projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
GRE: Add support for GRO/GSO of IPv6 GRE traffic
[deliverable/linux.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>
c12b395a 27#include <linux/init.h>
28#include <linux/in6.h>
29#include <linux/inetdevice.h>
30#include <linux/igmp.h>
31#include <linux/netfilter_ipv4.h>
32#include <linux/etherdevice.h>
33#include <linux/if_ether.h>
34#include <linux/hash.h>
35#include <linux/if_tunnel.h>
36#include <linux/ip6_tunnel.h>
37
38#include <net/sock.h>
39#include <net/ip.h>
c5441932 40#include <net/ip_tunnels.h>
c12b395a 41#include <net/icmp.h>
42#include <net/protocol.h>
43#include <net/addrconf.h>
44#include <net/arp.h>
45#include <net/checksum.h>
46#include <net/dsfield.h>
47#include <net/inet_ecn.h>
48#include <net/xfrm.h>
49#include <net/net_namespace.h>
50#include <net/netns/generic.h>
51#include <net/rtnetlink.h>
52
53#include <net/ipv6.h>
54#include <net/ip6_fib.h>
55#include <net/ip6_route.h>
56#include <net/ip6_tunnel.h>
57
58
eccc1bb8 59static bool log_ecn_error = true;
60module_param(log_ecn_error, bool, 0644);
61MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
62
c12b395a 63#define HASH_SIZE_SHIFT 5
64#define HASH_SIZE (1 << HASH_SIZE_SHIFT)
65
66static int ip6gre_net_id __read_mostly;
67struct ip6gre_net {
68 struct ip6_tnl __rcu *tunnels[4][HASH_SIZE];
69
70 struct net_device *fb_tunnel_dev;
71};
72
73static struct rtnl_link_ops ip6gre_link_ops __read_mostly;
22f08069 74static struct rtnl_link_ops ip6gre_tap_ops __read_mostly;
c12b395a 75static int ip6gre_tunnel_init(struct net_device *dev);
76static void ip6gre_tunnel_setup(struct net_device *dev);
77static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t);
78static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu);
79
80/* Tunnel hash table */
81
82/*
83 4 hash tables:
84
85 3: (remote,local)
86 2: (remote,*)
87 1: (*,local)
88 0: (*,*)
89
90 We require exact key match i.e. if a key is present in packet
91 it will match only tunnel with the same key; if it is not present,
92 it will match only keyless tunnel.
93
94 All keysless packets, if not matched configured keyless tunnels
95 will match fallback tunnel.
96 */
97
98#define HASH_KEY(key) (((__force u32)key^((__force u32)key>>4))&(HASH_SIZE - 1))
99static u32 HASH_ADDR(const struct in6_addr *addr)
100{
101 u32 hash = ipv6_addr_hash(addr);
102
103 return hash_32(hash, HASH_SIZE_SHIFT);
104}
105
106#define tunnels_r_l tunnels[3]
107#define tunnels_r tunnels[2]
108#define tunnels_l tunnels[1]
109#define tunnels_wc tunnels[0]
c12b395a 110
c12b395a 111/* Given src, dst and key, find appropriate for input tunnel. */
112
113static struct ip6_tnl *ip6gre_tunnel_lookup(struct net_device *dev,
114 const struct in6_addr *remote, const struct in6_addr *local,
115 __be32 key, __be16 gre_proto)
116{
117 struct net *net = dev_net(dev);
118 int link = dev->ifindex;
119 unsigned int h0 = HASH_ADDR(remote);
120 unsigned int h1 = HASH_KEY(key);
121 struct ip6_tnl *t, *cand = NULL;
122 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
123 int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
124 ARPHRD_ETHER : ARPHRD_IP6GRE;
125 int score, cand_score = 4;
126
e086cadc 127 for_each_ip_tunnel_rcu(t, ign->tunnels_r_l[h0 ^ h1]) {
c12b395a 128 if (!ipv6_addr_equal(local, &t->parms.laddr) ||
129 !ipv6_addr_equal(remote, &t->parms.raddr) ||
130 key != t->parms.i_key ||
131 !(t->dev->flags & IFF_UP))
132 continue;
133
134 if (t->dev->type != ARPHRD_IP6GRE &&
135 t->dev->type != dev_type)
136 continue;
137
138 score = 0;
139 if (t->parms.link != link)
140 score |= 1;
141 if (t->dev->type != dev_type)
142 score |= 2;
143 if (score == 0)
144 return t;
145
146 if (score < cand_score) {
147 cand = t;
148 cand_score = score;
149 }
150 }
151
e086cadc 152 for_each_ip_tunnel_rcu(t, ign->tunnels_r[h0 ^ h1]) {
c12b395a 153 if (!ipv6_addr_equal(remote, &t->parms.raddr) ||
154 key != t->parms.i_key ||
155 !(t->dev->flags & IFF_UP))
156 continue;
157
158 if (t->dev->type != ARPHRD_IP6GRE &&
159 t->dev->type != dev_type)
160 continue;
161
162 score = 0;
163 if (t->parms.link != link)
164 score |= 1;
165 if (t->dev->type != dev_type)
166 score |= 2;
167 if (score == 0)
168 return t;
169
170 if (score < cand_score) {
171 cand = t;
172 cand_score = score;
173 }
174 }
175
e086cadc 176 for_each_ip_tunnel_rcu(t, ign->tunnels_l[h1]) {
c12b395a 177 if ((!ipv6_addr_equal(local, &t->parms.laddr) &&
178 (!ipv6_addr_equal(local, &t->parms.raddr) ||
179 !ipv6_addr_is_multicast(local))) ||
180 key != t->parms.i_key ||
181 !(t->dev->flags & IFF_UP))
182 continue;
183
184 if (t->dev->type != ARPHRD_IP6GRE &&
185 t->dev->type != dev_type)
186 continue;
187
188 score = 0;
189 if (t->parms.link != link)
190 score |= 1;
191 if (t->dev->type != dev_type)
192 score |= 2;
193 if (score == 0)
194 return t;
195
196 if (score < cand_score) {
197 cand = t;
198 cand_score = score;
199 }
200 }
201
e086cadc 202 for_each_ip_tunnel_rcu(t, ign->tunnels_wc[h1]) {
c12b395a 203 if (t->parms.i_key != key ||
204 !(t->dev->flags & IFF_UP))
205 continue;
206
207 if (t->dev->type != ARPHRD_IP6GRE &&
208 t->dev->type != dev_type)
209 continue;
210
211 score = 0;
212 if (t->parms.link != link)
213 score |= 1;
214 if (t->dev->type != dev_type)
215 score |= 2;
216 if (score == 0)
217 return t;
218
219 if (score < cand_score) {
220 cand = t;
221 cand_score = score;
222 }
223 }
224
53b24b8f 225 if (cand)
c12b395a 226 return cand;
227
228 dev = ign->fb_tunnel_dev;
229 if (dev->flags & IFF_UP)
230 return netdev_priv(dev);
231
232 return NULL;
233}
234
235static struct ip6_tnl __rcu **__ip6gre_bucket(struct ip6gre_net *ign,
236 const struct __ip6_tnl_parm *p)
237{
238 const struct in6_addr *remote = &p->raddr;
239 const struct in6_addr *local = &p->laddr;
240 unsigned int h = HASH_KEY(p->i_key);
241 int prio = 0;
242
243 if (!ipv6_addr_any(local))
244 prio |= 1;
245 if (!ipv6_addr_any(remote) && !ipv6_addr_is_multicast(remote)) {
246 prio |= 2;
247 h ^= HASH_ADDR(remote);
248 }
249
250 return &ign->tunnels[prio][h];
251}
252
253static inline struct ip6_tnl __rcu **ip6gre_bucket(struct ip6gre_net *ign,
254 const struct ip6_tnl *t)
255{
256 return __ip6gre_bucket(ign, &t->parms);
257}
258
259static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t)
260{
261 struct ip6_tnl __rcu **tp = ip6gre_bucket(ign, t);
262
263 rcu_assign_pointer(t->next, rtnl_dereference(*tp));
264 rcu_assign_pointer(*tp, t);
265}
266
267static void ip6gre_tunnel_unlink(struct ip6gre_net *ign, struct ip6_tnl *t)
268{
269 struct ip6_tnl __rcu **tp;
270 struct ip6_tnl *iter;
271
272 for (tp = ip6gre_bucket(ign, t);
273 (iter = rtnl_dereference(*tp)) != NULL;
274 tp = &iter->next) {
275 if (t == iter) {
276 rcu_assign_pointer(*tp, t->next);
277 break;
278 }
279 }
280}
281
282static struct ip6_tnl *ip6gre_tunnel_find(struct net *net,
283 const struct __ip6_tnl_parm *parms,
284 int type)
285{
286 const struct in6_addr *remote = &parms->raddr;
287 const struct in6_addr *local = &parms->laddr;
288 __be32 key = parms->i_key;
289 int link = parms->link;
290 struct ip6_tnl *t;
291 struct ip6_tnl __rcu **tp;
292 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
293
294 for (tp = __ip6gre_bucket(ign, parms);
295 (t = rtnl_dereference(*tp)) != NULL;
296 tp = &t->next)
297 if (ipv6_addr_equal(local, &t->parms.laddr) &&
298 ipv6_addr_equal(remote, &t->parms.raddr) &&
299 key == t->parms.i_key &&
300 link == t->parms.link &&
301 type == t->dev->type)
302 break;
303
304 return t;
305}
306
307static struct ip6_tnl *ip6gre_tunnel_locate(struct net *net,
308 const struct __ip6_tnl_parm *parms, int create)
309{
310 struct ip6_tnl *t, *nt;
311 struct net_device *dev;
312 char name[IFNAMSIZ];
313 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
314
315 t = ip6gre_tunnel_find(net, parms, ARPHRD_IP6GRE);
cd0a0bd9
SK		 316 if (t && create)
317 return NULL;
c12b395a 318 if (t || !create)
319 return t;
320
321 if (parms->name[0])
322 strlcpy(name, parms->name, IFNAMSIZ);
323 else
324 strcpy(name, "ip6gre%d");
325
c835a677
TG		 326 dev = alloc_netdev(sizeof(*t), name, NET_NAME_UNKNOWN,
327 ip6gre_tunnel_setup);
c12b395a 328 if (!dev)
329 return NULL;
330
331 dev_net_set(dev, net);
332
333 nt = netdev_priv(dev);
334 nt->parms = *parms;
335 dev->rtnl_link_ops = &ip6gre_link_ops;
336
337 nt->dev = dev;
0bd87628 338 nt->net = dev_net(dev);
c12b395a 339 ip6gre_tnl_link_config(nt, 1);
340
341 if (register_netdevice(dev) < 0)
342 goto failed_free;
343
344 /* Can use a lockless transmit, unless we generate output sequences */
345 if (!(nt->parms.o_flags & GRE_SEQ))
346 dev->features |= NETIF_F_LLTX;
347
348 dev_hold(dev);
349 ip6gre_tunnel_link(ign, nt);
350 return nt;
351
352failed_free:
353 free_netdev(dev);
354 return NULL;
355}
356
357static void ip6gre_tunnel_uninit(struct net_device *dev)
358{
22f08069
ND		 359 struct ip6_tnl *t = netdev_priv(dev);
360 struct ip6gre_net *ign = net_generic(t->net, ip6gre_net_id);
c12b395a 361
22f08069 362 ip6gre_tunnel_unlink(ign, t);
607f725f 363 dst_cache_reset(&t->dst_cache);
c12b395a 364 dev_put(dev);
365}
366
367
368static void ip6gre_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
369 u8 type, u8 code, int offset, __be32 info)
370{
371 const struct ipv6hdr *ipv6h = (const struct ipv6hdr *)skb->data;
b87fb39e
ED		 372 __be16 *p = (__be16 *)(skb->data + offset);
373 int grehlen = offset + 4;
c12b395a 374 struct ip6_tnl *t;
375 __be16 flags;
376
377 flags = p[0];
378 if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
379 if (flags&(GRE_VERSION|GRE_ROUTING))
380 return;
381 if (flags&GRE_KEY) {
382 grehlen += 4;
383 if (flags&GRE_CSUM)
384 grehlen += 4;
385 }
386 }
387
388 /* If only 8 bytes returned, keyed message will be dropped here */
b87fb39e 389 if (!pskb_may_pull(skb, grehlen))
c12b395a 390 return;
b87fb39e
ED		 391 ipv6h = (const struct ipv6hdr *)skb->data;
392 p = (__be16 *)(skb->data + offset);
c12b395a 393
c12b395a 394 t = ip6gre_tunnel_lookup(skb->dev, &ipv6h->daddr, &ipv6h->saddr,
395 flags & GRE_KEY ?
396 *(((__be32 *)p) + (grehlen / 4) - 1) : 0,
397 p[1]);
63159f29 398 if (!t)
0c5794a6 399 return;
c12b395a 400
401 switch (type) {
402 __u32 teli;
403 struct ipv6_tlv_tnl_enc_lim *tel;
404 __u32 mtu;
405 case ICMPV6_DEST_UNREACH:
a46496ce
MB		 406 net_dbg_ratelimited("%s: Path to destination invalid or inactive!\n",
407 t->parms.name);
c12b395a 408 break;
409 case ICMPV6_TIME_EXCEED:
410 if (code == ICMPV6_EXC_HOPLIMIT) {
a46496ce
MB		 411 net_dbg_ratelimited("%s: Too small hop limit or routing loop in tunnel!\n",
412 t->parms.name);
c12b395a 413 }
414 break;
415 case ICMPV6_PARAMPROB:
416 teli = 0;
417 if (code == ICMPV6_HDR_FIELD)
418 teli = ip6_tnl_parse_tlv_enc_lim(skb, skb->data);
419
d1e158e2 420 if (teli && teli == be32_to_cpu(info) - 2) {
c12b395a 421 tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
422 if (tel->encap_limit == 0) {
a46496ce
MB		 423 net_dbg_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
424 t->parms.name);
c12b395a 425 }
426 } else {
a46496ce
MB		 427 net_dbg_ratelimited("%s: Recipient unable to parse tunneled packet!\n",
428 t->parms.name);
c12b395a 429 }
430 break;
431 case ICMPV6_PKT_TOOBIG:
d1e158e2 432 mtu = be32_to_cpu(info) - offset;
c12b395a 433 if (mtu < IPV6_MIN_MTU)
434 mtu = IPV6_MIN_MTU;
435 t->dev->mtu = mtu;
436 break;
437 }
438
439 if (time_before(jiffies, t->err_time + IP6TUNNEL_ERR_TIMEO))
440 t->err_count++;
441 else
442 t->err_count = 1;
443 t->err_time = jiffies;
c12b395a 444}
445
c12b395a 446static int ip6gre_rcv(struct sk_buff *skb)
447{
448 const struct ipv6hdr *ipv6h;
449 u8 *h;
450 __be16 flags;
451 __sum16 csum = 0;
452 __be32 key = 0;
453 u32 seqno = 0;
454 struct ip6_tnl *tunnel;
455 int offset = 4;
456 __be16 gre_proto;
eccc1bb8 457 int err;
c12b395a 458
459 if (!pskb_may_pull(skb, sizeof(struct in6_addr)))
0c5794a6 460 goto drop;
c12b395a 461
462 ipv6h = ipv6_hdr(skb);
463 h = skb->data;
464 flags = *(__be16 *)h;
465
466 if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
467 /* - Version must be 0.
468 - We do not support routing headers.
469 */
470 if (flags&(GRE_VERSION|GRE_ROUTING))
0c5794a6 471 goto drop;
c12b395a 472
473 if (flags&GRE_CSUM) {
81249bea 474 csum = skb_checksum_simple_validate(skb);
c12b395a 475 offset += 4;
476 }
477 if (flags&GRE_KEY) {
478 key = *(__be32 *)(h + offset);
479 offset += 4;
480 }
481 if (flags&GRE_SEQ) {
482 seqno = ntohl(*(__be32 *)(h + offset));
483 offset += 4;
484 }
485 }
486
487 gre_proto = *(__be16 *)(h + 2);
488
c12b395a 489 tunnel = ip6gre_tunnel_lookup(skb->dev,
490 &ipv6h->saddr, &ipv6h->daddr, key,
491 gre_proto);
492 if (tunnel) {
8f84985f 493 struct pcpu_sw_netstats *tstats;
c12b395a 494
495 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
496 goto drop;
497
498 if (!ip6_tnl_rcv_ctl(tunnel, &ipv6h->daddr, &ipv6h->saddr)) {
499 tunnel->dev->stats.rx_dropped++;
500 goto drop;
501 }
502
c12b395a 503 skb->protocol = gre_proto;
504 /* WCCP version 1 and 2 protocol decoding.
be6572fd 505 * - Change protocol to IPv6
c12b395a 506 * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
507 */
508 if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
be6572fd 509 skb->protocol = htons(ETH_P_IPV6);
c12b395a 510 if ((*(h + offset) & 0xF0) != 0x40)
511 offset += 4;
512 }
513
514 skb->mac_header = skb->network_header;
515 __pskb_pull(skb, offset);
516 skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
c12b395a 517
518 if (((flags&GRE_CSUM) && csum) ||
519 (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
520 tunnel->dev->stats.rx_crc_errors++;
521 tunnel->dev->stats.rx_errors++;
522 goto drop;
523 }
524 if (tunnel->parms.i_flags&GRE_SEQ) {
525 if (!(flags&GRE_SEQ) ||
526 (tunnel->i_seqno &&
527 (s32)(seqno - tunnel->i_seqno) < 0)) {
528 tunnel->dev->stats.rx_fifo_errors++;
529 tunnel->dev->stats.rx_errors++;
530 goto drop;
531 }
532 tunnel->i_seqno = seqno + 1;
533 }
534
535 /* Warning: All skb pointers will be invalidated! */
536 if (tunnel->dev->type == ARPHRD_ETHER) {
537 if (!pskb_may_pull(skb, ETH_HLEN)) {
538 tunnel->dev->stats.rx_length_errors++;
539 tunnel->dev->stats.rx_errors++;
540 goto drop;
541 }
542
543 ipv6h = ipv6_hdr(skb);
544 skb->protocol = eth_type_trans(skb, tunnel->dev);
545 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
546 }
547
ea23192e 548 __skb_tunnel_rx(skb, tunnel->dev, tunnel->net);
eccc1bb8 549
550 skb_reset_network_header(skb);
551
552 err = IP6_ECN_decapsulate(ipv6h, skb);
553 if (unlikely(err)) {
554 if (log_ecn_error)
555 net_info_ratelimited("non-ECT from %pI6 with dsfield=%#x\n",
556 &ipv6h->saddr,
557 ipv6_get_dsfield(ipv6h));
558 if (err > 1) {
559 ++tunnel->dev->stats.rx_frame_errors;
560 ++tunnel->dev->stats.rx_errors;
561 goto drop;
562 }
563 }
564
c12b395a 565 tstats = this_cpu_ptr(tunnel->dev->tstats);
566 u64_stats_update_begin(&tstats->syncp);
567 tstats->rx_packets++;
568 tstats->rx_bytes += skb->len;
569 u64_stats_update_end(&tstats->syncp);
570
c12b395a 571 netif_rx(skb);
572
c12b395a 573 return 0;
574 }
575 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
576
577drop:
c12b395a 578 kfree_skb(skb);
579 return 0;
580}
581
582struct ipv6_tel_txoption {
583 struct ipv6_txoptions ops;
584 __u8 dst_opt[8];
585};
586
587static void init_tel_txopt(struct ipv6_tel_txoption *opt, __u8 encap_limit)
588{
589 memset(opt, 0, sizeof(struct ipv6_tel_txoption));
590
591 opt->dst_opt[2] = IPV6_TLV_TNL_ENCAP_LIMIT;
592 opt->dst_opt[3] = 1;
593 opt->dst_opt[4] = encap_limit;
594 opt->dst_opt[5] = IPV6_TLV_PADN;
595 opt->dst_opt[6] = 1;
596
597 opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
598 opt->ops.opt_nflen = 8;
599}
600
ac4eb009
AD		 601static __sum16 gre6_checksum(struct sk_buff *skb)
602{
603 __wsum csum;
604
605 if (skb->ip_summed == CHECKSUM_PARTIAL)
606 csum = lco_csum(skb);
607 else
608 csum = skb_checksum(skb, sizeof(struct ipv6hdr),
609 skb->len - sizeof(struct ipv6hdr), 0);
610 return csum_fold(csum);
611}
612
c12b395a 613static netdev_tx_t ip6gre_xmit2(struct sk_buff *skb,
614 struct net_device *dev,
615 __u8 dsfield,
616 struct flowi6 *fl6,
617 int encap_limit,
618 __u32 *pmtu)
619{
c12b395a 620 struct ip6_tnl *tunnel = netdev_priv(dev);
22f08069 621 struct net *net = tunnel->net;
c12b395a 622 struct net_device *tdev; /* Device to other host */
623 struct ipv6hdr *ipv6h; /* Our new IP header */
3da812d8 624 unsigned int max_headroom = 0; /* The extra header space needed */
c12b395a 625 int gre_hlen;
626 struct ipv6_tel_txoption opt;
627 int mtu;
628 struct dst_entry *dst = NULL, *ndst = NULL;
629 struct net_device_stats *stats = &tunnel->dev->stats;
630 int err = -1;
631 u8 proto;
c12b395a 632 struct sk_buff *new_skb;
54bc9bac 633 __be16 protocol;
c12b395a 634
635 if (dev->type == ARPHRD_ETHER)
636 IPCB(skb)->flags = 0;
637
638 if (dev->header_ops && dev->type == ARPHRD_IP6GRE) {
639 gre_hlen = 0;
640 ipv6h = (struct ipv6hdr *)skb->data;
641 fl6->daddr = ipv6h->daddr;
642 } else {
643 gre_hlen = tunnel->hlen;
644 fl6->daddr = tunnel->parms.raddr;
645 }
646
647 if (!fl6->flowi6_mark)
607f725f 648 dst = dst_cache_get(&tunnel->dst_cache);
c12b395a 649
650 if (!dst) {
cdf3464e 651 dst = ip6_route_output(net, NULL, fl6);
c12b395a 652
cdf3464e 653 if (dst->error)
c12b395a 654 goto tx_err_link_failure;
cdf3464e
MKL		 655 dst = xfrm_lookup(net, dst, flowi6_to_flowi(fl6), NULL, 0);
656 if (IS_ERR(dst)) {
657 err = PTR_ERR(dst);
658 dst = NULL;
c12b395a 659 goto tx_err_link_failure;
660 }
cdf3464e 661 ndst = dst;
c12b395a 662 }
663
664 tdev = dst->dev;
665
666 if (tdev == dev) {
667 stats->collisions++;
668 net_warn_ratelimited("%s: Local routing loop detected!\n",
669 tunnel->parms.name);
670 goto tx_err_dst_release;
671 }
672
673 mtu = dst_mtu(dst) - sizeof(*ipv6h);
674 if (encap_limit >= 0) {
675 max_headroom += 8;
676 mtu -= 8;
677 }
678 if (mtu < IPV6_MIN_MTU)
679 mtu = IPV6_MIN_MTU;
680 if (skb_dst(skb))
681 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
682 if (skb->len > mtu) {
683 *pmtu = mtu;
684 err = -EMSGSIZE;
685 goto tx_err_dst_release;
686 }
687
688 if (tunnel->err_count > 0) {
689 if (time_before(jiffies,
690 tunnel->err_time + IP6TUNNEL_ERR_TIMEO)) {
691 tunnel->err_count--;
692
693 dst_link_failure(skb);
694 } else
695 tunnel->err_count = 0;
696 }
697
963a88b3
ND		 698 skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(dev)));
699
3da812d8 700 max_headroom += LL_RESERVED_SPACE(tdev) + gre_hlen + dst->header_len;
c12b395a 701
702 if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
703 (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
704 new_skb = skb_realloc_headroom(skb, max_headroom);
705 if (max_headroom > dev->needed_headroom)
706 dev->needed_headroom = max_headroom;
707 if (!new_skb)
708 goto tx_err_dst_release;
709
710 if (skb->sk)
711 skb_set_owner_w(new_skb, skb->sk);
712 consume_skb(skb);
713 skb = new_skb;
714 }
715
cdf3464e 716 if (!fl6->flowi6_mark && ndst)
607f725f 717 dst_cache_set_ip6(&tunnel->dst_cache, ndst, &fl6->saddr);
cdf3464e 718 skb_dst_set(skb, dst);
c12b395a 719
c12b395a 720 proto = NEXTHDR_GRE;
721 if (encap_limit >= 0) {
722 init_tel_txopt(&opt, encap_limit);
723 ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
724 }
725
3d483058
HFS		 726 if (likely(!skb->encapsulation)) {
727 skb_reset_inner_headers(skb);
728 skb->encapsulation = 1;
729 }
730
c12b395a 731 skb_push(skb, gre_hlen);
732 skb_reset_network_header(skb);
ae782bb1 733 skb_set_transport_header(skb, sizeof(*ipv6h));
c12b395a 734
735 /*
736 * Push down and install the IP header.
737 */
738 ipv6h = ipv6_hdr(skb);
cb1ce2ef 739 ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield),
42240901 740 ip6_make_flowlabel(net, skb, fl6->flowlabel, true, fl6));
c12b395a 741 ipv6h->hop_limit = tunnel->parms.hop_limit;
742 ipv6h->nexthdr = proto;
743 ipv6h->saddr = fl6->saddr;
744 ipv6h->daddr = fl6->daddr;
745
746 ((__be16 *)(ipv6h + 1))[0] = tunnel->parms.o_flags;
54bc9bac
TH		 747 protocol = (dev->type == ARPHRD_ETHER) ?
748 htons(ETH_P_TEB) : skb->protocol;
749 ((__be16 *)(ipv6h + 1))[1] = protocol;
c12b395a 750
751 if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
752 __be32 *ptr = (__be32 *)(((u8 *)ipv6h) + tunnel->hlen - 4);
753
754 if (tunnel->parms.o_flags&GRE_SEQ) {
755 ++tunnel->o_seqno;
756 *ptr = htonl(tunnel->o_seqno);
757 ptr--;
758 }
759 if (tunnel->parms.o_flags&GRE_KEY) {
760 *ptr = tunnel->parms.o_key;
761 ptr--;
762 }
763 if (tunnel->parms.o_flags&GRE_CSUM) {
764 *ptr = 0;
ac4eb009 765 *(__sum16 *)ptr = gre6_checksum(skb);
c12b395a 766 }
767 }
768
54bc9bac
TH		 769 skb_set_inner_protocol(skb, protocol);
770
79b16aad 771 ip6tunnel_xmit(NULL, skb, dev);
c12b395a 772 return 0;
773tx_err_link_failure:
774 stats->tx_carrier_errors++;
775 dst_link_failure(skb);
776tx_err_dst_release:
cdf3464e 777 dst_release(dst);
c12b395a 778 return err;
779}
780
781static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
782{
783 struct ip6_tnl *t = netdev_priv(dev);
784 const struct iphdr *iph = ip_hdr(skb);
785 int encap_limit = -1;
786 struct flowi6 fl6;
787 __u8 dsfield;
788 __u32 mtu;
789 int err;
790
5146d1f1
BH		 791 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
792
c12b395a 793 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
794 encap_limit = t->parms.encap_limit;
795
796 memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
3be07244 797 fl6.flowi6_proto = IPPROTO_GRE;
c12b395a 798
799 dsfield = ipv4_get_dsfield(iph);
800
801 if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
802 fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
803 & IPV6_TCLASS_MASK;
804 if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
805 fl6.flowi6_mark = skb->mark;
806
807 err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
808 if (err != 0) {
809 /* XXX: send ICMP error even if DF is not set. */
810 if (err == -EMSGSIZE)
811 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
812 htonl(mtu));
813 return -1;
814 }
815
816 return 0;
817}
818
819static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
820{
821 struct ip6_tnl *t = netdev_priv(dev);
822 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
823 int encap_limit = -1;
824 __u16 offset;
825 struct flowi6 fl6;
826 __u8 dsfield;
827 __u32 mtu;
828 int err;
829
830 if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
831 return -1;
832
833 offset = ip6_tnl_parse_tlv_enc_lim(skb, skb_network_header(skb));
834 if (offset > 0) {
835 struct ipv6_tlv_tnl_enc_lim *tel;
836 tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
837 if (tel->encap_limit == 0) {
838 icmpv6_send(skb, ICMPV6_PARAMPROB,
839 ICMPV6_HDR_FIELD, offset + 2);
840 return -1;
841 }
842 encap_limit = tel->encap_limit - 1;
843 } else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
844 encap_limit = t->parms.encap_limit;
845
846 memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
3be07244 847 fl6.flowi6_proto = IPPROTO_GRE;
c12b395a 848
849 dsfield = ipv6_get_dsfield(ipv6h);
850 if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
851 fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
852 if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
3308de2b 853 fl6.flowlabel |= ip6_flowlabel(ipv6h);
c12b395a 854 if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
855 fl6.flowi6_mark = skb->mark;
856
857 err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
858 if (err != 0) {
859 if (err == -EMSGSIZE)
860 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
861 return -1;
862 }
863
864 return 0;
865}
866
867/**
868 * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
869 * @t: the outgoing tunnel device
870 * @hdr: IPv6 header from the incoming packet
871 *
872 * Description:
873 * Avoid trivial tunneling loop by checking that tunnel exit-point
874 * doesn't match source of incoming packet.
875 *
876 * Return:
877 * 1 if conflict,
878 * 0 else
879 **/
880
881static inline bool ip6gre_tnl_addr_conflict(const struct ip6_tnl *t,
882 const struct ipv6hdr *hdr)
883{
884 return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
885}
886
887static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
888{
889 struct ip6_tnl *t = netdev_priv(dev);
890 int encap_limit = -1;
891 struct flowi6 fl6;
892 __u32 mtu;
893 int err;
894
895 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
896 encap_limit = t->parms.encap_limit;
897
898 memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
899 fl6.flowi6_proto = skb->protocol;
900
901 err = ip6gre_xmit2(skb, dev, 0, &fl6, encap_limit, &mtu);
902
903 return err;
904}
905
906static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
907 struct net_device *dev)
908{
909 struct ip6_tnl *t = netdev_priv(dev);
910 struct net_device_stats *stats = &t->dev->stats;
911 int ret;
912
d5005140 913 if (!ip6_tnl_xmit_ctl(t, &t->parms.laddr, &t->parms.raddr))
41ab3e31 914 goto tx_err;
c12b395a 915
916 switch (skb->protocol) {
917 case htons(ETH_P_IP):
918 ret = ip6gre_xmit_ipv4(skb, dev);
919 break;
920 case htons(ETH_P_IPV6):
921 ret = ip6gre_xmit_ipv6(skb, dev);
922 break;
923 default:
924 ret = ip6gre_xmit_other(skb, dev);
925 break;
926 }
927
928 if (ret < 0)
929 goto tx_err;
930
931 return NETDEV_TX_OK;
932
933tx_err:
934 stats->tx_errors++;
935 stats->tx_dropped++;
936 kfree_skb(skb);
937 return NETDEV_TX_OK;
938}
939
940static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
941{
942 struct net_device *dev = t->dev;
943 struct __ip6_tnl_parm *p = &t->parms;
944 struct flowi6 *fl6 = &t->fl.u.ip6;
945 int addend = sizeof(struct ipv6hdr) + 4;
946
947 if (dev->type != ARPHRD_ETHER) {
948 memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
949 memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
950 }
951
952 /* Set up flowi template */
953 fl6->saddr = p->laddr;
954 fl6->daddr = p->raddr;
955 fl6->flowi6_oif = p->link;
956 fl6->flowlabel = 0;
957
958 if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
959 fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
960 if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
961 fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
962
963 p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
964 p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
965
966 if (p->flags&IP6_TNL_F_CAP_XMIT &&
967 p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
968 dev->flags |= IFF_POINTOPOINT;
969 else
970 dev->flags &= ~IFF_POINTOPOINT;
971
c12b395a 972 /* Precalculate GRE options length */
973 if (t->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
974 if (t->parms.o_flags&GRE_CSUM)
975 addend += 4;
976 if (t->parms.o_flags&GRE_KEY)
977 addend += 4;
978 if (t->parms.o_flags&GRE_SEQ)
979 addend += 4;
980 }
bf581759 981 t->hlen = addend;
c12b395a 982
983 if (p->flags & IP6_TNL_F_CAP_XMIT) {
984 int strict = (ipv6_addr_type(&p->raddr) &
985 (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
986
22f08069 987 struct rt6_info *rt = rt6_lookup(t->net,
c12b395a 988 &p->raddr, &p->laddr,
989 p->link, strict);
990
63159f29 991 if (!rt)
c12b395a 992 return;
993
994 if (rt->dst.dev) {
995 dev->hard_header_len = rt->dst.dev->hard_header_len + addend;
996
997 if (set_mtu) {
998 dev->mtu = rt->dst.dev->mtu - addend;
999 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1000 dev->mtu -= 8;
a9e242ca
AD		 1001 if (dev->type == ARPHRD_ETHER)
1002 dev->mtu -= ETH_HLEN;
c12b395a 1003
1004 if (dev->mtu < IPV6_MIN_MTU)
1005 dev->mtu = IPV6_MIN_MTU;
1006 }
1007 }
94e187c0 1008 ip6_rt_put(rt);
c12b395a 1009 }
c12b395a 1010}
1011
1012static int ip6gre_tnl_change(struct ip6_tnl *t,
1013 const struct __ip6_tnl_parm *p, int set_mtu)
1014{
1015 t->parms.laddr = p->laddr;
1016 t->parms.raddr = p->raddr;
1017 t->parms.flags = p->flags;
1018 t->parms.hop_limit = p->hop_limit;
1019 t->parms.encap_limit = p->encap_limit;
1020 t->parms.flowinfo = p->flowinfo;
1021 t->parms.link = p->link;
1022 t->parms.proto = p->proto;
1023 t->parms.i_key = p->i_key;
1024 t->parms.o_key = p->o_key;
1025 t->parms.i_flags = p->i_flags;
1026 t->parms.o_flags = p->o_flags;
607f725f 1027 dst_cache_reset(&t->dst_cache);
c12b395a 1028 ip6gre_tnl_link_config(t, set_mtu);
1029 return 0;
1030}
1031
1032static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
1033 const struct ip6_tnl_parm2 *u)
1034{
1035 p->laddr = u->laddr;
1036 p->raddr = u->raddr;
1037 p->flags = u->flags;
1038 p->hop_limit = u->hop_limit;
1039 p->encap_limit = u->encap_limit;
1040 p->flowinfo = u->flowinfo;
1041 p->link = u->link;
1042 p->i_key = u->i_key;
1043 p->o_key = u->o_key;
1044 p->i_flags = u->i_flags;
1045 p->o_flags = u->o_flags;
1046 memcpy(p->name, u->name, sizeof(u->name));
1047}
1048
1049static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
1050 const struct __ip6_tnl_parm *p)
1051{
1052 u->proto = IPPROTO_GRE;
1053 u->laddr = p->laddr;
1054 u->raddr = p->raddr;
1055 u->flags = p->flags;
1056 u->hop_limit = p->hop_limit;
1057 u->encap_limit = p->encap_limit;
1058 u->flowinfo = p->flowinfo;
1059 u->link = p->link;
1060 u->i_key = p->i_key;
1061 u->o_key = p->o_key;
1062 u->i_flags = p->i_flags;
1063 u->o_flags = p->o_flags;
1064 memcpy(u->name, p->name, sizeof(u->name));
1065}
1066
1067static int ip6gre_tunnel_ioctl(struct net_device *dev,
1068 struct ifreq *ifr, int cmd)
1069{
1070 int err = 0;
1071 struct ip6_tnl_parm2 p;
1072 struct __ip6_tnl_parm p1;
22f08069
ND		 1073 struct ip6_tnl *t = netdev_priv(dev);
1074 struct net *net = t->net;
c12b395a 1075 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1076
1077 switch (cmd) {
1078 case SIOCGETTUNNEL:
c12b395a 1079 if (dev == ign->fb_tunnel_dev) {
1080 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1081 err = -EFAULT;
1082 break;
1083 }
1084 ip6gre_tnl_parm_from_user(&p1, &p);
1085 t = ip6gre_tunnel_locate(net, &p1, 0);
63159f29 1086 if (!t)
22f08069 1087 t = netdev_priv(dev);
c12b395a 1088 }
5dbd5068 1089 memset(&p, 0, sizeof(p));
c12b395a 1090 ip6gre_tnl_parm_to_user(&p, &t->parms);
1091 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1092 err = -EFAULT;
1093 break;
1094
1095 case SIOCADDTUNNEL:
1096 case SIOCCHGTUNNEL:
1097 err = -EPERM;
af31f412 1098 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
c12b395a 1099 goto done;
1100
1101 err = -EFAULT;
1102 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1103 goto done;
1104
1105 err = -EINVAL;
1106 if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
1107 goto done;
1108
1109 if (!(p.i_flags&GRE_KEY))
1110 p.i_key = 0;
1111 if (!(p.o_flags&GRE_KEY))
1112 p.o_key = 0;
1113
1114 ip6gre_tnl_parm_from_user(&p1, &p);
1115 t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
1116
1117 if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
53b24b8f 1118 if (t) {
c12b395a 1119 if (t->dev != dev) {
1120 err = -EEXIST;
1121 break;
1122 }
1123 } else {
1124 t = netdev_priv(dev);
1125
1126 ip6gre_tunnel_unlink(ign, t);
1127 synchronize_net();
1128 ip6gre_tnl_change(t, &p1, 1);
1129 ip6gre_tunnel_link(ign, t);
1130 netdev_state_change(dev);
1131 }
1132 }
1133
1134 if (t) {
1135 err = 0;
1136
5dbd5068 1137 memset(&p, 0, sizeof(p));
c12b395a 1138 ip6gre_tnl_parm_to_user(&p, &t->parms);
1139 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1140 err = -EFAULT;
1141 } else
1142 err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1143 break;
1144
1145 case SIOCDELTUNNEL:
1146 err = -EPERM;
af31f412 1147 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
c12b395a 1148 goto done;
1149
1150 if (dev == ign->fb_tunnel_dev) {
1151 err = -EFAULT;
1152 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1153 goto done;
1154 err = -ENOENT;
1155 ip6gre_tnl_parm_from_user(&p1, &p);
1156 t = ip6gre_tunnel_locate(net, &p1, 0);
63159f29 1157 if (!t)
c12b395a 1158 goto done;
1159 err = -EPERM;
1160 if (t == netdev_priv(ign->fb_tunnel_dev))
1161 goto done;
1162 dev = t->dev;
1163 }
1164 unregister_netdevice(dev);
1165 err = 0;
1166 break;
1167
1168 default:
1169 err = -EINVAL;
1170 }
1171
1172done:
1173 return err;
1174}
1175
1176static int ip6gre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1177{
c12b395a 1178 if (new_mtu < 68 ||
0e719e3a 1179 new_mtu > 0xFFF8 - dev->hard_header_len)
c12b395a 1180 return -EINVAL;
1181 dev->mtu = new_mtu;
1182 return 0;
1183}
1184
1185static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
1186 unsigned short type,
1187 const void *daddr, const void *saddr, unsigned int len)
1188{
1189 struct ip6_tnl *t = netdev_priv(dev);
1190 struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);
1191 __be16 *p = (__be16 *)(ipv6h+1);
1192
cb1ce2ef
TH		 1193 ip6_flow_hdr(ipv6h, 0,
1194 ip6_make_flowlabel(dev_net(dev), skb,
42240901 1195 t->fl.u.ip6.flowlabel, true,
67800f9b 1196 &t->fl.u.ip6));
c12b395a 1197 ipv6h->hop_limit = t->parms.hop_limit;
1198 ipv6h->nexthdr = NEXTHDR_GRE;
1199 ipv6h->saddr = t->parms.laddr;
1200 ipv6h->daddr = t->parms.raddr;
1201
1202 p[0] = t->parms.o_flags;
1203 p[1] = htons(type);
1204
1205 /*
1206 * Set the source hardware address.
1207 */
1208
1209 if (saddr)
1210 memcpy(&ipv6h->saddr, saddr, sizeof(struct in6_addr));
1211 if (daddr)
1212 memcpy(&ipv6h->daddr, daddr, sizeof(struct in6_addr));
1213 if (!ipv6_addr_any(&ipv6h->daddr))
1214 return t->hlen;
1215
1216 return -t->hlen;
1217}
1218
c12b395a 1219static const struct header_ops ip6gre_header_ops = {
1220 .create = ip6gre_header,
c12b395a 1221};
1222
1223static const struct net_device_ops ip6gre_netdev_ops = {
1224 .ndo_init = ip6gre_tunnel_init,
1225 .ndo_uninit = ip6gre_tunnel_uninit,
1226 .ndo_start_xmit = ip6gre_tunnel_xmit,
1227 .ndo_do_ioctl = ip6gre_tunnel_ioctl,
1228 .ndo_change_mtu = ip6gre_tunnel_change_mtu,
f61dd388 1229 .ndo_get_stats64 = ip_tunnel_get_stats64,
ecf2c06a 1230 .ndo_get_iflink = ip6_tnl_get_iflink,
c12b395a 1231};
1232
1233static void ip6gre_dev_free(struct net_device *dev)
1234{
cdf3464e
MKL		 1235 struct ip6_tnl *t = netdev_priv(dev);
1236
607f725f 1237 dst_cache_destroy(&t->dst_cache);
c12b395a 1238 free_percpu(dev->tstats);
1239 free_netdev(dev);
1240}
1241
1242static void ip6gre_tunnel_setup(struct net_device *dev)
1243{
1244 struct ip6_tnl *t;
1245
1246 dev->netdev_ops = &ip6gre_netdev_ops;
1247 dev->destructor = ip6gre_dev_free;
1248
1249 dev->type = ARPHRD_IP6GRE;
1250 dev->hard_header_len = LL_MAX_HEADER + sizeof(struct ipv6hdr) + 4;
1251 dev->mtu = ETH_DATA_LEN - sizeof(struct ipv6hdr) - 4;
1252 t = netdev_priv(dev);
1253 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1254 dev->mtu -= 8;
1255 dev->flags |= IFF_NOARP;
c12b395a 1256 dev->addr_len = sizeof(struct in6_addr);
02875878 1257 netif_keep_dst(dev);
c12b395a 1258}
1259
a3c119d3 1260static int ip6gre_tunnel_init_common(struct net_device *dev)
c12b395a 1261{
1262 struct ip6_tnl *tunnel;
cdf3464e 1263 int ret;
c12b395a 1264
1265 tunnel = netdev_priv(dev);
1266
1267 tunnel->dev = dev;
0bd87628 1268 tunnel->net = dev_net(dev);
c12b395a 1269 strcpy(tunnel->parms.name, dev->name);
1270
a3c119d3
MKL		 1271 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1272 if (!dev->tstats)
1273 return -ENOMEM;
1274
607f725f 1275 ret = dst_cache_init(&tunnel->dst_cache, GFP_KERNEL);
cdf3464e
MKL		 1276 if (ret) {
1277 free_percpu(dev->tstats);
1278 dev->tstats = NULL;
1279 return ret;
1280 }
1281
a3c119d3
MKL		 1282 return 0;
1283}
1284
1285static int ip6gre_tunnel_init(struct net_device *dev)
1286{
1287 struct ip6_tnl *tunnel;
1288 int ret;
1289
1290 ret = ip6gre_tunnel_init_common(dev);
1291 if (ret)
1292 return ret;
1293
1294 tunnel = netdev_priv(dev);
1295
c12b395a 1296 memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1297 memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1298
1299 if (ipv6_addr_any(&tunnel->parms.raddr))
1300 dev->header_ops = &ip6gre_header_ops;
1301
c12b395a 1302 return 0;
1303}
1304
1305static void ip6gre_fb_tunnel_init(struct net_device *dev)
1306{
1307 struct ip6_tnl *tunnel = netdev_priv(dev);
1308
1309 tunnel->dev = dev;
0bd87628 1310 tunnel->net = dev_net(dev);
c12b395a 1311 strcpy(tunnel->parms.name, dev->name);
1312
1313 tunnel->hlen = sizeof(struct ipv6hdr) + 4;
1314
1315 dev_hold(dev);
1316}
1317
1318
1319static struct inet6_protocol ip6gre_protocol __read_mostly = {
1320 .handler = ip6gre_rcv,
1321 .err_handler = ip6gre_err,
1322 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1323};
1324
22f08069 1325static void ip6gre_destroy_tunnels(struct net *net, struct list_head *head)
c12b395a 1326{
22f08069
ND		 1327 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1328 struct net_device *dev, *aux;
c12b395a 1329 int prio;
1330
22f08069
ND		 1331 for_each_netdev_safe(net, dev, aux)
1332 if (dev->rtnl_link_ops == &ip6gre_link_ops ||
1333 dev->rtnl_link_ops == &ip6gre_tap_ops)
1334 unregister_netdevice_queue(dev, head);
1335
c12b395a 1336 for (prio = 0; prio < 4; prio++) {
1337 int h;
1338 for (h = 0; h < HASH_SIZE; h++) {
1339 struct ip6_tnl *t;
1340
1341 t = rtnl_dereference(ign->tunnels[prio][h]);
1342
53b24b8f 1343 while (t) {
22f08069
ND		 1344 /* If dev is in the same netns, it has already
1345 * been added to the list by the previous loop.
1346 */
1347 if (!net_eq(dev_net(t->dev), net))
1348 unregister_netdevice_queue(t->dev,
1349 head);
c12b395a 1350 t = rtnl_dereference(t->next);
1351 }
1352 }
1353 }
1354}
1355
1356static int __net_init ip6gre_init_net(struct net *net)
1357{
1358 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1359 int err;
1360
1361 ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
c835a677
TG		 1362 NET_NAME_UNKNOWN,
1363 ip6gre_tunnel_setup);
c12b395a 1364 if (!ign->fb_tunnel_dev) {
1365 err = -ENOMEM;
1366 goto err_alloc_dev;
1367 }
1368 dev_net_set(ign->fb_tunnel_dev, net);
22f08069
ND		 1369 /* FB netdevice is special: we have one, and only one per netns.
1370 * Allowing to move it to another netns is clearly unsafe.
1371 */
1372 ign->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1373
c12b395a 1374
1375 ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1376 ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1377
1378 err = register_netdev(ign->fb_tunnel_dev);
1379 if (err)
1380 goto err_reg_dev;
1381
1382 rcu_assign_pointer(ign->tunnels_wc[0],
1383 netdev_priv(ign->fb_tunnel_dev));
1384 return 0;
1385
1386err_reg_dev:
1387 ip6gre_dev_free(ign->fb_tunnel_dev);
1388err_alloc_dev:
1389 return err;
1390}
1391
1392static void __net_exit ip6gre_exit_net(struct net *net)
1393{
c12b395a 1394 LIST_HEAD(list);
1395
c12b395a 1396 rtnl_lock();
22f08069 1397 ip6gre_destroy_tunnels(net, &list);
c12b395a 1398 unregister_netdevice_many(&list);
1399 rtnl_unlock();
1400}
1401
1402static struct pernet_operations ip6gre_net_ops = {
1403 .init = ip6gre_init_net,
1404 .exit = ip6gre_exit_net,
1405 .id = &ip6gre_net_id,
1406 .size = sizeof(struct ip6gre_net),
1407};
1408
1409static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1410{
1411 __be16 flags;
1412
1413 if (!data)
1414 return 0;
1415
1416 flags = 0;
1417 if (data[IFLA_GRE_IFLAGS])
1418 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1419 if (data[IFLA_GRE_OFLAGS])
1420 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1421 if (flags & (GRE_VERSION|GRE_ROUTING))
1422 return -EINVAL;
1423
1424 return 0;
1425}
1426
1427static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1428{
1429 struct in6_addr daddr;
1430
1431 if (tb[IFLA_ADDRESS]) {
1432 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1433 return -EINVAL;
1434 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1435 return -EADDRNOTAVAIL;
1436 }
1437
1438 if (!data)
1439 goto out;
1440
1441 if (data[IFLA_GRE_REMOTE]) {
67b61f6c 1442 daddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
c12b395a 1443 if (ipv6_addr_any(&daddr))
1444 return -EINVAL;
1445 }
1446
1447out:
1448 return ip6gre_tunnel_validate(tb, data);
1449}
1450
1451
1452static void ip6gre_netlink_parms(struct nlattr *data[],
1453 struct __ip6_tnl_parm *parms)
1454{
1455 memset(parms, 0, sizeof(*parms));
1456
1457 if (!data)
1458 return;
1459
1460 if (data[IFLA_GRE_LINK])
1461 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1462
1463 if (data[IFLA_GRE_IFLAGS])
1464 parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1465
1466 if (data[IFLA_GRE_OFLAGS])
1467 parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1468
1469 if (data[IFLA_GRE_IKEY])
1470 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1471
1472 if (data[IFLA_GRE_OKEY])
1473 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1474
1475 if (data[IFLA_GRE_LOCAL])
67b61f6c 1476 parms->laddr = nla_get_in6_addr(data[IFLA_GRE_LOCAL]);
c12b395a 1477
1478 if (data[IFLA_GRE_REMOTE])
67b61f6c 1479 parms->raddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
c12b395a 1480
1481 if (data[IFLA_GRE_TTL])
1482 parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1483
1484 if (data[IFLA_GRE_ENCAP_LIMIT])
1485 parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1486
1487 if (data[IFLA_GRE_FLOWINFO])
1488 parms->flowinfo = nla_get_u32(data[IFLA_GRE_FLOWINFO]);
1489
1490 if (data[IFLA_GRE_FLAGS])
1491 parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1492}
1493
1494static int ip6gre_tap_init(struct net_device *dev)
1495{
1496 struct ip6_tnl *tunnel;
a3c119d3 1497 int ret;
c12b395a 1498
a3c119d3
MKL		 1499 ret = ip6gre_tunnel_init_common(dev);
1500 if (ret)
1501 return ret;
c12b395a 1502
a3c119d3 1503 tunnel = netdev_priv(dev);
c12b395a 1504
1505 ip6gre_tnl_link_config(tunnel, 1);
1506
c12b395a 1507 return 0;
1508}
1509
1510static const struct net_device_ops ip6gre_tap_netdev_ops = {
1511 .ndo_init = ip6gre_tap_init,
1512 .ndo_uninit = ip6gre_tunnel_uninit,
1513 .ndo_start_xmit = ip6gre_tunnel_xmit,
1514 .ndo_set_mac_address = eth_mac_addr,
1515 .ndo_validate_addr = eth_validate_addr,
1516 .ndo_change_mtu = ip6gre_tunnel_change_mtu,
f61dd388 1517 .ndo_get_stats64 = ip_tunnel_get_stats64,
ecf2c06a 1518 .ndo_get_iflink = ip6_tnl_get_iflink,
c12b395a 1519};
1520
ac4eb009
AD		 1521#define GRE6_FEATURES (NETIF_F_SG | \
1522 NETIF_F_FRAGLIST | \
1523 NETIF_F_HIGHDMA | \
1524 NETIF_F_HW_CSUM)
1525
c12b395a 1526static void ip6gre_tap_setup(struct net_device *dev)
1527{
1528
1529 ether_setup(dev);
1530
1531 dev->netdev_ops = &ip6gre_tap_netdev_ops;
1532 dev->destructor = ip6gre_dev_free;
1533
c12b395a 1534 dev->features |= NETIF_F_NETNS_LOCAL;
d13b161c 1535 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
c12b395a 1536}
1537
1538static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1539 struct nlattr *tb[], struct nlattr *data[])
1540{
1541 struct ip6_tnl *nt;
1542 struct net *net = dev_net(dev);
1543 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1544 int err;
1545
1546 nt = netdev_priv(dev);
1547 ip6gre_netlink_parms(data, &nt->parms);
1548
1549 if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1550 return -EEXIST;
1551
1552 if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1553 eth_hw_addr_random(dev);
1554
1555 nt->dev = dev;
0bd87628 1556 nt->net = dev_net(dev);
c12b395a 1557 ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1558
ac4eb009
AD		 1559 dev->features |= GRE6_FEATURES;
1560 dev->hw_features |= GRE6_FEATURES;
1561
c12b395a 1562 /* Can use a lockless transmit, unless we generate output sequences */
1563 if (!(nt->parms.o_flags & GRE_SEQ))
1564 dev->features |= NETIF_F_LLTX;
1565
1566 err = register_netdevice(dev);
1567 if (err)
1568 goto out;
1569
1570 dev_hold(dev);
1571 ip6gre_tunnel_link(ign, nt);
1572
1573out:
1574 return err;
1575}
1576
1577static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1578 struct nlattr *data[])
1579{
22f08069
ND		 1580 struct ip6_tnl *t, *nt = netdev_priv(dev);
1581 struct net *net = nt->net;
c12b395a 1582 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1583 struct __ip6_tnl_parm p;
1584
1585 if (dev == ign->fb_tunnel_dev)
1586 return -EINVAL;
1587
c12b395a 1588 ip6gre_netlink_parms(data, &p);
1589
1590 t = ip6gre_tunnel_locate(net, &p, 0);
1591
1592 if (t) {
1593 if (t->dev != dev)
1594 return -EEXIST;
1595 } else {
1596 t = nt;
c12b395a 1597 }
1598
6a61d4db
ND		 1599 ip6gre_tunnel_unlink(ign, t);
1600 ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1601 ip6gre_tunnel_link(ign, t);
c12b395a 1602 return 0;
1603}
1604
54d63f78
ND		 1605static void ip6gre_dellink(struct net_device *dev, struct list_head *head)
1606{
1607 struct net *net = dev_net(dev);
1608 struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1609
1610 if (dev != ign->fb_tunnel_dev)
1611 unregister_netdevice_queue(dev, head);
1612}
1613
c12b395a 1614static size_t ip6gre_get_size(const struct net_device *dev)
1615{
1616 return
1617 /* IFLA_GRE_LINK */
1618 nla_total_size(4) +
1619 /* IFLA_GRE_IFLAGS */
1620 nla_total_size(2) +
1621 /* IFLA_GRE_OFLAGS */
1622 nla_total_size(2) +
1623 /* IFLA_GRE_IKEY */
1624 nla_total_size(4) +
1625 /* IFLA_GRE_OKEY */
1626 nla_total_size(4) +
1627 /* IFLA_GRE_LOCAL */
a3754133 1628 nla_total_size(sizeof(struct in6_addr)) +
c12b395a 1629 /* IFLA_GRE_REMOTE */
a3754133 1630 nla_total_size(sizeof(struct in6_addr)) +
c12b395a 1631 /* IFLA_GRE_TTL */
1632 nla_total_size(1) +
1633 /* IFLA_GRE_TOS */
1634 nla_total_size(1) +
1635 /* IFLA_GRE_ENCAP_LIMIT */
1636 nla_total_size(1) +
1637 /* IFLA_GRE_FLOWINFO */
1638 nla_total_size(4) +
1639 /* IFLA_GRE_FLAGS */
1640 nla_total_size(4) +
1641 0;
1642}
1643
1644static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1645{
1646 struct ip6_tnl *t = netdev_priv(dev);
1647 struct __ip6_tnl_parm *p = &t->parms;
1648
1649 if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1650 nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1651 nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1652 nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1653 nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
930345ea
JB		 1654 nla_put_in6_addr(skb, IFLA_GRE_LOCAL, &p->laddr) ||
1655 nla_put_in6_addr(skb, IFLA_GRE_REMOTE, &p->raddr) ||
c12b395a 1656 nla_put_u8(skb, IFLA_GRE_TTL, p->hop_limit) ||
1657 /*nla_put_u8(skb, IFLA_GRE_TOS, t->priority) ||*/
1658 nla_put_u8(skb, IFLA_GRE_ENCAP_LIMIT, p->encap_limit) ||
1659 nla_put_be32(skb, IFLA_GRE_FLOWINFO, p->flowinfo) ||
1660 nla_put_u32(skb, IFLA_GRE_FLAGS, p->flags))
1661 goto nla_put_failure;
1662 return 0;
1663
1664nla_put_failure:
1665 return -EMSGSIZE;
1666}
1667
1668static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
1669 [IFLA_GRE_LINK] = { .type = NLA_U32 },
1670 [IFLA_GRE_IFLAGS] = { .type = NLA_U16 },
1671 [IFLA_GRE_OFLAGS] = { .type = NLA_U16 },
1672 [IFLA_GRE_IKEY] = { .type = NLA_U32 },
1673 [IFLA_GRE_OKEY] = { .type = NLA_U32 },
1674 [IFLA_GRE_LOCAL] = { .len = FIELD_SIZEOF(struct ipv6hdr, saddr) },
1675 [IFLA_GRE_REMOTE] = { .len = FIELD_SIZEOF(struct ipv6hdr, daddr) },
1676 [IFLA_GRE_TTL] = { .type = NLA_U8 },
1677 [IFLA_GRE_ENCAP_LIMIT] = { .type = NLA_U8 },
1678 [IFLA_GRE_FLOWINFO] = { .type = NLA_U32 },
1679 [IFLA_GRE_FLAGS] = { .type = NLA_U32 },
1680};
1681
1682static struct rtnl_link_ops ip6gre_link_ops __read_mostly = {
1683 .kind = "ip6gre",
1684 .maxtype = IFLA_GRE_MAX,
1685 .policy = ip6gre_policy,
1686 .priv_size = sizeof(struct ip6_tnl),
1687 .setup = ip6gre_tunnel_setup,
1688 .validate = ip6gre_tunnel_validate,
1689 .newlink = ip6gre_newlink,
1690 .changelink = ip6gre_changelink,
54d63f78 1691 .dellink = ip6gre_dellink,
c12b395a 1692 .get_size = ip6gre_get_size,
1693 .fill_info = ip6gre_fill_info,
1728d4fa 1694 .get_link_net = ip6_tnl_get_link_net,
c12b395a 1695};
1696
1697static struct rtnl_link_ops ip6gre_tap_ops __read_mostly = {
1698 .kind = "ip6gretap",
1699 .maxtype = IFLA_GRE_MAX,
1700 .policy = ip6gre_policy,
1701 .priv_size = sizeof(struct ip6_tnl),
1702 .setup = ip6gre_tap_setup,
1703 .validate = ip6gre_tap_validate,
1704 .newlink = ip6gre_newlink,
1705 .changelink = ip6gre_changelink,
1706 .get_size = ip6gre_get_size,
1707 .fill_info = ip6gre_fill_info,
3390e397 1708 .get_link_net = ip6_tnl_get_link_net,
c12b395a 1709};
1710
1711/*
1712 * And now the modules code and kernel interface.
1713 */
1714
1715static int __init ip6gre_init(void)
1716{
1717 int err;
1718
1719 pr_info("GRE over IPv6 tunneling driver\n");
1720
1721 err = register_pernet_device(&ip6gre_net_ops);
1722 if (err < 0)
1723 return err;
1724
1725 err = inet6_add_protocol(&ip6gre_protocol, IPPROTO_GRE);
1726 if (err < 0) {
1727 pr_info("%s: can't add protocol\n", __func__);
1728 goto add_proto_failed;
1729 }
1730
1731 err = rtnl_link_register(&ip6gre_link_ops);
1732 if (err < 0)
1733 goto rtnl_link_failed;
1734
1735 err = rtnl_link_register(&ip6gre_tap_ops);
1736 if (err < 0)
1737 goto tap_ops_failed;
1738
1739out:
1740 return err;
1741
1742tap_ops_failed:
1743 rtnl_link_unregister(&ip6gre_link_ops);
1744rtnl_link_failed:
1745 inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1746add_proto_failed:
1747 unregister_pernet_device(&ip6gre_net_ops);
1748 goto out;
1749}
1750
1751static void __exit ip6gre_fini(void)
1752{
1753 rtnl_link_unregister(&ip6gre_tap_ops);
1754 rtnl_link_unregister(&ip6gre_link_ops);
1755 inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1756 unregister_pernet_device(&ip6gre_net_ops);
1757}
1758
1759module_init(ip6gre_init);
1760module_exit(ip6gre_fini);
1761MODULE_LICENSE("GPL");
1762MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
1763MODULE_DESCRIPTION("GRE over IPv6 tunneling device");
1764MODULE_ALIAS_RTNL_LINK("ip6gre");
5a4ee9a9 1765MODULE_ALIAS_RTNL_LINK("ip6gretap");
c12b395a 1766MODULE_ALIAS_NETDEV("ip6gre0");
Linux kernel - Deliverables
This page took 0.465567 seconds and 5 git commands to generate.