ip_md_tunnel_xmit);
681
682void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev,
683 const struct iphdr *tnl_params, u8 protocol)
684{
685 struct ip_tunnel *tunnel = netdev_priv(dev);
686 struct ip_tunnel_info *tun_info = NULL;
687 const struct iphdr *inner_iph;
688 unsigned int max_headroom; /* The extra header space needed */
689 struct rtable *rt = NULL; /* Route to the other host */
690 __be16 payload_protocol;
691 bool use_cache = false;
692 struct flowi4 fl4;
693 bool md = false;
694 bool connected;
695 u8 tos, ttl;
696 __be32 dst;
697 __be16 df;
698
699 inner_iph = (const struct iphdr *)skb_inner_network_header(skb);
700 connected = (tunnel->parms.iph.daddr != 0);
701 payload_protocol = skb_protocol(skb, true);
702
703 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
704
705 dst = tnl_params->daddr;
706 if (dst == 0) {
707 /* NBMA tunnel */
708
709 if (!skb_dst(skb)) {
710 DEV_STATS_INC(dev, tx_fifo_errors);
711 goto tx_error;
712 }
713
714 tun_info = skb_tunnel_info(skb);
715 if (tun_info && (tun_info->mode & IP_TUNNEL_INFO_TX) &&
716 ip_tunnel_info_af(tun_info) == AF_INET &&
717 tun_info->key.u.ipv4.dst) {
718 dst = tun_info->key.u.ipv4.dst;
719 md = true;
720 connected = true;
721 } else if (payload_protocol == htons(ETH_P_IP)) {
722 rt = skb_rtable(skb);
723 dst = rt_nexthop(rt, inner_iph->daddr);
724 }
725#if IS_ENABLED(CONFIG_IPV6)
726 else if (payload_protocol == htons(ETH_P_IPV6)) {
727 const struct in6_addr *addr6;
728 struct neighbour *neigh;
729 bool do_tx_error_icmp;
730 int addr_type;
731
732 neigh = dst_neigh_lookup(skb_dst(skb),
733 &ipv6_hdr(skb)->daddr);
734 if (!neigh)
735 goto tx_error;
736
737 addr6 = (const struct in6_addr *)&neigh->primary_key;
738 addr_type = ipv6_addr_type(addr6);
739
740 if (addr_type == IPV6_ADDR_ANY) {
741 addr6 = &ipv6_hdr(skb)->daddr;
742 addr_type = ipv6_addr_type(addr6);
743 }
744
745 if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
746 do_tx_error_icmp = true;
747 else {
748 do_tx_error_icmp = false;
749 dst = addr6->s6_addr32[3];
750 }
751 neigh_release(neigh);
752 if (do_tx_error_icmp)
753 goto tx_error_icmp;
754 }
755#endif
756 else
757 goto tx_error;
758
759 if (!md)
760 connected = false;
761 }
762
763 tos = tnl_params->tos;
764 if (tos & 0x1) {
765 tos &= ~0x1;
766 if (payload_protocol == htons(ETH_P_IP)) {
767 tos = inner_iph->tos;
768 connected = false;
769 } else if (payload_protocol == htons(ETH_P_IPV6)) {
770 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph);
771 connected = false;
772 }
773 }
774
775 ip_tunnel_init_flow(&fl4, protocol, dst, tnl_params->saddr,
776 tunnel->parms.o_key, tos & INET_DSCP_MASK,
777 tunnel->net, READ_ONCE(tunnel->parms.link),
778 tunnel->fwmark, skb_get_hash(skb), 0);
779
780 if (ip_tunnel_encap(skb, &tunnel->encap, &protocol, &fl4) < 0)
781 goto tx_error;
782
783 if (connected && md) {
784 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info);
785 if (use_cache)
786 rt = dst_cache_get_ip4(&tun_info->dst_cache,
787 &fl4.saddr);
788 } else {
789 rt = connected ? dst_cache_get_ip4(&tunnel->dst_cache,
790 &fl4.saddr) : NULL;
791 }
792
793 if (!rt) {
794 rt = ip_route_output_key(tunnel->net, &fl4);
795
796 if (IS_ERR(rt)) {
797 DEV_STATS_INC(dev, tx_carrier_errors);
798 goto tx_error;
799 }
800 if (use_cache)
801 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst,
802 fl4.saddr);
803 else if (!md && connected)
804 dst_cache_set_ip4(&tunnel->dst_cache, &rt->dst,
805 fl4.saddr);
806 }
807
808 if (rt->dst.dev == dev) {
809 ip_rt_put(rt);
810 DEV_STATS_INC(dev, collisions);
811 goto tx_error;
812 }
813
814 df = tnl_params->frag_off;
815 if (payload_protocol == htons(ETH_P_IP) && !tunnel->ignore_df)
816 df |= (inner_iph->frag_off & htons(IP_DF));
817
818 if (tnl_update_pmtu(dev, skb, rt, df, inner_iph, 0, 0, false)) {
819 ip_rt_put(rt);
820 goto tx_error;
821 }
822
823 if (tunnel->err_count > 0) {
824 if (time_before(jiffies,
825 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
826 tunnel->err_count--;
827
828 dst_link_failure(skb);
829 } else
830 tunnel->err_count = 0;
831 }
832
833 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb);
834 ttl = tnl_params->ttl;
835 if (ttl == 0) {
836 if (payload_protocol == htons(ETH_P_IP))
837 ttl = inner_iph->ttl;
838#if IS_ENABLED(CONFIG_IPV6)
839 else if (payload_protocol == htons(ETH_P_IPV6))
840 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit;
841#endif
842 else
843 ttl = ip4_dst_hoplimit(&rt->dst);
844 }
845
846 max_headroom = LL_RESERVED_SPACE(rt->dst.dev) + sizeof(struct iphdr)
847 + rt->dst.header_len + ip_encap_hlen(&tunnel->encap);
848
849 if (skb_cow_head(skb, max_headroom)) {
850 ip_rt_put(rt);
851 DEV_STATS_INC(dev, tx_dropped);
852 kfree_skb(skb);
853 return;
854 }
855
856 ip_tunnel_adj_headroom(dev, max_headroom);
857
858 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, protocol, tos, ttl,
859 df, !net_eq(tunnel->net, dev_net(dev)));
860 return;
861
862#if IS_ENABLED(CONFIG_IPV6)
863tx_error_icmp:
864 dst_link_failure(skb);
865#endif
866tx_error:
867 DEV_STATS_INC(dev, tx_errors);
868 kfree_skb(skb);
869}
870EXPORT_SYMBOL_GPL(ip_tunnel_xmit);
871
872static void ip_tunnel_update(struct ip_tunnel_net *itn,
873 struct ip_tunnel *t,
874 struct net_device *dev,
875 struct ip_tunnel_parm_kern *p,
876 bool set_mtu,
877 __u32 fwmark)
878{
879 ip_tunnel_del(itn, t);
880 t->parms.iph.saddr = p->iph.saddr;
881 t->parms.iph.daddr = p->iph.daddr;
882 t->parms.i_key = p->i_key;
883 t->parms.o_key = p->o_key;
884 if (dev->type != ARPHRD_ETHER) {
885 __dev_addr_set(dev, &p->iph.saddr, 4);
886 memcpy(dev->broadcast, &p->iph.daddr, 4);
887 }
888 ip_tunnel_add(itn, t);
889
890 t->parms.iph.ttl = p->iph.ttl;
891 t->parms.iph.tos = p->iph.tos;
892 t->parms.iph.frag_off = p->iph.frag_off;
893
894 if (t->parms.link != p->link || t->fwmark != fwmark) {
895 int mtu;
896
897 WRITE_ONCE(t->parms.link, p->link);
898 t->fwmark = fwmark;
899 mtu = ip_tunnel_bind_dev(dev);
900 if (set_mtu)
901 WRITE_ONCE(dev->mtu, mtu);
902 }
903 dst_cache_reset(&t->dst_cache);
904 netdev_state_change(dev);
905}
906
907int ip_tunnel_ctl(struct net_device *dev, struct ip_tunnel_parm_kern *p,
908 int cmd)
909{
910 int err = 0;
911 struct ip_tunnel *t = netdev_priv(dev);
912 struct net *net = t->net;
913 struct ip_tunnel_net *itn = net_generic(net, t->ip_tnl_net_id);
914
915 switch (cmd) {
916 case SIOCGETTUNNEL:
917 if (dev == itn->fb_tunnel_dev) {
918 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type);
919 if (!t)
920 t = netdev_priv(dev);
921 }
922 memcpy(p, &t->parms, sizeof(*p));
923 break;
924
925 case SIOCADDTUNNEL:
926 case SIOCCHGTUNNEL:
927 err = -EPERM;
928 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
929 goto done;
930 if (p->iph.ttl)
931 p->iph.frag_off |= htons(IP_DF);
932 if (!test_bit(IP_TUNNEL_VTI_BIT, p->i_flags)) {
933 if (!test_bit(IP_TUNNEL_KEY_BIT, p->i_flags))
934 p->i_key = 0;
935 if (!test_bit(IP_TUNNEL_KEY_BIT, p->o_flags))
936 p->o_key = 0;
937 }
938
939 t = ip_tunnel_find(itn, p, itn->type);
940
941 if (cmd == SIOCADDTUNNEL) {
942 if (!t) {
943 t = ip_tunnel_create(net, itn, p);
944 err = PTR_ERR_OR_ZERO(t);
945 break;
946 }
947
948 err = -EEXIST;
949 break;
950 }
951 if (dev != itn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
952 if (t) {
953 if (t->dev != dev) {
954 err = -EEXIST;
955 break;
956 }
957 } else {
958 unsigned int nflags = 0;
959
960 if (ipv4_is_multicast(p->iph.daddr))
961 nflags = IFF_BROADCAST;
962 else if (p->iph.daddr)
963 nflags = IFF_POINTOPOINT;
964
965 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
966 err = -EINVAL;
967 break;
968 }
969
970 t = netdev_priv(dev);
971 }
972 }
973
974 if (t) {
975 err = 0;
976 ip_tunnel_update(itn, t, dev, p, true, 0);
977 } else {
978 err = -ENOENT;
979 }
980 break;
981
982 case SIOCDELTUNNEL:
983 err = -EPERM;
984 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
985 goto done;
986
987 if (dev == itn->fb_tunnel_dev) {
988 err = -ENOENT;
989 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type);
990 if (!t)
991 goto done;
992 err = -EPERM;
993 if (t == netdev_priv(itn->fb_tunnel_dev))
994 goto done;
995 dev = t->dev;
996 }
997 unregister_netdevice(dev);
998 err = 0;
999 break;
1000
1001 default:
1002 err = -EINVAL;
1003 }
1004
1005done:
1006 return err;
1007}
1008EXPORT_SYMBOL_GPL(ip_tunnel_ctl);
1009
1010bool ip_tunnel_parm_from_user(struct ip_tunnel_parm_kern *kp,
1011 const void __user *data)
1012{
1013 struct ip_tunnel_parm p;
1014
1015 if (copy_from_user(&p, data, sizeof(p)))
1016 return false;
1017
1018 strscpy(kp->name, p.name);
1019 kp->link = p.link;
1020 ip_tunnel_flags_from_be16(kp->i_flags, p.i_flags);
1021 ip_tunnel_flags_from_be16(kp->o_flags, p.o_flags);
1022 kp->i_key = p.i_key;
1023 kp->o_key = p.o_key;
1024 memcpy(&kp->iph, &p.iph, min(sizeof(kp->iph), sizeof(p.iph)));
1025
1026 return true;
1027}
1028EXPORT_SYMBOL_GPL(ip_tunnel_parm_from_user);
1029
1030bool ip_tunnel_parm_to_user(void __user *data, struct ip_tunnel_parm_kern *kp)
1031{
1032 struct ip_tunnel_parm p;
1033
1034 if (!ip_tunnel_flags_is_be16_compat(kp->i_flags) ||
1035 !ip_tunnel_flags_is_be16_compat(kp->o_flags))
1036 return false;
1037
1038 memset(&p, 0, sizeof(p));
1039
1040 strscpy(p.name, kp->name);
1041 p.link = kp->link;
1042 p.i_flags = ip_tunnel_flags_to_be16(kp->i_flags);
1043 p.o_flags = ip_tunnel_flags_to_be16(kp->o_flags);
1044 p.i_key = kp->i_key;
1045 p.o_key = kp->o_key;
1046 memcpy(&p.iph, &kp->iph, min(sizeof(p.iph), sizeof(kp->iph)));
1047
1048 return !copy_to_user(data, &p, sizeof(p));
1049}
1050EXPORT_SYMBOL_GPL(ip_tunnel_parm_to_user);
1051
1052int ip_tunnel_siocdevprivate(struct net_device *dev, struct ifreq *ifr,
1053 void __user *data, int cmd)
1054{
1055 struct ip_tunnel_parm_kern p;
1056 int err;
1057
1058 if (!ip_tunnel_parm_from_user(&p, data))
1059 return -EFAULT;
1060 err = dev->netdev_ops->ndo_tunnel_ctl(dev, &p, cmd);
1061 if (!err && !ip_tunnel_parm_to_user(data, &p))
1062 return -EFAULT;
1063 return err;
1064}
1065EXPORT_SYMBOL_GPL(ip_tunnel_siocdevprivate);
1066
1067int __ip_tunnel_change_mtu(struct net_device *dev, int new_mtu, bool strict)
1068{
1069 struct ip_tunnel *tunnel = netdev_priv(dev);
1070 int t_hlen = tunnel->hlen + sizeof(struct iphdr);
1071 int max_mtu = IP_MAX_MTU - t_hlen;
1072
1073 if (dev->type == ARPHRD_ETHER)
1074 max_mtu -= dev->hard_header_len;
1075
1076 if (new_mtu < ETH_MIN_MTU)
1077 return -EINVAL;
1078
1079 if (new_mtu > max_mtu) {
1080 if (strict)
1081 return -EINVAL;
1082
1083 new_mtu = max_mtu;
1084 }
1085
1086 WRITE_ONCE(dev->mtu, new_mtu);
1087 return 0;
1088}
1089EXPORT_SYMBOL_GPL(__ip_tunnel_change_mtu);
1090
1091int ip_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1092{
1093 return __ip_tunnel_change_mtu(dev, new_mtu, true);
1094}
1095EXPORT_SYMBOL_GPL(ip_tunnel_change_mtu);
1096
1097static void ip_tunnel_dev_free(struct net_device *dev)
1098{
1099 struct ip_tunnel *tunnel = netdev_priv(dev);
1100
1101 gro_cells_destroy(&tunnel->gro_cells);
1102 dst_cache_destroy(&tunnel->dst_cache);
1103}
1104
1105void ip_tunnel_dellink(struct net_device *dev, struct list_head *head)
1106{
1107 struct ip_tunnel *tunnel = netdev_priv(dev);
1108 struct ip_tunnel_net *itn;
1109
1110 itn = net_generic(tunnel->net, tunnel->ip_tnl_net_id);
1111
1112 if (itn->fb_tunnel_dev != dev) {
1113 ip_tunnel_del(itn, netdev_priv(dev));
1114 unregister_netdevice_queue(dev, head);
1115 }
1116}
1117EXPORT_SYMBOL_GPL(ip_tunnel_dellink);
1118
1119struct net *ip_tunnel_get_link_net(const struct net_device *dev)
1120{
1121 struct ip_tunnel *tunnel = netdev_priv(dev);
1122
1123 return READ_ONCE(tunnel->net);
1124}
1125EXPORT_SYMBOL(ip_tunnel_get_link_net);
1126
1127int ip_tunnel_get_iflink(const struct net_device *dev)
1128{
1129 const struct ip_tunnel *tunnel = netdev_priv(dev);
1130
1131 return READ_ONCE(tunnel->parms.link);
1132}
1133EXPORT_SYMBOL(ip_tunnel_get_iflink);
1134
1135int ip_tunnel_init_net(struct net *net, unsigned int ip_tnl_net_id,
1136 struct rtnl_link_ops *ops, char *devname)
1137{
1138 struct ip_tunnel_net *itn = net_generic(net, ip_tnl_net_id);
1139 struct ip_tunnel_parm_kern parms;
1140 unsigned int i;
1141
1142 itn->rtnl_link_ops = ops;
1143 for (i = 0; i < IP_TNL_HASH_SIZE; i++)
1144 INIT_HLIST_HEAD(&itn->tunnels[i]);
1145
1146 if (!ops || !net_has_fallback_tunnels(net)) {
1147 struct ip_tunnel_net *it_init_net;
1148
1149 it_init_net = net_generic(&init_net, ip_tnl_net_id);
1150 itn->type = it_init_net->type;
1151 itn->fb_tunnel_dev = NULL;
1152 return 0;
1153 }
1154
1155 memset(&parms, 0, sizeof(parms));
1156 if (devname)
1157 strscpy(parms.name, devname, IFNAMSIZ);
1158
1159 rtnl_lock();
1160 itn->fb_tunnel_dev = __ip_tunnel_create(net, ops, &parms);
1161 /* FB netdevice is special: we have one, and only one per netns.
1162 * Allowing to move it to another netns is clearly unsafe.
1163 */
1164 if (!IS_ERR(itn->fb_tunnel_dev)) {
1165 itn->fb_tunnel_dev->netns_local = true;
1166 itn->fb_tunnel_dev->mtu = ip_tunnel_bind_dev(itn->fb_tunnel_dev);
1167 ip_tunnel_add(itn, netdev_priv(itn->fb_tunnel_dev));
1168 itn->type = itn->fb_tunnel_dev->type;
1169 }
1170 rtnl_unlock();
1171
1172 return PTR_ERR_OR_ZERO(itn->fb_tunnel_dev);
1173}
1174EXPORT_SYMBOL_GPL(ip_tunnel_init_net);
1175
1176static void ip_tunnel_destroy(struct net *net, struct ip_tunnel_net *itn,
1177 struct list_head *head,
1178 struct rtnl_link_ops *ops)
1179{
1180 struct net_device *dev, *aux;
1181 int h;
1182
1183 for_each_netdev_safe(net, dev, aux)
1184 if (dev->rtnl_link_ops == ops)
1185 unregister_netdevice_queue(dev, head);
1186
1187 for (h = 0; h < IP_TNL_HASH_SIZE; h++) {
1188 struct ip_tunnel *t;
1189 struct hlist_node *n;
1190 struct hlist_head *thead = &itn->tunnels[h];
1191
1192 hlist_for_each_entry_safe(t, n, thead, hash_node)
1193 /* If dev is in the same netns, it has already
1194 * been added to the list by the previous loop.
1195 */
1196 if (!net_eq(dev_net(t->dev), net))
1197 unregister_netdevice_queue(t->dev, head);
1198 }
1199}
1200
1201void ip_tunnel_delete_nets(struct list_head *net_list, unsigned int id,
1202 struct rtnl_link_ops *ops,
1203 struct list_head *dev_to_kill)
1204{
1205 struct ip_tunnel_net *itn;
1206 struct net *net;
1207
1208 ASSERT_RTNL();
1209 list_for_each_entry(net, net_list, exit_list) {
1210 itn = net_generic(net, id);
1211 ip_tunnel_destroy(net, itn, dev_to_kill, ops);
1212 }
1213}
1214EXPORT_SYMBOL_GPL(ip_tunnel_delete_nets);
1215
1216int ip_tunnel_newlink(struct net_device *dev, struct nlattr *tb[],
1217 struct ip_tunnel_parm_kern *p, __u32 fwmark)
1218{
1219 struct ip_tunnel *nt;
1220 struct net *net = dev_net(dev);
1221 struct ip_tunnel_net *itn;
1222 int mtu;
1223 int err;
1224
1225 nt = netdev_priv(dev);
1226 itn = net_generic(net, nt->ip_tnl_net_id);
1227
1228 if (nt->collect_md) {
1229 if (rtnl_dereference(itn->collect_md_tun))
1230 return -EEXIST;
1231 } else {
1232 if (ip_tunnel_find(itn, p, dev->type))
1233 return -EEXIST;
1234 }
1235
1236 nt->net = net;
1237 nt->parms = *p;
1238 nt->fwmark = fwmark;
1239 err = register_netdevice(dev);
1240 if (err)
1241 goto err_register_netdevice;
1242
1243 if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1244 eth_hw_addr_random(dev);
1245
1246 mtu = ip_tunnel_bind_dev(dev);
1247 if (tb[IFLA_MTU]) {
1248 unsigned int max = IP_MAX_MTU - (nt->hlen + sizeof(struct iphdr));
1249
1250 if (dev->type == ARPHRD_ETHER)
1251 max -= dev->hard_header_len;
1252
1253 mtu = clamp(dev->mtu, (unsigned int)ETH_MIN_MTU, max);
1254 }
1255
1256 err = dev_set_mtu(dev, mtu);
1257 if (err)
1258 goto err_dev_set_mtu;
1259
1260 ip_tunnel_add(itn, nt);
1261 return 0;
1262
1263err_dev_set_mtu:
1264 unregister_netdevice(dev);
1265err_register_netdevice:
1266 return err;
1267}
1268EXPORT_SYMBOL_GPL(ip_tunnel_newlink);
1269
1270int ip_tunnel_changelink(struct net_device *dev, struct nlattr *tb[],
1271 struct ip_tunnel_parm_kern *p, __u32 fwmark)
1272{
1273 struct ip_tunnel *t;
1274 struct ip_tunnel *tunnel = netdev_priv(dev);
1275 struct net *net = tunnel->net;
1276 struct ip_tunnel_net *itn = net_generic(net, tunnel->ip_tnl_net_id);
1277
1278 if (dev == itn->fb_tunnel_dev)
1279 return -EINVAL;
1280
1281 t = ip_tunnel_find(itn, p, dev->type);
1282
1283 if (t) {
1284 if (t->dev != dev)
1285 return -EEXIST;
1286 } else {
1287 t = tunnel;
1288
1289 if (dev->type != ARPHRD_ETHER) {
1290 unsigned int nflags = 0;
1291
1292 if (ipv4_is_multicast(p->iph.daddr))
1293 nflags = IFF_BROADCAST;
1294 else if (p->iph.daddr)
1295 nflags = IFF_POINTOPOINT;
1296
1297 if ((dev->flags ^ nflags) &
1298 (IFF_POINTOPOINT | IFF_BROADCAST))
1299 return -EINVAL;
1300 }
1301 }
1302
1303 ip_tunnel_update(itn, t, dev, p, !tb[IFLA_MTU], fwmark);
1304 return 0;
1305}
1306EXPORT_SYMBOL_GPL(ip_tunnel_changelink);
1307
1308int ip_tunnel_init(struct net_device *dev)
1309{
1310 struct ip_tunnel *tunnel = netdev_priv(dev);
1311 struct iphdr *iph = &tunnel->parms.iph;
1312 int err;
1313
1314 dev->needs_free_netdev = true;
1315 dev->priv_destructor = ip_tunnel_dev_free;
1316 dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
1317
1318 err = dst_cache_init(&tunnel->dst_cache, GFP_KERNEL);
1319 if (err)
1320 return err;
1321
1322 err = gro_cells_init(&tunnel->gro_cells, dev);
1323 if (err) {
1324 dst_cache_destroy(&tunnel->dst_cache);
1325 return err;
1326 }
1327
1328 tunnel->dev = dev;
1329 tunnel->net = dev_net(dev);
1330 strscpy(tunnel->parms.name, dev->name);
1331 iph->version = 4;
1332 iph->ihl = 5;
1333
1334 if (tunnel->collect_md)
1335 netif_keep_dst(dev);
1336 netdev_lockdep_set_classes(dev);
1337 return 0;
1338}
1339EXPORT_SYMBOL_GPL(ip_tunnel_init);
1340
1341void ip_tunnel_uninit(struct net_device *dev)
1342{
1343 struct ip_tunnel *tunnel = netdev_priv(dev);
1344 struct net *net = tunnel->net;
1345 struct ip_tunnel_net *itn;
1346
1347 itn = net_generic(net, tunnel->ip_tnl_net_id);
1348 ip_tunnel_del(itn, netdev_priv(dev));
1349 if (itn->fb_tunnel_dev == dev)
1350 WRITE_ONCE(itn->fb_tunnel_dev, NULL);
1351
1352 dst_cache_reset(&tunnel->dst_cache);
1353}
1354EXPORT_SYMBOL_GPL(ip_tunnel_uninit);
1355
1356/* Do least required initialization, rest of init is done in tunnel_init call */
1357void ip_tunnel_setup(struct net_device *dev, unsigned int net_id)
1358{
1359 struct ip_tunnel *tunnel = netdev_priv(dev);
1360 tunnel->ip_tnl_net_id = net_id;
1361}
1362EXPORT_SYMBOL_GPL(ip_tunnel_setup);
1363
1364MODULE_DESCRIPTION("IPv4 tunnel implementation library");
1365MODULE_LICENSE("GPL");