projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ip6_gre: only increase err_count for some certain type icmpv6 in ip6gre_err
[GitHub/exynos8895/android_kernel_samsung_universal8895.git] / net / ipv6 / ip6_gre.c
CommitLineData
c12b395a 1/*
2 * GRE over IPv6 protocol decoder.
3 *
4 * Authors: Dmitry Kozlov (xeb@mail.ru)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12
13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14
15#include <linux/capability.h>
16#include <linux/module.h>
17#include <linux/types.h>
18#include <linux/kernel.h>
19#include <linux/slab.h>
20#include <linux/uaccess.h>
21#include <linux/skbuff.h>
22#include <linux/netdevice.h>
23#include <linux/in.h>
24#include <linux/tcp.h>
25#include <linux/udp.h>
26#include <linux/if_arp.h>
27#include <linux/mroute.h>
28#include <linux/init.h>
29#include <linux/in6.h>
30#include <linux/inetdevice.h>
31#include <linux/igmp.h>
32#include <linux/netfilter_ipv4.h>
33#include <linux/etherdevice.h>
34#include <linux/if_ether.h>
35#include <linux/hash.h>
36#include <linux/if_tunnel.h>
37#include <linux/ip6_tunnel.h>
38
39#include <net/sock.h>
40#include <net/ip.h>
c5441932 41#include <net/ip_tunnels.h>
c12b395a 42#include <net/icmp.h>
43#include <net/protocol.h>
44#include <net/addrconf.h>
45#include <net/arp.h>
46#include <net/checksum.h>
47#include <net/dsfield.h>
48#include <net/inet_ecn.h>
49#include <net/xfrm.h>
50#include <net/net_namespace.h>
51#include <net/netns/generic.h>
52#include <net/rtnetlink.h>
53
54#include <net/ipv6.h>
55#include <net/ip6_fib.h>
56#include <net/ip6_route.h>
57#include <net/ip6_tunnel.h>
eaa3a58f 58#include <net/gre.h>
c12b395a 59
60
eccc1bb8 61static bool log_ecn_error = true;
62module_param(log_ecn_error, bool, 0644);
63MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
64
c12b395a 65#define HASH_SIZE_SHIFT 5
66#define HASH_SIZE (1 << HASH_SIZE_SHIFT)
67
68static int ip6gre_net_id __read_mostly;
69struct ip6gre_net {
70 struct ip6_tnl __rcu *tunnels[4][HASH_SIZE];
71
72 struct net_device *fb_tunnel_dev;
73};
74
75static struct rtnl_link_ops ip6gre_link_ops __read_mostly;
22f08069 76static struct rtnl_link_ops ip6gre_tap_ops __read_mostly;
c12b395a 77static int ip6gre_tunnel_init(struct net_device *dev);
78static void ip6gre_tunnel_setup(struct net_device *dev);
79static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t);
80static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu);
81
82/* Tunnel hash table */
83
84/*
85 4 hash tables:
86
87 3: (remote,local)
88 2: (remote,*)
89 1: (*,local)
90 0: (*,*)
91
92 We require exact key match i.e. if a key is present in packet
93 it will match only tunnel with the same key; if it is not present,
94 it will match only keyless tunnel.
95
96 All keysless packets, if not matched configured keyless tunnels
97 will match fallback tunnel.
98 */
99
100#define HASH_KEY(key) (((__force u32)key^((__force u32)key>>4))&(HASH_SIZE - 1))
101static u32 HASH_ADDR(const struct in6_addr *addr)
102{
103 u32 hash = ipv6_addr_hash(addr);
104
105 return hash_32(hash, HASH_SIZE_SHIFT);
106}
107
108#define tunnels_r_l tunnels[3]
109#define tunnels_r tunnels[2]
110#define tunnels_l tunnels[1]
111#define tunnels_wc tunnels[0]
c12b395a 112
c12b395a 113/* Given src, dst and key, find appropriate for input tunnel. */
114
115static struct ip6_tnl *ip6gre_tunnel_lookup(struct net_device *dev,
116 const struct in6_addr *remote, const struct in6_addr *local,
117 __be32 key, __be16 gre_proto)
118{
119 struct net *net = dev_net(dev);
120 int link = dev->ifindex;
121 unsigned int h0 = HASH_ADDR(remote);
122 unsigned int h1 = HASH_KEY(key);
123 struct ip6_tnl *t, *cand = NULL;
124 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
125 int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
126 ARPHRD_ETHER : ARPHRD_IP6GRE;
127 int score, cand_score = 4;
128
e086cadc 129 for_each_ip_tunnel_rcu(t, ign->tunnels_r_l[h0 ^ h1]) {
c12b395a 130 if (!ipv6_addr_equal(local, &t->parms.laddr) ||
131 !ipv6_addr_equal(remote, &t->parms.raddr) ||
132 key != t->parms.i_key ||
133 !(t->dev->flags & IFF_UP))
134 continue;
135
136 if (t->dev->type != ARPHRD_IP6GRE &&
137 t->dev->type != dev_type)
138 continue;
139
140 score = 0;
141 if (t->parms.link != link)
142 score |= 1;
143 if (t->dev->type != dev_type)
144 score |= 2;
145 if (score == 0)
146 return t;
147
148 if (score < cand_score) {
149 cand = t;
150 cand_score = score;
151 }
152 }
153
e086cadc 154 for_each_ip_tunnel_rcu(t, ign->tunnels_r[h0 ^ h1]) {
c12b395a 155 if (!ipv6_addr_equal(remote, &t->parms.raddr) ||
156 key != t->parms.i_key ||
157 !(t->dev->flags & IFF_UP))
158 continue;
159
160 if (t->dev->type != ARPHRD_IP6GRE &&
161 t->dev->type != dev_type)
162 continue;
163
164 score = 0;
165 if (t->parms.link != link)
166 score |= 1;
167 if (t->dev->type != dev_type)
168 score |= 2;
169 if (score == 0)
170 return t;
171
172 if (score < cand_score) {
173 cand = t;
174 cand_score = score;
175 }
176 }
177
e086cadc 178 for_each_ip_tunnel_rcu(t, ign->tunnels_l[h1]) {
c12b395a 179 if ((!ipv6_addr_equal(local, &t->parms.laddr) &&
180 (!ipv6_addr_equal(local, &t->parms.raddr) ||
181 !ipv6_addr_is_multicast(local))) ||
182 key != t->parms.i_key ||
183 !(t->dev->flags & IFF_UP))
184 continue;
185
186 if (t->dev->type != ARPHRD_IP6GRE &&
187 t->dev->type != dev_type)
188 continue;
189
190 score = 0;
191 if (t->parms.link != link)
192 score |= 1;
193 if (t->dev->type != dev_type)
194 score |= 2;
195 if (score == 0)
196 return t;
197
198 if (score < cand_score) {
199 cand = t;
200 cand_score = score;
201 }
202 }
203
e086cadc 204 for_each_ip_tunnel_rcu(t, ign->tunnels_wc[h1]) {
c12b395a 205 if (t->parms.i_key != key ||
206 !(t->dev->flags & IFF_UP))
207 continue;
208
209 if (t->dev->type != ARPHRD_IP6GRE &&
210 t->dev->type != dev_type)
211 continue;
212
213 score = 0;
214 if (t->parms.link != link)
215 score |= 1;
216 if (t->dev->type != dev_type)
217 score |= 2;
218 if (score == 0)
219 return t;
220
221 if (score < cand_score) {
222 cand = t;
223 cand_score = score;
224 }
225 }
226
53b24b8f 227 if (cand)
c12b395a 228 return cand;
229
230 dev = ign->fb_tunnel_dev;
231 if (dev->flags & IFF_UP)
232 return netdev_priv(dev);
233
234 return NULL;
235}
236
237static struct ip6_tnl __rcu **__ip6gre_bucket(struct ip6gre_net *ign,
238 const struct __ip6_tnl_parm *p)
239{
240 const struct in6_addr *remote = &p->raddr;
241 const struct in6_addr *local = &p->laddr;
242 unsigned int h = HASH_KEY(p->i_key);
243 int prio = 0;
244
245 if (!ipv6_addr_any(local))
246 prio |= 1;
247 if (!ipv6_addr_any(remote) && !ipv6_addr_is_multicast(remote)) {
248 prio |= 2;
249 h ^= HASH_ADDR(remote);
250 }
251
252 return &ign->tunnels[prio][h];
253}
254
255static inline struct ip6_tnl __rcu **ip6gre_bucket(struct ip6gre_net *ign,
256 const struct ip6_tnl *t)
257{
258 return __ip6gre_bucket(ign, &t->parms);
259}
260
261static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t)
262{
263 struct ip6_tnl __rcu **tp = ip6gre_bucket(ign, t);
264
265 rcu_assign_pointer(t->next, rtnl_dereference(*tp));
266 rcu_assign_pointer(*tp, t);
267}
268
269static void ip6gre_tunnel_unlink(struct ip6gre_net *ign, struct ip6_tnl *t)
270{
271 struct ip6_tnl __rcu **tp;
272 struct ip6_tnl *iter;
273
274 for (tp = ip6gre_bucket(ign, t);
275 (iter = rtnl_dereference(*tp)) != NULL;
276 tp = &iter->next) {
277 if (t == iter) {
278 rcu_assign_pointer(*tp, t->next);
279 break;
280 }
281 }
282}
283
284static struct ip6_tnl *ip6gre_tunnel_find(struct net *net,
285 const struct __ip6_tnl_parm *parms,
286 int type)
287{
288 const struct in6_addr *remote = &parms->raddr;
289 const struct in6_addr *local = &parms->laddr;
290 __be32 key = parms->i_key;
291 int link = parms->link;
292 struct ip6_tnl *t;
293 struct ip6_tnl __rcu **tp;
294 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
295
296 for (tp = __ip6gre_bucket(ign, parms);
297 (t = rtnl_dereference(*tp)) != NULL;
298 tp = &t->next)
299 if (ipv6_addr_equal(local, &t->parms.laddr) &&
300 ipv6_addr_equal(remote, &t->parms.raddr) &&
301 key == t->parms.i_key &&
302 link == t->parms.link &&
303 type == t->dev->type)
304 break;
305
306 return t;
307}
308
309static struct ip6_tnl *ip6gre_tunnel_locate(struct net *net,
310 const struct __ip6_tnl_parm *parms, int create)
311{
312 struct ip6_tnl *t, *nt;
313 struct net_device *dev;
314 char name[IFNAMSIZ];
315 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
316
317 t = ip6gre_tunnel_find(net, parms, ARPHRD_IP6GRE);
cd0a0bd9
SK		 318 if (t && create)
319 return NULL;
c12b395a 320 if (t || !create)
321 return t;
322
323 if (parms->name[0])
324 strlcpy(name, parms->name, IFNAMSIZ);
325 else
326 strcpy(name, "ip6gre%d");
327
c835a677
TG		 328 dev = alloc_netdev(sizeof(*t), name, NET_NAME_UNKNOWN,
329 ip6gre_tunnel_setup);
c12b395a 330 if (!dev)
331 return NULL;
332
333 dev_net_set(dev, net);
334
335 nt = netdev_priv(dev);
336 nt->parms = *parms;
337 dev->rtnl_link_ops = &ip6gre_link_ops;
338
339 nt->dev = dev;
0bd87628 340 nt->net = dev_net(dev);
c12b395a 341 ip6gre_tnl_link_config(nt, 1);
342
343 if (register_netdevice(dev) < 0)
344 goto failed_free;
345
346 /* Can use a lockless transmit, unless we generate output sequences */
347 if (!(nt->parms.o_flags & GRE_SEQ))
348 dev->features |= NETIF_F_LLTX;
349
350 dev_hold(dev);
351 ip6gre_tunnel_link(ign, nt);
352 return nt;
353
354failed_free:
355 free_netdev(dev);
356 return NULL;
357}
358
359static void ip6gre_tunnel_uninit(struct net_device *dev)
360{
22f08069
ND		 361 struct ip6_tnl *t = netdev_priv(dev);
362 struct ip6gre_net *ign = net_generic(t->net, ip6gre_net_id);
c12b395a 363
22f08069 364 ip6gre_tunnel_unlink(ign, t);
d4257295 365 ip6_tnl_dst_reset(t);
c12b395a 366 dev_put(dev);
367}
368
369
370static void ip6gre_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
eaa3a58f 371 u8 type, u8 code, int offset, __be32 info)
c12b395a 372{
eaa3a58f
ED		 373 const struct gre_base_hdr *greh;
374 const struct ipv6hdr *ipv6h;
375 int grehlen = sizeof(*greh);
c12b395a 376 struct ip6_tnl *t;
eaa3a58f 377 int key_off = 0;
c12b395a 378 __be16 flags;
eaa3a58f 379 __be32 key;
c12b395a 380
eaa3a58f
ED		 381 if (!pskb_may_pull(skb, offset + grehlen))
382 return;
383 greh = (const struct gre_base_hdr *)(skb->data + offset);
384 flags = greh->flags;
385 if (flags & (GRE_VERSION | GRE_ROUTING))
386 return;
387 if (flags & GRE_CSUM)
388 grehlen += 4;
389 if (flags & GRE_KEY) {
390 key_off = grehlen + offset;
391 grehlen += 4;
c12b395a 392 }
393
eaa3a58f 394 if (!pskb_may_pull(skb, offset + grehlen))
c12b395a 395 return;
b87fb39e 396 ipv6h = (const struct ipv6hdr *)skb->data;
eaa3a58f
ED		 397 greh = (const struct gre_base_hdr *)(skb->data + offset);
398 key = key_off ? *(__be32 *)(skb->data + key_off) : 0;
c12b395a 399
c12b395a 400 t = ip6gre_tunnel_lookup(skb->dev, &ipv6h->daddr, &ipv6h->saddr,
eaa3a58f 401 key, greh->protocol);
63159f29 402 if (!t)
0c5794a6 403 return;
c12b395a 404
405 switch (type) {
406 __u32 teli;
407 struct ipv6_tlv_tnl_enc_lim *tel;
408 __u32 mtu;
409 case ICMPV6_DEST_UNREACH:
a46496ce
MB		 410 net_dbg_ratelimited("%s: Path to destination invalid or inactive!\n",
411 t->parms.name);
5c8a0850
XL		 412 if (code != ICMPV6_PORT_UNREACH)
413 break;
414 return;
c12b395a 415 case ICMPV6_TIME_EXCEED:
416 if (code == ICMPV6_EXC_HOPLIMIT) {
a46496ce
MB		 417 net_dbg_ratelimited("%s: Too small hop limit or routing loop in tunnel!\n",
418 t->parms.name);
5c8a0850 419 break;
c12b395a 420 }
5c8a0850 421 return;
c12b395a 422 case ICMPV6_PARAMPROB:
423 teli = 0;
424 if (code == ICMPV6_HDR_FIELD)
425 teli = ip6_tnl_parse_tlv_enc_lim(skb, skb->data);
426
d1e158e2 427 if (teli && teli == be32_to_cpu(info) - 2) {
c12b395a 428 tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
429 if (tel->encap_limit == 0) {
a46496ce
MB		 430 net_dbg_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
431 t->parms.name);
c12b395a 432 }
433 } else {
a46496ce
MB		 434 net_dbg_ratelimited("%s: Recipient unable to parse tunneled packet!\n",
435 t->parms.name);
c12b395a 436 }
5c8a0850 437 return;
c12b395a 438 case ICMPV6_PKT_TOOBIG:
d1e158e2 439 mtu = be32_to_cpu(info) - offset;
c12b395a 440 if (mtu < IPV6_MIN_MTU)
441 mtu = IPV6_MIN_MTU;
442 t->dev->mtu = mtu;
5c8a0850 443 return;
c12b395a 444 }
445
446 if (time_before(jiffies, t->err_time + IP6TUNNEL_ERR_TIMEO))
447 t->err_count++;
448 else
449 t->err_count = 1;
450 t->err_time = jiffies;
c12b395a 451}
452
c12b395a 453static int ip6gre_rcv(struct sk_buff *skb)
454{
455 const struct ipv6hdr *ipv6h;
456 u8 *h;
457 __be16 flags;
458 __sum16 csum = 0;
459 __be32 key = 0;
460 u32 seqno = 0;
461 struct ip6_tnl *tunnel;
462 int offset = 4;
463 __be16 gre_proto;
eccc1bb8 464 int err;
c12b395a 465
466 if (!pskb_may_pull(skb, sizeof(struct in6_addr)))
0c5794a6 467 goto drop;
c12b395a 468
469 ipv6h = ipv6_hdr(skb);
470 h = skb->data;
471 flags = *(__be16 *)h;
472
473 if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
474 /* - Version must be 0.
475 - We do not support routing headers.
476 */
477 if (flags&(GRE_VERSION|GRE_ROUTING))
0c5794a6 478 goto drop;
c12b395a 479
480 if (flags&GRE_CSUM) {
81249bea 481 csum = skb_checksum_simple_validate(skb);
c12b395a 482 offset += 4;
483 }
484 if (flags&GRE_KEY) {
485 key = *(__be32 *)(h + offset);
486 offset += 4;
487 }
488 if (flags&GRE_SEQ) {
489 seqno = ntohl(*(__be32 *)(h + offset));
490 offset += 4;
491 }
492 }
493
494 gre_proto = *(__be16 *)(h + 2);
495
c12b395a 496 tunnel = ip6gre_tunnel_lookup(skb->dev,
497 &ipv6h->saddr, &ipv6h->daddr, key,
498 gre_proto);
499 if (tunnel) {
8f84985f 500 struct pcpu_sw_netstats *tstats;
c12b395a 501
502 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
503 goto drop;
504
505 if (!ip6_tnl_rcv_ctl(tunnel, &ipv6h->daddr, &ipv6h->saddr)) {
506 tunnel->dev->stats.rx_dropped++;
507 goto drop;
508 }
509
c12b395a 510 skb->protocol = gre_proto;
511 /* WCCP version 1 and 2 protocol decoding.
be6572fd 512 * - Change protocol to IPv6
c12b395a 513 * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
514 */
515 if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
be6572fd 516 skb->protocol = htons(ETH_P_IPV6);
c12b395a 517 if ((*(h + offset) & 0xF0) != 0x40)
518 offset += 4;
519 }
520
521 skb->mac_header = skb->network_header;
522 __pskb_pull(skb, offset);
523 skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
c12b395a 524
525 if (((flags&GRE_CSUM) && csum) ||
526 (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
527 tunnel->dev->stats.rx_crc_errors++;
528 tunnel->dev->stats.rx_errors++;
529 goto drop;
530 }
531 if (tunnel->parms.i_flags&GRE_SEQ) {
532 if (!(flags&GRE_SEQ) ||
533 (tunnel->i_seqno &&
534 (s32)(seqno - tunnel->i_seqno) < 0)) {
535 tunnel->dev->stats.rx_fifo_errors++;
536 tunnel->dev->stats.rx_errors++;
537 goto drop;
538 }
539 tunnel->i_seqno = seqno + 1;
540 }
541
542 /* Warning: All skb pointers will be invalidated! */
543 if (tunnel->dev->type == ARPHRD_ETHER) {
544 if (!pskb_may_pull(skb, ETH_HLEN)) {
545 tunnel->dev->stats.rx_length_errors++;
546 tunnel->dev->stats.rx_errors++;
547 goto drop;
548 }
549
550 ipv6h = ipv6_hdr(skb);
551 skb->protocol = eth_type_trans(skb, tunnel->dev);
552 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
553 }
554
ea23192e 555 __skb_tunnel_rx(skb, tunnel->dev, tunnel->net);
eccc1bb8 556
557 skb_reset_network_header(skb);
558
559 err = IP6_ECN_decapsulate(ipv6h, skb);
560 if (unlikely(err)) {
561 if (log_ecn_error)
562 net_info_ratelimited("non-ECT from %pI6 with dsfield=%#x\n",
563 &ipv6h->saddr,
564 ipv6_get_dsfield(ipv6h));
565 if (err > 1) {
566 ++tunnel->dev->stats.rx_frame_errors;
567 ++tunnel->dev->stats.rx_errors;
568 goto drop;
569 }
570 }
571
c12b395a 572 tstats = this_cpu_ptr(tunnel->dev->tstats);
573 u64_stats_update_begin(&tstats->syncp);
574 tstats->rx_packets++;
575 tstats->rx_bytes += skb->len;
576 u64_stats_update_end(&tstats->syncp);
577
c12b395a 578 netif_rx(skb);
579
c12b395a 580 return 0;
581 }
582 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
583
584drop:
c12b395a 585 kfree_skb(skb);
586 return 0;
587}
588
589struct ipv6_tel_txoption {
590 struct ipv6_txoptions ops;
591 __u8 dst_opt[8];
592};
593
594static void init_tel_txopt(struct ipv6_tel_txoption *opt, __u8 encap_limit)
595{
596 memset(opt, 0, sizeof(struct ipv6_tel_txoption));
597
598 opt->dst_opt[2] = IPV6_TLV_TNL_ENCAP_LIMIT;
599 opt->dst_opt[3] = 1;
600 opt->dst_opt[4] = encap_limit;
601 opt->dst_opt[5] = IPV6_TLV_PADN;
602 opt->dst_opt[6] = 1;
603
604 opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
605 opt->ops.opt_nflen = 8;
606}
607
608static netdev_tx_t ip6gre_xmit2(struct sk_buff *skb,
609 struct net_device *dev,
610 __u8 dsfield,
611 struct flowi6 *fl6,
612 int encap_limit,
613 __u32 *pmtu)
614{
c12b395a 615 struct ip6_tnl *tunnel = netdev_priv(dev);
22f08069 616 struct net *net = tunnel->net;
c12b395a 617 struct net_device *tdev; /* Device to other host */
618 struct ipv6hdr *ipv6h; /* Our new IP header */
3da812d8 619 unsigned int max_headroom = 0; /* The extra header space needed */
c12b395a 620 int gre_hlen;
621 struct ipv6_tel_txoption opt;
622 int mtu;
623 struct dst_entry *dst = NULL, *ndst = NULL;
624 struct net_device_stats *stats = &tunnel->dev->stats;
625 int err = -1;
626 u8 proto;
c12b395a 627 struct sk_buff *new_skb;
54bc9bac 628 __be16 protocol;
c12b395a 629
630 if (dev->type == ARPHRD_ETHER)
631 IPCB(skb)->flags = 0;
632
633 if (dev->header_ops && dev->type == ARPHRD_IP6GRE) {
634 gre_hlen = 0;
635 ipv6h = (struct ipv6hdr *)skb->data;
636 fl6->daddr = ipv6h->daddr;
637 } else {
638 gre_hlen = tunnel->hlen;
639 fl6->daddr = tunnel->parms.raddr;
640 }
641
642 if (!fl6->flowi6_mark)
f230d1e8 643 dst = ip6_tnl_dst_get(tunnel);
c12b395a 644
645 if (!dst) {
cdf3464e 646 dst = ip6_route_output(net, NULL, fl6);
c12b395a 647
cdf3464e 648 if (dst->error)
c12b395a 649 goto tx_err_link_failure;
cdf3464e
MKL		 650 dst = xfrm_lookup(net, dst, flowi6_to_flowi(fl6), NULL, 0);
651 if (IS_ERR(dst)) {
652 err = PTR_ERR(dst);
653 dst = NULL;
c12b395a 654 goto tx_err_link_failure;
655 }
cdf3464e 656 ndst = dst;
c12b395a 657 }
658
659 tdev = dst->dev;
660
661 if (tdev == dev) {
662 stats->collisions++;
663 net_warn_ratelimited("%s: Local routing loop detected!\n",
664 tunnel->parms.name);
665 goto tx_err_dst_release;
666 }
667
668 mtu = dst_mtu(dst) - sizeof(*ipv6h);
669 if (encap_limit >= 0) {
670 max_headroom += 8;
671 mtu -= 8;
672 }
673 if (mtu < IPV6_MIN_MTU)
674 mtu = IPV6_MIN_MTU;
675 if (skb_dst(skb))
676 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
677 if (skb->len > mtu) {
678 *pmtu = mtu;
679 err = -EMSGSIZE;
680 goto tx_err_dst_release;
681 }
682
683 if (tunnel->err_count > 0) {
684 if (time_before(jiffies,
685 tunnel->err_time + IP6TUNNEL_ERR_TIMEO)) {
686 tunnel->err_count--;
687
688 dst_link_failure(skb);
689 } else
690 tunnel->err_count = 0;
691 }
692
963a88b3
ND		 693 skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(dev)));
694
3da812d8 695 max_headroom += LL_RESERVED_SPACE(tdev) + gre_hlen + dst->header_len;
c12b395a 696
697 if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
698 (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
699 new_skb = skb_realloc_headroom(skb, max_headroom);
700 if (max_headroom > dev->needed_headroom)
701 dev->needed_headroom = max_headroom;
702 if (!new_skb)
703 goto tx_err_dst_release;
704
705 if (skb->sk)
706 skb_set_owner_w(new_skb, skb->sk);
707 consume_skb(skb);
708 skb = new_skb;
709 }
710
cdf3464e
MKL		 711 if (!fl6->flowi6_mark && ndst)
712 ip6_tnl_dst_set(tunnel, ndst);
713 skb_dst_set(skb, dst);
c12b395a 714
c12b395a 715 proto = NEXTHDR_GRE;
716 if (encap_limit >= 0) {
717 init_tel_txopt(&opt, encap_limit);
718 ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
719 }
720
3d483058
HFS		 721 if (likely(!skb->encapsulation)) {
722 skb_reset_inner_headers(skb);
723 skb->encapsulation = 1;
724 }
725
c12b395a 726 skb_push(skb, gre_hlen);
727 skb_reset_network_header(skb);
ae782bb1 728 skb_set_transport_header(skb, sizeof(*ipv6h));
c12b395a 729
730 /*
731 * Push down and install the IP header.
732 */
733 ipv6h = ipv6_hdr(skb);
cb1ce2ef 734 ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield),
42240901 735 ip6_make_flowlabel(net, skb, fl6->flowlabel, true, fl6));
c12b395a 736 ipv6h->hop_limit = tunnel->parms.hop_limit;
737 ipv6h->nexthdr = proto;
738 ipv6h->saddr = fl6->saddr;
739 ipv6h->daddr = fl6->daddr;
740
741 ((__be16 *)(ipv6h + 1))[0] = tunnel->parms.o_flags;
54bc9bac
TH		 742 protocol = (dev->type == ARPHRD_ETHER) ?
743 htons(ETH_P_TEB) : skb->protocol;
744 ((__be16 *)(ipv6h + 1))[1] = protocol;
c12b395a 745
746 if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
747 __be32 *ptr = (__be32 *)(((u8 *)ipv6h) + tunnel->hlen - 4);
748
749 if (tunnel->parms.o_flags&GRE_SEQ) {
750 ++tunnel->o_seqno;
751 *ptr = htonl(tunnel->o_seqno);
752 ptr--;
753 }
754 if (tunnel->parms.o_flags&GRE_KEY) {
755 *ptr = tunnel->parms.o_key;
756 ptr--;
757 }
758 if (tunnel->parms.o_flags&GRE_CSUM) {
759 *ptr = 0;
760 *(__sum16 *)ptr = ip_compute_csum((void *)(ipv6h+1),
761 skb->len - sizeof(struct ipv6hdr));
762 }
763 }
764
54bc9bac
TH		 765 skb_set_inner_protocol(skb, protocol);
766
79b16aad 767 ip6tunnel_xmit(NULL, skb, dev);
c12b395a 768 return 0;
769tx_err_link_failure:
770 stats->tx_carrier_errors++;
771 dst_link_failure(skb);
772tx_err_dst_release:
cdf3464e 773 dst_release(dst);
c12b395a 774 return err;
775}
776
777static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
778{
779 struct ip6_tnl *t = netdev_priv(dev);
780 const struct iphdr *iph = ip_hdr(skb);
781 int encap_limit = -1;
782 struct flowi6 fl6;
783 __u8 dsfield;
784 __u32 mtu;
785 int err;
786
207485dc
BH		 787 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
788
c12b395a 789 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
790 encap_limit = t->parms.encap_limit;
791
792 memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
3be07244 793 fl6.flowi6_proto = IPPROTO_GRE;
c12b395a 794
795 dsfield = ipv4_get_dsfield(iph);
796
797 if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
798 fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
799 & IPV6_TCLASS_MASK;
800 if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
801 fl6.flowi6_mark = skb->mark;
802
803 err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
804 if (err != 0) {
805 /* XXX: send ICMP error even if DF is not set. */
806 if (err == -EMSGSIZE)
807 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
808 htonl(mtu));
809 return -1;
810 }
811
812 return 0;
813}
814
815static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
816{
817 struct ip6_tnl *t = netdev_priv(dev);
818 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
819 int encap_limit = -1;
820 __u16 offset;
821 struct flowi6 fl6;
822 __u8 dsfield;
823 __u32 mtu;
824 int err;
825
826 if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
827 return -1;
828
829 offset = ip6_tnl_parse_tlv_enc_lim(skb, skb_network_header(skb));
830 if (offset > 0) {
831 struct ipv6_tlv_tnl_enc_lim *tel;
832 tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
833 if (tel->encap_limit == 0) {
834 icmpv6_send(skb, ICMPV6_PARAMPROB,
835 ICMPV6_HDR_FIELD, offset + 2);
836 return -1;
837 }
838 encap_limit = tel->encap_limit - 1;
839 } else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
840 encap_limit = t->parms.encap_limit;
841
842 memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
3be07244 843 fl6.flowi6_proto = IPPROTO_GRE;
c12b395a 844
845 dsfield = ipv6_get_dsfield(ipv6h);
846 if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
847 fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
848 if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
3308de2b 849 fl6.flowlabel |= ip6_flowlabel(ipv6h);
c12b395a 850 if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
851 fl6.flowi6_mark = skb->mark;
852
853 err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
854 if (err != 0) {
855 if (err == -EMSGSIZE)
856 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
857 return -1;
858 }
859
860 return 0;
861}
862
863/**
864 * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
865 * @t: the outgoing tunnel device
866 * @hdr: IPv6 header from the incoming packet
867 *
868 * Description:
869 * Avoid trivial tunneling loop by checking that tunnel exit-point
870 * doesn't match source of incoming packet.
871 *
872 * Return:
873 * 1 if conflict,
874 * 0 else
875 **/
876
877static inline bool ip6gre_tnl_addr_conflict(const struct ip6_tnl *t,
878 const struct ipv6hdr *hdr)
879{
880 return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
881}
882
883static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
884{
885 struct ip6_tnl *t = netdev_priv(dev);
886 int encap_limit = -1;
887 struct flowi6 fl6;
888 __u32 mtu;
889 int err;
890
891 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
892 encap_limit = t->parms.encap_limit;
893
894 memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
c12b395a 895
896 err = ip6gre_xmit2(skb, dev, 0, &fl6, encap_limit, &mtu);
897
898 return err;
899}
900
901static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
902 struct net_device *dev)
903{
904 struct ip6_tnl *t = netdev_priv(dev);
905 struct net_device_stats *stats = &t->dev->stats;
906 int ret;
907
d5005140 908 if (!ip6_tnl_xmit_ctl(t, &t->parms.laddr, &t->parms.raddr))
41ab3e31 909 goto tx_err;
c12b395a 910
911 switch (skb->protocol) {
912 case htons(ETH_P_IP):
913 ret = ip6gre_xmit_ipv4(skb, dev);
914 break;
915 case htons(ETH_P_IPV6):
916 ret = ip6gre_xmit_ipv6(skb, dev);
917 break;
918 default:
919 ret = ip6gre_xmit_other(skb, dev);
920 break;
921 }
922
923 if (ret < 0)
924 goto tx_err;
925
926 return NETDEV_TX_OK;
927
928tx_err:
929 stats->tx_errors++;
930 stats->tx_dropped++;
931 kfree_skb(skb);
932 return NETDEV_TX_OK;
933}
934
935static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
936{
937 struct net_device *dev = t->dev;
938 struct __ip6_tnl_parm *p = &t->parms;
939 struct flowi6 *fl6 = &t->fl.u.ip6;
940 int addend = sizeof(struct ipv6hdr) + 4;
941
942 if (dev->type != ARPHRD_ETHER) {
943 memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
944 memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
945 }
946
947 /* Set up flowi template */
948 fl6->saddr = p->laddr;
949 fl6->daddr = p->raddr;
950 fl6->flowi6_oif = p->link;
951 fl6->flowlabel = 0;
952
953 if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
954 fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
955 if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
956 fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
957
958 p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
959 p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
960
961 if (p->flags&IP6_TNL_F_CAP_XMIT &&
962 p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
963 dev->flags |= IFF_POINTOPOINT;
964 else
965 dev->flags &= ~IFF_POINTOPOINT;
966
c12b395a 967 /* Precalculate GRE options length */
968 if (t->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
969 if (t->parms.o_flags&GRE_CSUM)
970 addend += 4;
971 if (t->parms.o_flags&GRE_KEY)
972 addend += 4;
973 if (t->parms.o_flags&GRE_SEQ)
974 addend += 4;
975 }
bf581759 976 t->hlen = addend;
c12b395a 977
978 if (p->flags & IP6_TNL_F_CAP_XMIT) {
979 int strict = (ipv6_addr_type(&p->raddr) &
980 (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
981
22f08069 982 struct rt6_info *rt = rt6_lookup(t->net,
c12b395a 983 &p->raddr, &p->laddr,
984 p->link, strict);
985
63159f29 986 if (!rt)
c12b395a 987 return;
988
989 if (rt->dst.dev) {
990 dev->hard_header_len = rt->dst.dev->hard_header_len + addend;
991
992 if (set_mtu) {
993 dev->mtu = rt->dst.dev->mtu - addend;
994 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
995 dev->mtu -= 8;
996
997 if (dev->mtu < IPV6_MIN_MTU)
998 dev->mtu = IPV6_MIN_MTU;
999 }
1000 }
94e187c0 1001 ip6_rt_put(rt);
c12b395a 1002 }
c12b395a 1003}
1004
1005static int ip6gre_tnl_change(struct ip6_tnl *t,
1006 const struct __ip6_tnl_parm *p, int set_mtu)
1007{
1008 t->parms.laddr = p->laddr;
1009 t->parms.raddr = p->raddr;
1010 t->parms.flags = p->flags;
1011 t->parms.hop_limit = p->hop_limit;
1012 t->parms.encap_limit = p->encap_limit;
1013 t->parms.flowinfo = p->flowinfo;
1014 t->parms.link = p->link;
1015 t->parms.proto = p->proto;
1016 t->parms.i_key = p->i_key;
1017 t->parms.o_key = p->o_key;
1018 t->parms.i_flags = p->i_flags;
1019 t->parms.o_flags = p->o_flags;
1020 ip6_tnl_dst_reset(t);
1021 ip6gre_tnl_link_config(t, set_mtu);
1022 return 0;
1023}
1024
1025static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
1026 const struct ip6_tnl_parm2 *u)
1027{
1028 p->laddr = u->laddr;
1029 p->raddr = u->raddr;
1030 p->flags = u->flags;
1031 p->hop_limit = u->hop_limit;
1032 p->encap_limit = u->encap_limit;
1033 p->flowinfo = u->flowinfo;
1034 p->link = u->link;
1035 p->i_key = u->i_key;
1036 p->o_key = u->o_key;
1037 p->i_flags = u->i_flags;
1038 p->o_flags = u->o_flags;
1039 memcpy(p->name, u->name, sizeof(u->name));
1040}
1041
1042static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
1043 const struct __ip6_tnl_parm *p)
1044{
1045 u->proto = IPPROTO_GRE;
1046 u->laddr = p->laddr;
1047 u->raddr = p->raddr;
1048 u->flags = p->flags;
1049 u->hop_limit = p->hop_limit;
1050 u->encap_limit = p->encap_limit;
1051 u->flowinfo = p->flowinfo;
1052 u->link = p->link;
1053 u->i_key = p->i_key;
1054 u->o_key = p->o_key;
1055 u->i_flags = p->i_flags;
1056 u->o_flags = p->o_flags;
1057 memcpy(u->name, p->name, sizeof(u->name));
1058}
1059
1060static int ip6gre_tunnel_ioctl(struct net_device *dev,
1061 struct ifreq *ifr, int cmd)
1062{
1063 int err = 0;
1064 struct ip6_tnl_parm2 p;
1065 struct __ip6_tnl_parm p1;
22f08069
ND		 1066 struct ip6_tnl *t = netdev_priv(dev);
1067 struct net *net = t->net;
c12b395a 1068 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1069
1070 switch (cmd) {
1071 case SIOCGETTUNNEL:
c12b395a 1072 if (dev == ign->fb_tunnel_dev) {
1073 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1074 err = -EFAULT;
1075 break;
1076 }
1077 ip6gre_tnl_parm_from_user(&p1, &p);
1078 t = ip6gre_tunnel_locate(net, &p1, 0);
63159f29 1079 if (!t)
22f08069 1080 t = netdev_priv(dev);
c12b395a 1081 }
5dbd5068 1082 memset(&p, 0, sizeof(p));
c12b395a 1083 ip6gre_tnl_parm_to_user(&p, &t->parms);
1084 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1085 err = -EFAULT;
1086 break;
1087
1088 case SIOCADDTUNNEL:
1089 case SIOCCHGTUNNEL:
1090 err = -EPERM;
af31f412 1091 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
c12b395a 1092 goto done;
1093
1094 err = -EFAULT;
1095 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1096 goto done;
1097
1098 err = -EINVAL;
1099 if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
1100 goto done;
1101
1102 if (!(p.i_flags&GRE_KEY))
1103 p.i_key = 0;
1104 if (!(p.o_flags&GRE_KEY))
1105 p.o_key = 0;
1106
1107 ip6gre_tnl_parm_from_user(&p1, &p);
1108 t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
1109
1110 if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
53b24b8f 1111 if (t) {
c12b395a 1112 if (t->dev != dev) {
1113 err = -EEXIST;
1114 break;
1115 }
1116 } else {
1117 t = netdev_priv(dev);
1118
1119 ip6gre_tunnel_unlink(ign, t);
1120 synchronize_net();
1121 ip6gre_tnl_change(t, &p1, 1);
1122 ip6gre_tunnel_link(ign, t);
1123 netdev_state_change(dev);
1124 }
1125 }
1126
1127 if (t) {
1128 err = 0;
1129
5dbd5068 1130 memset(&p, 0, sizeof(p));
c12b395a 1131 ip6gre_tnl_parm_to_user(&p, &t->parms);
1132 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1133 err = -EFAULT;
1134 } else
1135 err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1136 break;
1137
1138 case SIOCDELTUNNEL:
1139 err = -EPERM;
af31f412 1140 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
c12b395a 1141 goto done;
1142
1143 if (dev == ign->fb_tunnel_dev) {
1144 err = -EFAULT;
1145 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1146 goto done;
1147 err = -ENOENT;
1148 ip6gre_tnl_parm_from_user(&p1, &p);
1149 t = ip6gre_tunnel_locate(net, &p1, 0);
63159f29 1150 if (!t)
c12b395a 1151 goto done;
1152 err = -EPERM;
1153 if (t == netdev_priv(ign->fb_tunnel_dev))
1154 goto done;
1155 dev = t->dev;
1156 }
1157 unregister_netdevice(dev);
1158 err = 0;
1159 break;
1160
1161 default:
1162 err = -EINVAL;
1163 }
1164
1165done:
1166 return err;
1167}
1168
1169static int ip6gre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1170{
c12b395a 1171 if (new_mtu < 68 ||
0e719e3a 1172 new_mtu > 0xFFF8 - dev->hard_header_len)
c12b395a 1173 return -EINVAL;
1174 dev->mtu = new_mtu;
1175 return 0;
1176}
1177
1178static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
3483c650
XL		 1179 unsigned short type, const void *daddr,
1180 const void *saddr, unsigned int len)
c12b395a 1181{
1182 struct ip6_tnl *t = netdev_priv(dev);
3483c650
XL		 1183 struct ipv6hdr *ipv6h;
1184 __be16 *p;
c12b395a 1185
3483c650
XL		 1186 ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen + sizeof(*ipv6h));
1187 ip6_flow_hdr(ipv6h, 0, ip6_make_flowlabel(dev_net(dev), skb,
1188 t->fl.u.ip6.flowlabel,
1189 true, &t->fl.u.ip6));
c12b395a 1190 ipv6h->hop_limit = t->parms.hop_limit;
1191 ipv6h->nexthdr = NEXTHDR_GRE;
1192 ipv6h->saddr = t->parms.laddr;
1193 ipv6h->daddr = t->parms.raddr;
1194
3483c650
XL		 1195 p = (__be16 *)(ipv6h + 1);
1196 p[0] = t->parms.o_flags;
1197 p[1] = htons(type);
c12b395a 1198
1199 /*
1200 * Set the source hardware address.
1201 */
1202
1203 if (saddr)
1204 memcpy(&ipv6h->saddr, saddr, sizeof(struct in6_addr));
1205 if (daddr)
1206 memcpy(&ipv6h->daddr, daddr, sizeof(struct in6_addr));
1207 if (!ipv6_addr_any(&ipv6h->daddr))
1208 return t->hlen;
1209
1210 return -t->hlen;
1211}
1212
c12b395a 1213static const struct header_ops ip6gre_header_ops = {
1214 .create = ip6gre_header,
c12b395a 1215};
1216
1217static const struct net_device_ops ip6gre_netdev_ops = {
1218 .ndo_init = ip6gre_tunnel_init,
1219 .ndo_uninit = ip6gre_tunnel_uninit,
1220 .ndo_start_xmit = ip6gre_tunnel_xmit,
1221 .ndo_do_ioctl = ip6gre_tunnel_ioctl,
1222 .ndo_change_mtu = ip6gre_tunnel_change_mtu,
f61dd388 1223 .ndo_get_stats64 = ip_tunnel_get_stats64,
ecf2c06a 1224 .ndo_get_iflink = ip6_tnl_get_iflink,
c12b395a 1225};
1226
1227static void ip6gre_dev_free(struct net_device *dev)
1228{
cdf3464e
MKL		 1229 struct ip6_tnl *t = netdev_priv(dev);
1230
1231 ip6_tnl_dst_destroy(t);
c12b395a 1232 free_percpu(dev->tstats);
1233 free_netdev(dev);
1234}
1235
1236static void ip6gre_tunnel_setup(struct net_device *dev)
1237{
1238 struct ip6_tnl *t;
1239
1240 dev->netdev_ops = &ip6gre_netdev_ops;
1241 dev->destructor = ip6gre_dev_free;
1242
1243 dev->type = ARPHRD_IP6GRE;
1244 dev->hard_header_len = LL_MAX_HEADER + sizeof(struct ipv6hdr) + 4;
1245 dev->mtu = ETH_DATA_LEN - sizeof(struct ipv6hdr) - 4;
1246 t = netdev_priv(dev);
1247 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1248 dev->mtu -= 8;
1249 dev->flags |= IFF_NOARP;
c12b395a 1250 dev->addr_len = sizeof(struct in6_addr);
02875878 1251 netif_keep_dst(dev);
c12b395a 1252}
1253
a3c119d3 1254static int ip6gre_tunnel_init_common(struct net_device *dev)
c12b395a 1255{
1256 struct ip6_tnl *tunnel;
cdf3464e 1257 int ret;
c12b395a 1258
1259 tunnel = netdev_priv(dev);
1260
1261 tunnel->dev = dev;
0bd87628 1262 tunnel->net = dev_net(dev);
c12b395a 1263 strcpy(tunnel->parms.name, dev->name);
1264
a3c119d3
MKL		 1265 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1266 if (!dev->tstats)
1267 return -ENOMEM;
1268
cdf3464e
MKL		 1269 ret = ip6_tnl_dst_init(tunnel);
1270 if (ret) {
1271 free_percpu(dev->tstats);
1272 dev->tstats = NULL;
1273 return ret;
1274 }
1275
a3c119d3
MKL		 1276 return 0;
1277}
1278
1279static int ip6gre_tunnel_init(struct net_device *dev)
1280{
1281 struct ip6_tnl *tunnel;
1282 int ret;
1283
1284 ret = ip6gre_tunnel_init_common(dev);
1285 if (ret)
1286 return ret;
1287
1288 tunnel = netdev_priv(dev);
1289
c12b395a 1290 memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1291 memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1292
1293 if (ipv6_addr_any(&tunnel->parms.raddr))
1294 dev->header_ops = &ip6gre_header_ops;
1295
c12b395a 1296 return 0;
1297}
1298
1299static void ip6gre_fb_tunnel_init(struct net_device *dev)
1300{
1301 struct ip6_tnl *tunnel = netdev_priv(dev);
1302
1303 tunnel->dev = dev;
0bd87628 1304 tunnel->net = dev_net(dev);
c12b395a 1305 strcpy(tunnel->parms.name, dev->name);
1306
1307 tunnel->hlen = sizeof(struct ipv6hdr) + 4;
1308
1309 dev_hold(dev);
1310}
1311
1312
1313static struct inet6_protocol ip6gre_protocol __read_mostly = {
1314 .handler = ip6gre_rcv,
1315 .err_handler = ip6gre_err,
1316 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1317};
1318
22f08069 1319static void ip6gre_destroy_tunnels(struct net *net, struct list_head *head)
c12b395a 1320{
22f08069
ND		 1321 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1322 struct net_device *dev, *aux;
c12b395a 1323 int prio;
1324
22f08069
ND		 1325 for_each_netdev_safe(net, dev, aux)
1326 if (dev->rtnl_link_ops == &ip6gre_link_ops ||
1327 dev->rtnl_link_ops == &ip6gre_tap_ops)
1328 unregister_netdevice_queue(dev, head);
1329
c12b395a 1330 for (prio = 0; prio < 4; prio++) {
1331 int h;
1332 for (h = 0; h < HASH_SIZE; h++) {
1333 struct ip6_tnl *t;
1334
1335 t = rtnl_dereference(ign->tunnels[prio][h]);
1336
53b24b8f 1337 while (t) {
22f08069
ND		 1338 /* If dev is in the same netns, it has already
1339 * been added to the list by the previous loop.
1340 */
1341 if (!net_eq(dev_net(t->dev), net))
1342 unregister_netdevice_queue(t->dev,
1343 head);
c12b395a 1344 t = rtnl_dereference(t->next);
1345 }
1346 }
1347 }
1348}
1349
1350static int __net_init ip6gre_init_net(struct net *net)
1351{
1352 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1353 int err;
1354
1355 ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
c835a677
TG		 1356 NET_NAME_UNKNOWN,
1357 ip6gre_tunnel_setup);
c12b395a 1358 if (!ign->fb_tunnel_dev) {
1359 err = -ENOMEM;
1360 goto err_alloc_dev;
1361 }
1362 dev_net_set(ign->fb_tunnel_dev, net);
22f08069
ND		 1363 /* FB netdevice is special: we have one, and only one per netns.
1364 * Allowing to move it to another netns is clearly unsafe.
1365 */
1366 ign->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1367
c12b395a 1368
1369 ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1370 ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1371
1372 err = register_netdev(ign->fb_tunnel_dev);
1373 if (err)
1374 goto err_reg_dev;
1375
1376 rcu_assign_pointer(ign->tunnels_wc[0],
1377 netdev_priv(ign->fb_tunnel_dev));
1378 return 0;
1379
1380err_reg_dev:
1381 ip6gre_dev_free(ign->fb_tunnel_dev);
1382err_alloc_dev:
1383 return err;
1384}
1385
1386static void __net_exit ip6gre_exit_net(struct net *net)
1387{
c12b395a 1388 LIST_HEAD(list);
1389
c12b395a 1390 rtnl_lock();
22f08069 1391 ip6gre_destroy_tunnels(net, &list);
c12b395a 1392 unregister_netdevice_many(&list);
1393 rtnl_unlock();
1394}
1395
1396static struct pernet_operations ip6gre_net_ops = {
1397 .init = ip6gre_init_net,
1398 .exit = ip6gre_exit_net,
1399 .id = &ip6gre_net_id,
1400 .size = sizeof(struct ip6gre_net),
1401};
1402
1403static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1404{
1405 __be16 flags;
1406
1407 if (!data)
1408 return 0;
1409
1410 flags = 0;
1411 if (data[IFLA_GRE_IFLAGS])
1412 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1413 if (data[IFLA_GRE_OFLAGS])
1414 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1415 if (flags & (GRE_VERSION|GRE_ROUTING))
1416 return -EINVAL;
1417
1418 return 0;
1419}
1420
1421static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1422{
1423 struct in6_addr daddr;
1424
1425 if (tb[IFLA_ADDRESS]) {
1426 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1427 return -EINVAL;
1428 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1429 return -EADDRNOTAVAIL;
1430 }
1431
1432 if (!data)
1433 goto out;
1434
1435 if (data[IFLA_GRE_REMOTE]) {
67b61f6c 1436 daddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
c12b395a 1437 if (ipv6_addr_any(&daddr))
1438 return -EINVAL;
1439 }
1440
1441out:
1442 return ip6gre_tunnel_validate(tb, data);
1443}
1444
1445
1446static void ip6gre_netlink_parms(struct nlattr *data[],
1447 struct __ip6_tnl_parm *parms)
1448{
1449 memset(parms, 0, sizeof(*parms));
1450
1451 if (!data)
1452 return;
1453
1454 if (data[IFLA_GRE_LINK])
1455 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1456
1457 if (data[IFLA_GRE_IFLAGS])
1458 parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1459
1460 if (data[IFLA_GRE_OFLAGS])
1461 parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1462
1463 if (data[IFLA_GRE_IKEY])
1464 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1465
1466 if (data[IFLA_GRE_OKEY])
1467 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1468
1469 if (data[IFLA_GRE_LOCAL])
67b61f6c 1470 parms->laddr = nla_get_in6_addr(data[IFLA_GRE_LOCAL]);
c12b395a 1471
1472 if (data[IFLA_GRE_REMOTE])
67b61f6c 1473 parms->raddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
c12b395a 1474
1475 if (data[IFLA_GRE_TTL])
1476 parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1477
1478 if (data[IFLA_GRE_ENCAP_LIMIT])
1479 parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1480
1481 if (data[IFLA_GRE_FLOWINFO])
1482 parms->flowinfo = nla_get_u32(data[IFLA_GRE_FLOWINFO]);
1483
1484 if (data[IFLA_GRE_FLAGS])
1485 parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1486}
1487
1488static int ip6gre_tap_init(struct net_device *dev)
1489{
1490 struct ip6_tnl *tunnel;
a3c119d3 1491 int ret;
c12b395a 1492
a3c119d3
MKL		 1493 ret = ip6gre_tunnel_init_common(dev);
1494 if (ret)
1495 return ret;
c12b395a 1496
a3c119d3 1497 tunnel = netdev_priv(dev);
c12b395a 1498
1499 ip6gre_tnl_link_config(tunnel, 1);
1500
c12b395a 1501 return 0;
1502}
1503
1504static const struct net_device_ops ip6gre_tap_netdev_ops = {
1505 .ndo_init = ip6gre_tap_init,
1506 .ndo_uninit = ip6gre_tunnel_uninit,
1507 .ndo_start_xmit = ip6gre_tunnel_xmit,
1508 .ndo_set_mac_address = eth_mac_addr,
1509 .ndo_validate_addr = eth_validate_addr,
1510 .ndo_change_mtu = ip6gre_tunnel_change_mtu,
f61dd388 1511 .ndo_get_stats64 = ip_tunnel_get_stats64,
ecf2c06a 1512 .ndo_get_iflink = ip6_tnl_get_iflink,
c12b395a 1513};
1514
1515static void ip6gre_tap_setup(struct net_device *dev)
1516{
1517
1518 ether_setup(dev);
1519
1520 dev->netdev_ops = &ip6gre_tap_netdev_ops;
1521 dev->destructor = ip6gre_dev_free;
1522
c12b395a 1523 dev->features |= NETIF_F_NETNS_LOCAL;
1524}
1525
1526static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1527 struct nlattr *tb[], struct nlattr *data[])
1528{
1529 struct ip6_tnl *nt;
1530 struct net *net = dev_net(dev);
1531 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1532 int err;
1533
1534 nt = netdev_priv(dev);
1535 ip6gre_netlink_parms(data, &nt->parms);
1536
1537 if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1538 return -EEXIST;
1539
1540 if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1541 eth_hw_addr_random(dev);
1542
1543 nt->dev = dev;
0bd87628 1544 nt->net = dev_net(dev);
c12b395a 1545 ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1546
1547 /* Can use a lockless transmit, unless we generate output sequences */
1548 if (!(nt->parms.o_flags & GRE_SEQ))
1549 dev->features |= NETIF_F_LLTX;
1550
1551 err = register_netdevice(dev);
1552 if (err)
1553 goto out;
1554
1555 dev_hold(dev);
1556 ip6gre_tunnel_link(ign, nt);
1557
1558out:
1559 return err;
1560}
1561
1562static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1563 struct nlattr *data[])
1564{
22f08069
ND		 1565 struct ip6_tnl *t, *nt = netdev_priv(dev);
1566 struct net *net = nt->net;
c12b395a 1567 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1568 struct __ip6_tnl_parm p;
1569
1570 if (dev == ign->fb_tunnel_dev)
1571 return -EINVAL;
1572
c12b395a 1573 ip6gre_netlink_parms(data, &p);
1574
1575 t = ip6gre_tunnel_locate(net, &p, 0);
1576
1577 if (t) {
1578 if (t->dev != dev)
1579 return -EEXIST;
1580 } else {
1581 t = nt;
c12b395a 1582 }
1583
6a61d4db
ND		 1584 ip6gre_tunnel_unlink(ign, t);
1585 ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1586 ip6gre_tunnel_link(ign, t);
c12b395a 1587 return 0;
1588}
1589
54d63f78
ND		 1590static void ip6gre_dellink(struct net_device *dev, struct list_head *head)
1591{
1592 struct net *net = dev_net(dev);
1593 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1594
1595 if (dev != ign->fb_tunnel_dev)
1596 unregister_netdevice_queue(dev, head);
1597}
1598
c12b395a 1599static size_t ip6gre_get_size(const struct net_device *dev)
1600{
1601 return
1602 /* IFLA_GRE_LINK */
1603 nla_total_size(4) +
1604 /* IFLA_GRE_IFLAGS */
1605 nla_total_size(2) +
1606 /* IFLA_GRE_OFLAGS */
1607 nla_total_size(2) +
1608 /* IFLA_GRE_IKEY */
1609 nla_total_size(4) +
1610 /* IFLA_GRE_OKEY */
1611 nla_total_size(4) +
1612 /* IFLA_GRE_LOCAL */
a3754133 1613 nla_total_size(sizeof(struct in6_addr)) +
c12b395a 1614 /* IFLA_GRE_REMOTE */
a3754133 1615 nla_total_size(sizeof(struct in6_addr)) +
c12b395a 1616 /* IFLA_GRE_TTL */
1617 nla_total_size(1) +
1618 /* IFLA_GRE_TOS */
1619 nla_total_size(1) +
1620 /* IFLA_GRE_ENCAP_LIMIT */
1621 nla_total_size(1) +
1622 /* IFLA_GRE_FLOWINFO */
1623 nla_total_size(4) +
1624 /* IFLA_GRE_FLAGS */
1625 nla_total_size(4) +
1626 0;
1627}
1628
1629static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1630{
1631 struct ip6_tnl *t = netdev_priv(dev);
1632 struct __ip6_tnl_parm *p = &t->parms;
1633
1634 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1635 nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1636 nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1637 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1638 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
930345ea
JB		 1639 nla_put_in6_addr(skb, IFLA_GRE_LOCAL, &p->laddr) ||
1640 nla_put_in6_addr(skb, IFLA_GRE_REMOTE, &p->raddr) ||
c12b395a 1641 nla_put_u8(skb, IFLA_GRE_TTL, p->hop_limit) ||
1642 /*nla_put_u8(skb, IFLA_GRE_TOS, t->priority) ||*/
1643 nla_put_u8(skb, IFLA_GRE_ENCAP_LIMIT, p->encap_limit) ||
1644 nla_put_be32(skb, IFLA_GRE_FLOWINFO, p->flowinfo) ||
1645 nla_put_u32(skb, IFLA_GRE_FLAGS, p->flags))
1646 goto nla_put_failure;
1647 return 0;
1648
1649nla_put_failure:
1650 return -EMSGSIZE;
1651}
1652
1653static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
1654 [IFLA_GRE_LINK] = { .type = NLA_U32 },
1655 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 },
1656 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 },
1657 [IFLA_GRE_IKEY] = { .type = NLA_U32 },
1658 [IFLA_GRE_OKEY] = { .type = NLA_U32 },
1659 [IFLA_GRE_LOCAL] = { .len = FIELD_SIZEOF(struct ipv6hdr, saddr) },
1660 [IFLA_GRE_REMOTE] = { .len = FIELD_SIZEOF(struct ipv6hdr, daddr) },
1661 [IFLA_GRE_TTL] = { .type = NLA_U8 },
1662 [IFLA_GRE_ENCAP_LIMIT] = { .type = NLA_U8 },
1663 [IFLA_GRE_FLOWINFO] = { .type = NLA_U32 },
1664 [IFLA_GRE_FLAGS] = { .type = NLA_U32 },
1665};
1666
1667static struct rtnl_link_ops ip6gre_link_ops __read_mostly = {
1668 .kind = "ip6gre",
1669 .maxtype = IFLA_GRE_MAX,
1670 .policy = ip6gre_policy,
1671 .priv_size = sizeof(struct ip6_tnl),
1672 .setup = ip6gre_tunnel_setup,
1673 .validate = ip6gre_tunnel_validate,
1674 .newlink = ip6gre_newlink,
1675 .changelink = ip6gre_changelink,
54d63f78 1676 .dellink = ip6gre_dellink,
c12b395a 1677 .get_size = ip6gre_get_size,
1678 .fill_info = ip6gre_fill_info,
1728d4fa 1679 .get_link_net = ip6_tnl_get_link_net,
c12b395a 1680};
1681
1682static struct rtnl_link_ops ip6gre_tap_ops __read_mostly = {
1683 .kind = "ip6gretap",
1684 .maxtype = IFLA_GRE_MAX,
1685 .policy = ip6gre_policy,
1686 .priv_size = sizeof(struct ip6_tnl),
1687 .setup = ip6gre_tap_setup,
1688 .validate = ip6gre_tap_validate,
1689 .newlink = ip6gre_newlink,
1690 .changelink = ip6gre_changelink,
1691 .get_size = ip6gre_get_size,
1692 .fill_info = ip6gre_fill_info,
3390e397 1693 .get_link_net = ip6_tnl_get_link_net,
c12b395a 1694};
1695
1696/*
1697 * And now the modules code and kernel interface.
1698 */
1699
1700static int __init ip6gre_init(void)
1701{
1702 int err;
1703
1704 pr_info("GRE over IPv6 tunneling driver\n");
1705
1706 err = register_pernet_device(&ip6gre_net_ops);
1707 if (err < 0)
1708 return err;
1709
1710 err = inet6_add_protocol(&ip6gre_protocol, IPPROTO_GRE);
1711 if (err < 0) {
1712 pr_info("%s: can't add protocol\n", __func__);
1713 goto add_proto_failed;
1714 }
1715
1716 err = rtnl_link_register(&ip6gre_link_ops);
1717 if (err < 0)
1718 goto rtnl_link_failed;
1719
1720 err = rtnl_link_register(&ip6gre_tap_ops);
1721 if (err < 0)
1722 goto tap_ops_failed;
1723
1724out:
1725 return err;
1726
1727tap_ops_failed:
1728 rtnl_link_unregister(&ip6gre_link_ops);
1729rtnl_link_failed:
1730 inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1731add_proto_failed:
1732 unregister_pernet_device(&ip6gre_net_ops);
1733 goto out;
1734}
1735
1736static void __exit ip6gre_fini(void)
1737{
1738 rtnl_link_unregister(&ip6gre_tap_ops);
1739 rtnl_link_unregister(&ip6gre_link_ops);
1740 inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1741 unregister_pernet_device(&ip6gre_net_ops);
1742}
1743
1744module_init(ip6gre_init);
1745module_exit(ip6gre_fini);
1746MODULE_LICENSE("GPL");
1747MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
1748MODULE_DESCRIPTION("GRE over IPv6 tunneling device");
1749MODULE_ALIAS_RTNL_LINK("ip6gre");
5a4ee9a9 1750MODULE_ALIAS_RTNL_LINK("ip6gretap");
c12b395a 1751MODULE_ALIAS_NETDEV("ip6gre0");