ip6_tunnel.c 34.8 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
2
 *	IPv6 tunneling device
L
Linus Torvalds 已提交
3 4 5
 *	Linux INET6 implementation
 *
 *	Authors:
6
 *	Ville Nuorvala		<vnuorval@tcs.hut.fi>
7
 *	Yasuyuki Kozakai	<kozakai@linux-ipv6.org>
L
Linus Torvalds 已提交
8 9 10 11
 *
 *	$Id$
 *
 *      Based on:
12
 *      linux/net/ipv6/sit.c and linux/net/ipv4/ipip.c
L
Linus Torvalds 已提交
13 14 15 16 17 18 19 20 21 22 23
 *
 *      RFC 2473
 *
 *	This program is free software; you can redistribute it and/or
 *      modify it under the terms of the GNU General Public License
 *      as published by the Free Software Foundation; either version
 *      2 of the License, or (at your option) any later version.
 *
 */

#include <linux/module.h>
24
#include <linux/capability.h>
L
Linus Torvalds 已提交
25 26 27
#include <linux/errno.h>
#include <linux/types.h>
#include <linux/sockios.h>
28
#include <linux/icmp.h>
L
Linus Torvalds 已提交
29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
#include <linux/if.h>
#include <linux/in.h>
#include <linux/ip.h>
#include <linux/if_tunnel.h>
#include <linux/net.h>
#include <linux/in6.h>
#include <linux/netdevice.h>
#include <linux/if_arp.h>
#include <linux/icmpv6.h>
#include <linux/init.h>
#include <linux/route.h>
#include <linux/rtnetlink.h>
#include <linux/netfilter_ipv6.h>

#include <asm/uaccess.h>
#include <asm/atomic.h>

46
#include <net/icmp.h>
L
Linus Torvalds 已提交
47 48 49 50 51 52 53 54
#include <net/ip.h>
#include <net/ipv6.h>
#include <net/ip6_route.h>
#include <net/addrconf.h>
#include <net/ip6_tunnel.h>
#include <net/xfrm.h>
#include <net/dsfield.h>
#include <net/inet_ecn.h>
55 56
#include <net/net_namespace.h>
#include <net/netns/generic.h>
L
Linus Torvalds 已提交
57 58

MODULE_AUTHOR("Ville Nuorvala");
59
MODULE_DESCRIPTION("IPv6 tunneling device");
L
Linus Torvalds 已提交
60 61 62 63 64
MODULE_LICENSE("GPL");

#define IPV6_TLV_TEL_DST_SIZE 8

#ifdef IP6_TNL_DEBUG
65
#define IP6_TNL_TRACE(x...) printk(KERN_DEBUG "%s:" x "\n", __func__)
L
Linus Torvalds 已提交
66 67 68 69 70
#else
#define IP6_TNL_TRACE(x...) do {;} while(0)
#endif

#define IPV6_TCLASS_MASK (IPV6_FLOWINFO_MASK & ~IPV6_FLOWLABEL_MASK)
71
#define IPV6_TCLASS_SHIFT 20
L
Linus Torvalds 已提交
72 73 74

#define HASH_SIZE  32

A
Al Viro 已提交
75
#define HASH(addr) ((__force u32)((addr)->s6_addr32[0] ^ (addr)->s6_addr32[1] ^ \
76 77
		     (addr)->s6_addr32[2] ^ (addr)->s6_addr32[3]) & \
		    (HASH_SIZE - 1))
L
Linus Torvalds 已提交
78

79 80 81
static int ip6_fb_tnl_dev_init(struct net_device *dev);
static int ip6_tnl_dev_init(struct net_device *dev);
static void ip6_tnl_dev_setup(struct net_device *dev);
L
Linus Torvalds 已提交
82

83 84 85 86
static int ip6_tnl_net_id;
struct ip6_tnl_net {
};

L
Linus Torvalds 已提交
87
/* the IPv6 tunnel fallback device */
88
static struct net_device *ip6_fb_tnl_dev;
L
Linus Torvalds 已提交
89 90 91 92 93 94 95 96


/* lists for storing tunnels in use */
static struct ip6_tnl *tnls_r_l[HASH_SIZE];
static struct ip6_tnl *tnls_wc[1];
static struct ip6_tnl **tnls[2] = { tnls_wc, tnls_r_l };

/* lock for the tunnel lists */
97
static DEFINE_RWLOCK(ip6_tnl_lock);
L
Linus Torvalds 已提交
98 99 100 101 102

static inline struct dst_entry *ip6_tnl_dst_check(struct ip6_tnl *t)
{
	struct dst_entry *dst = t->dst_cache;

103
	if (dst && dst->obsolete &&
L
Linus Torvalds 已提交
104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127
	    dst->ops->check(dst, t->dst_cookie) == NULL) {
		t->dst_cache = NULL;
		dst_release(dst);
		return NULL;
	}

	return dst;
}

static inline void ip6_tnl_dst_reset(struct ip6_tnl *t)
{
	dst_release(t->dst_cache);
	t->dst_cache = NULL;
}

static inline void ip6_tnl_dst_store(struct ip6_tnl *t, struct dst_entry *dst)
{
	struct rt6_info *rt = (struct rt6_info *) dst;
	t->dst_cookie = rt->rt6i_node ? rt->rt6i_node->fn_sernum : 0;
	dst_release(t->dst_cache);
	t->dst_cache = dst;
}

/**
128
 * ip6_tnl_lookup - fetch tunnel matching the end-point addresses
129 130
 *   @remote: the address of the tunnel exit-point
 *   @local: the address of the tunnel entry-point
L
Linus Torvalds 已提交
131
 *
132
 * Return:
L
Linus Torvalds 已提交
133
 *   tunnel matching given end-points if found,
134
 *   else fallback tunnel if its device is up,
L
Linus Torvalds 已提交
135 136 137 138
 *   else %NULL
 **/

static struct ip6_tnl *
139
ip6_tnl_lookup(struct net *net, struct in6_addr *remote, struct in6_addr *local)
L
Linus Torvalds 已提交
140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157
{
	unsigned h0 = HASH(remote);
	unsigned h1 = HASH(local);
	struct ip6_tnl *t;

	for (t = tnls_r_l[h0 ^ h1]; t; t = t->next) {
		if (ipv6_addr_equal(local, &t->parms.laddr) &&
		    ipv6_addr_equal(remote, &t->parms.raddr) &&
		    (t->dev->flags & IFF_UP))
			return t;
	}
	if ((t = tnls_wc[0]) != NULL && (t->dev->flags & IFF_UP))
		return t;

	return NULL;
}

/**
158
 * ip6_tnl_bucket - get head of list matching given tunnel parameters
159
 *   @p: parameters containing tunnel end-points
L
Linus Torvalds 已提交
160 161
 *
 * Description:
162
 *   ip6_tnl_bucket() returns the head of the list matching the
L
Linus Torvalds 已提交
163 164
 *   &struct in6_addr entries laddr and raddr in @p.
 *
165
 * Return: head of IPv6 tunnel list
L
Linus Torvalds 已提交
166 167 168
 **/

static struct ip6_tnl **
169
ip6_tnl_bucket(struct ip6_tnl_net *ip6n, struct ip6_tnl_parm *p)
L
Linus Torvalds 已提交
170 171 172 173 174 175 176 177 178 179 180 181 182 183
{
	struct in6_addr *remote = &p->raddr;
	struct in6_addr *local = &p->laddr;
	unsigned h = 0;
	int prio = 0;

	if (!ipv6_addr_any(remote) || !ipv6_addr_any(local)) {
		prio = 1;
		h = HASH(remote) ^ HASH(local);
	}
	return &tnls[prio][h];
}

/**
184
 * ip6_tnl_link - add tunnel to hash table
L
Linus Torvalds 已提交
185 186 187 188
 *   @t: tunnel to be added
 **/

static void
189
ip6_tnl_link(struct ip6_tnl_net *ip6n, struct ip6_tnl *t)
L
Linus Torvalds 已提交
190
{
191
	struct ip6_tnl **tp = ip6_tnl_bucket(ip6n, &t->parms);
L
Linus Torvalds 已提交
192 193

	t->next = *tp;
194
	write_lock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
195
	*tp = t;
196
	write_unlock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
197 198 199
}

/**
200
 * ip6_tnl_unlink - remove tunnel from hash table
L
Linus Torvalds 已提交
201 202 203 204
 *   @t: tunnel to be removed
 **/

static void
205
ip6_tnl_unlink(struct ip6_tnl_net *ip6n, struct ip6_tnl *t)
L
Linus Torvalds 已提交
206 207 208
{
	struct ip6_tnl **tp;

209
	for (tp = ip6_tnl_bucket(ip6n, &t->parms); *tp; tp = &(*tp)->next) {
L
Linus Torvalds 已提交
210
		if (t == *tp) {
211
			write_lock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
212
			*tp = t->next;
213
			write_unlock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
214 215 216 217 218 219 220 221 222 223 224 225
			break;
		}
	}
}

/**
 * ip6_tnl_create() - create a new tunnel
 *   @p: tunnel parameters
 *   @pt: pointer to new tunnel
 *
 * Description:
 *   Create tunnel matching given parameters.
226 227
 *
 * Return:
228
 *   created tunnel or NULL
L
Linus Torvalds 已提交
229 230
 **/

231
static struct ip6_tnl *ip6_tnl_create(struct net *net, struct ip6_tnl_parm *p)
L
Linus Torvalds 已提交
232 233 234 235 236
{
	struct net_device *dev;
	struct ip6_tnl *t;
	char name[IFNAMSIZ];
	int err;
237
	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
L
Linus Torvalds 已提交
238

239
	if (p->name[0])
L
Linus Torvalds 已提交
240
		strlcpy(name, p->name, IFNAMSIZ);
241 242 243
	else
		sprintf(name, "ip6tnl%%d");

244
	dev = alloc_netdev(sizeof (*t), name, ip6_tnl_dev_setup);
L
Linus Torvalds 已提交
245
	if (dev == NULL)
246
		goto failed;
L
Linus Torvalds 已提交
247

248 249 250 251 252
	if (strchr(name, '%')) {
		if (dev_alloc_name(dev, name) < 0)
			goto failed_free;
	}

253
	t = netdev_priv(dev);
254
	dev->init = ip6_tnl_dev_init;
L
Linus Torvalds 已提交
255 256
	t->parms = *p;

257 258 259
	if ((err = register_netdevice(dev)) < 0)
		goto failed_free;

L
Linus Torvalds 已提交
260
	dev_hold(dev);
261
	ip6_tnl_link(ip6n, t);
262
	return t;
263 264 265

failed_free:
	free_netdev(dev);
266 267
failed:
	return NULL;
L
Linus Torvalds 已提交
268 269 270
}

/**
271
 * ip6_tnl_locate - find or create tunnel matching given parameters
272
 *   @p: tunnel parameters
L
Linus Torvalds 已提交
273 274 275
 *   @create: != 0 if allowed to create new tunnel if no match found
 *
 * Description:
276
 *   ip6_tnl_locate() first tries to locate an existing tunnel
L
Linus Torvalds 已提交
277 278 279 280
 *   based on @parms. If this is unsuccessful, but @create is set a new
 *   tunnel device is created and registered for use.
 *
 * Return:
281
 *   matching tunnel or NULL
L
Linus Torvalds 已提交
282 283
 **/

284 285
static struct ip6_tnl *ip6_tnl_locate(struct net *net,
		struct ip6_tnl_parm *p, int create)
L
Linus Torvalds 已提交
286 287 288 289
{
	struct in6_addr *remote = &p->raddr;
	struct in6_addr *local = &p->laddr;
	struct ip6_tnl *t;
290
	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
L
Linus Torvalds 已提交
291

292
	for (t = *ip6_tnl_bucket(ip6n, p); t; t = t->next) {
L
Linus Torvalds 已提交
293
		if (ipv6_addr_equal(local, &t->parms.laddr) &&
294 295
		    ipv6_addr_equal(remote, &t->parms.raddr))
			return t;
L
Linus Torvalds 已提交
296 297
	}
	if (!create)
298
		return NULL;
299
	return ip6_tnl_create(net, p);
L
Linus Torvalds 已提交
300 301 302
}

/**
303
 * ip6_tnl_dev_uninit - tunnel device uninitializer
L
Linus Torvalds 已提交
304
 *   @dev: the device to be destroyed
305
 *
L
Linus Torvalds 已提交
306
 * Description:
307
 *   ip6_tnl_dev_uninit() removes tunnel from its list
L
Linus Torvalds 已提交
308 309 310
 **/

static void
311
ip6_tnl_dev_uninit(struct net_device *dev)
L
Linus Torvalds 已提交
312
{
313
	struct ip6_tnl *t = netdev_priv(dev);
314 315
	struct net *net = dev_net(dev);
	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
L
Linus Torvalds 已提交
316

317 318
	if (dev == ip6_fb_tnl_dev) {
		write_lock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
319
		tnls_wc[0] = NULL;
320
		write_unlock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
321
	} else {
322
		ip6_tnl_unlink(ip6n, t);
L
Linus Torvalds 已提交
323 324 325 326 327 328 329 330 331
	}
	ip6_tnl_dst_reset(t);
	dev_put(dev);
}

/**
 * parse_tvl_tnl_enc_lim - handle encapsulation limit option
 *   @skb: received socket buffer
 *
332 333
 * Return:
 *   0 if none was found,
L
Linus Torvalds 已提交
334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389
 *   else index to encapsulation limit
 **/

static __u16
parse_tlv_tnl_enc_lim(struct sk_buff *skb, __u8 * raw)
{
	struct ipv6hdr *ipv6h = (struct ipv6hdr *) raw;
	__u8 nexthdr = ipv6h->nexthdr;
	__u16 off = sizeof (*ipv6h);

	while (ipv6_ext_hdr(nexthdr) && nexthdr != NEXTHDR_NONE) {
		__u16 optlen = 0;
		struct ipv6_opt_hdr *hdr;
		if (raw + off + sizeof (*hdr) > skb->data &&
		    !pskb_may_pull(skb, raw - skb->data + off + sizeof (*hdr)))
			break;

		hdr = (struct ipv6_opt_hdr *) (raw + off);
		if (nexthdr == NEXTHDR_FRAGMENT) {
			struct frag_hdr *frag_hdr = (struct frag_hdr *) hdr;
			if (frag_hdr->frag_off)
				break;
			optlen = 8;
		} else if (nexthdr == NEXTHDR_AUTH) {
			optlen = (hdr->hdrlen + 2) << 2;
		} else {
			optlen = ipv6_optlen(hdr);
		}
		if (nexthdr == NEXTHDR_DEST) {
			__u16 i = off + 2;
			while (1) {
				struct ipv6_tlv_tnl_enc_lim *tel;

				/* No more room for encapsulation limit */
				if (i + sizeof (*tel) > off + optlen)
					break;

				tel = (struct ipv6_tlv_tnl_enc_lim *) &raw[i];
				/* return index of option if found and valid */
				if (tel->type == IPV6_TLV_TNL_ENCAP_LIMIT &&
				    tel->length == 1)
					return i;
				/* else jump to next option */
				if (tel->type)
					i += tel->length + 2;
				else
					i++;
			}
		}
		nexthdr = hdr->nexthdr;
		off += optlen;
	}
	return 0;
}

/**
390
 * ip6_tnl_err - tunnel error handler
L
Linus Torvalds 已提交
391 392
 *
 * Description:
393
 *   ip6_tnl_err() should handle errors in the tunnel according
L
Linus Torvalds 已提交
394 395 396
 *   to the specifications in RFC 2473.
 **/

H
Herbert Xu 已提交
397
static int
398
ip6_tnl_err(struct sk_buff *skb, __u8 ipproto, struct inet6_skb_parm *opt,
A
Al Viro 已提交
399
	    int *type, int *code, int *msg, __u32 *info, int offset)
L
Linus Torvalds 已提交
400 401 402 403 404 405 406 407
{
	struct ipv6hdr *ipv6h = (struct ipv6hdr *) skb->data;
	struct ip6_tnl *t;
	int rel_msg = 0;
	int rel_type = ICMPV6_DEST_UNREACH;
	int rel_code = ICMPV6_ADDR_UNREACH;
	__u32 rel_info = 0;
	__u16 len;
H
Herbert Xu 已提交
408
	int err = -ENOENT;
L
Linus Torvalds 已提交
409

410 411
	/* If the packet doesn't contain the original IPv6 header we are
	   in trouble since we might need the source address for further
L
Linus Torvalds 已提交
412 413
	   processing of the error. */

414
	read_lock(&ip6_tnl_lock);
415
	if ((t = ip6_tnl_lookup(dev_net(skb->dev), &ipv6h->daddr,
416
					&ipv6h->saddr)) == NULL)
L
Linus Torvalds 已提交
417 418
		goto out;

419 420 421
	if (t->parms.proto != ipproto && t->parms.proto != 0)
		goto out;

H
Herbert Xu 已提交
422 423
	err = 0;

424
	switch (*type) {
L
Linus Torvalds 已提交
425 426 427 428 429 430 431 432 433 434 435
		__u32 teli;
		struct ipv6_tlv_tnl_enc_lim *tel;
		__u32 mtu;
	case ICMPV6_DEST_UNREACH:
		if (net_ratelimit())
			printk(KERN_WARNING
			       "%s: Path to destination invalid "
			       "or inactive!\n", t->parms.name);
		rel_msg = 1;
		break;
	case ICMPV6_TIME_EXCEED:
436
		if ((*code) == ICMPV6_EXC_HOPLIMIT) {
L
Linus Torvalds 已提交
437 438 439
			if (net_ratelimit())
				printk(KERN_WARNING
				       "%s: Too small hop limit or "
440
				       "routing loop in tunnel!\n",
L
Linus Torvalds 已提交
441 442 443 444 445
				       t->parms.name);
			rel_msg = 1;
		}
		break;
	case ICMPV6_PARAMPROB:
446
		teli = 0;
447
		if ((*code) == ICMPV6_HDR_FIELD)
448
			teli = parse_tlv_tnl_enc_lim(skb, skb->data);
L
Linus Torvalds 已提交
449

A
Al Viro 已提交
450
		if (teli && teli == *info - 2) {
L
Linus Torvalds 已提交
451 452 453 454 455 456 457 458 459
			tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
			if (tel->encap_limit == 0) {
				if (net_ratelimit())
					printk(KERN_WARNING
					       "%s: Too small encapsulation "
					       "limit or routing loop in "
					       "tunnel!\n", t->parms.name);
				rel_msg = 1;
			}
460 461 462 463
		} else if (net_ratelimit()) {
			printk(KERN_WARNING
			       "%s: Recipient unable to parse tunneled "
			       "packet!\n ", t->parms.name);
L
Linus Torvalds 已提交
464 465 466
		}
		break;
	case ICMPV6_PKT_TOOBIG:
A
Al Viro 已提交
467
		mtu = *info - offset;
L
Linus Torvalds 已提交
468 469 470 471
		if (mtu < IPV6_MIN_MTU)
			mtu = IPV6_MIN_MTU;
		t->dev->mtu = mtu;

A
Al Viro 已提交
472
		if ((len = sizeof (*ipv6h) + ntohs(ipv6h->payload_len)) > mtu) {
L
Linus Torvalds 已提交
473 474 475 476 477 478 479
			rel_type = ICMPV6_PKT_TOOBIG;
			rel_code = 0;
			rel_info = mtu;
			rel_msg = 1;
		}
		break;
	}
480 481 482 483 484 485 486

	*type = rel_type;
	*code = rel_code;
	*info = rel_info;
	*msg = rel_msg;

out:
487
	read_unlock(&ip6_tnl_lock);
488 489 490
	return err;
}

491 492
static int
ip4ip6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
A
Al Viro 已提交
493
	   int type, int code, int offset, __be32 info)
494 495 496 497
{
	int rel_msg = 0;
	int rel_type = type;
	int rel_code = code;
A
Al Viro 已提交
498
	__u32 rel_info = ntohl(info);
499 500 501 502 503 504
	int err;
	struct sk_buff *skb2;
	struct iphdr *eiph;
	struct flowi fl;
	struct rtable *rt;

505 506
	err = ip6_tnl_err(skb, IPPROTO_IPIP, opt, &rel_type, &rel_code,
			  &rel_msg, &rel_info, offset);
507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539
	if (err < 0)
		return err;

	if (rel_msg == 0)
		return 0;

	switch (rel_type) {
	case ICMPV6_DEST_UNREACH:
		if (rel_code != ICMPV6_ADDR_UNREACH)
			return 0;
		rel_type = ICMP_DEST_UNREACH;
		rel_code = ICMP_HOST_UNREACH;
		break;
	case ICMPV6_PKT_TOOBIG:
		if (rel_code != 0)
			return 0;
		rel_type = ICMP_DEST_UNREACH;
		rel_code = ICMP_FRAG_NEEDED;
		break;
	default:
		return 0;
	}

	if (!pskb_may_pull(skb, offset + sizeof(struct iphdr)))
		return 0;

	skb2 = skb_clone(skb, GFP_ATOMIC);
	if (!skb2)
		return 0;

	dst_release(skb2->dst);
	skb2->dst = NULL;
	skb_pull(skb2, offset);
540
	skb_reset_network_header(skb2);
541
	eiph = ip_hdr(skb2);
542 543 544 545 546 547

	/* Try to guess incoming interface */
	memset(&fl, 0, sizeof(fl));
	fl.fl4_dst = eiph->saddr;
	fl.fl4_tos = RT_TOS(eiph->tos);
	fl.proto = IPPROTO_IPIP;
548
	if (ip_route_output_key(&init_net, &rt, &fl))
549 550 551 552 553 554 555 556 557 558 559
		goto out;

	skb2->dev = rt->u.dst.dev;

	/* route "incoming" packet */
	if (rt->rt_flags & RTCF_LOCAL) {
		ip_rt_put(rt);
		rt = NULL;
		fl.fl4_dst = eiph->daddr;
		fl.fl4_src = eiph->saddr;
		fl.fl4_tos = eiph->tos;
560
		if (ip_route_output_key(&init_net, &rt, &fl) ||
561 562 563 564
		    rt->u.dst.dev->type != ARPHRD_TUNNEL) {
			ip_rt_put(rt);
			goto out;
		}
565
		skb2->dst = (struct dst_entry *)rt;
566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581
	} else {
		ip_rt_put(rt);
		if (ip_route_input(skb2, eiph->daddr, eiph->saddr, eiph->tos,
				   skb2->dev) ||
		    skb2->dst->dev->type != ARPHRD_TUNNEL)
			goto out;
	}

	/* change mtu on this route */
	if (rel_type == ICMP_DEST_UNREACH && rel_code == ICMP_FRAG_NEEDED) {
		if (rel_info > dst_mtu(skb2->dst))
			goto out;

		skb2->dst->ops->update_pmtu(skb2->dst, rel_info);
	}

A
Al Viro 已提交
582
	icmp_send(skb2, rel_type, rel_code, htonl(rel_info));
583 584 585 586 587 588

out:
	kfree_skb(skb2);
	return 0;
}

589 590
static int
ip6ip6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
A
Al Viro 已提交
591
	   int type, int code, int offset, __be32 info)
592 593 594 595
{
	int rel_msg = 0;
	int rel_type = type;
	int rel_code = code;
A
Al Viro 已提交
596
	__u32 rel_info = ntohl(info);
597 598
	int err;

599 600
	err = ip6_tnl_err(skb, IPPROTO_IPV6, opt, &rel_type, &rel_code,
			  &rel_msg, &rel_info, offset);
601 602 603 604
	if (err < 0)
		return err;

	if (rel_msg && pskb_may_pull(skb, offset + sizeof(struct ipv6hdr))) {
L
Linus Torvalds 已提交
605 606
		struct rt6_info *rt;
		struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
607

L
Linus Torvalds 已提交
608
		if (!skb2)
609
			return 0;
L
Linus Torvalds 已提交
610 611 612 613

		dst_release(skb2->dst);
		skb2->dst = NULL;
		skb_pull(skb2, offset);
614
		skb_reset_network_header(skb2);
L
Linus Torvalds 已提交
615 616

		/* Try to guess incoming interface */
617
		rt = rt6_lookup(&init_net, &ipv6_hdr(skb2)->saddr, NULL, 0, 0);
L
Linus Torvalds 已提交
618 619 620 621 622 623 624 625 626 627 628

		if (rt && rt->rt6i_dev)
			skb2->dev = rt->rt6i_dev;

		icmpv6_send(skb2, rel_type, rel_code, rel_info, skb2->dev);

		if (rt)
			dst_release(&rt->u.dst);

		kfree_skb(skb2);
	}
629 630

	return 0;
L
Linus Torvalds 已提交
631 632
}

633 634 635 636 637 638 639
static void ip4ip6_dscp_ecn_decapsulate(struct ip6_tnl *t,
					struct ipv6hdr *ipv6h,
					struct sk_buff *skb)
{
	__u8 dsfield = ipv6_get_dsfield(ipv6h) & ~INET_ECN_MASK;

	if (t->parms.flags & IP6_TNL_F_RCV_DSCP_COPY)
640
		ipv4_change_dsfield(ip_hdr(skb), INET_ECN_MASK, dsfield);
641 642

	if (INET_ECN_is_ce(dsfield))
643
		IP_ECN_set_ce(ip_hdr(skb));
644 645
}

646 647 648
static void ip6ip6_dscp_ecn_decapsulate(struct ip6_tnl *t,
					struct ipv6hdr *ipv6h,
					struct sk_buff *skb)
L
Linus Torvalds 已提交
649
{
650
	if (t->parms.flags & IP6_TNL_F_RCV_DSCP_COPY)
651
		ipv6_copy_dscp(ipv6_get_dsfield(ipv6h), ipv6_hdr(skb));
L
Linus Torvalds 已提交
652

653
	if (INET_ECN_is_ce(ipv6_get_dsfield(ipv6h)))
654
		IP6_ECN_set_ce(ipv6_hdr(skb));
L
Linus Torvalds 已提交
655
}
656

657 658 659 660 661 662
static inline int ip6_tnl_rcv_ctl(struct ip6_tnl *t)
{
	struct ip6_tnl_parm *p = &t->parms;
	int ret = 0;

	if (p->flags & IP6_TNL_F_CAP_RCV) {
663
		struct net_device *ldev = NULL;
664 665

		if (p->link)
666
			ldev = dev_get_by_index(&init_net, p->link);
667 668

		if ((ipv6_addr_is_multicast(&p->laddr) ||
669 670
		     likely(ipv6_chk_addr(&init_net, &p->laddr, ldev, 0))) &&
		    likely(!ipv6_chk_addr(&init_net, &p->raddr, NULL, 0)))
671 672 673 674 675 676 677
			ret = 1;

		if (ldev)
			dev_put(ldev);
	}
	return ret;
}
L
Linus Torvalds 已提交
678 679

/**
680
 * ip6_tnl_rcv - decapsulate IPv6 packet and retransmit it locally
L
Linus Torvalds 已提交
681
 *   @skb: received socket buffer
682 683
 *   @protocol: ethernet protocol ID
 *   @dscp_ecn_decapsulate: the function to decapsulate DSCP code and ECN
L
Linus Torvalds 已提交
684 685 686 687
 *
 * Return: 0
 **/

688
static int ip6_tnl_rcv(struct sk_buff *skb, __u16 protocol,
689
		       __u8 ipproto,
690 691 692
		       void (*dscp_ecn_decapsulate)(struct ip6_tnl *t,
						    struct ipv6hdr *ipv6h,
						    struct sk_buff *skb))
L
Linus Torvalds 已提交
693 694
{
	struct ip6_tnl *t;
695
	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
L
Linus Torvalds 已提交
696

697
	read_lock(&ip6_tnl_lock);
L
Linus Torvalds 已提交
698

699
	if ((t = ip6_tnl_lookup(dev_net(skb->dev), &ipv6h->saddr,
700
					&ipv6h->daddr)) != NULL) {
701 702 703 704 705
		if (t->parms.proto != ipproto && t->parms.proto != 0) {
			read_unlock(&ip6_tnl_lock);
			goto discard;
		}

L
Linus Torvalds 已提交
706
		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
707
			read_unlock(&ip6_tnl_lock);
708
			goto discard;
L
Linus Torvalds 已提交
709 710
		}

711
		if (!ip6_tnl_rcv_ctl(t)) {
L
Linus Torvalds 已提交
712
			t->stat.rx_dropped++;
713
			read_unlock(&ip6_tnl_lock);
L
Linus Torvalds 已提交
714 715 716
			goto discard;
		}
		secpath_reset(skb);
717
		skb->mac_header = skb->network_header;
718
		skb_reset_network_header(skb);
719
		skb->protocol = htons(protocol);
L
Linus Torvalds 已提交
720 721 722 723 724
		skb->pkt_type = PACKET_HOST;
		memset(skb->cb, 0, sizeof(struct inet6_skb_parm));
		skb->dev = t->dev;
		dst_release(skb->dst);
		skb->dst = NULL;
725
		nf_reset(skb);
726 727 728

		dscp_ecn_decapsulate(t, ipv6h, skb);

L
Linus Torvalds 已提交
729 730 731
		t->stat.rx_packets++;
		t->stat.rx_bytes += skb->len;
		netif_rx(skb);
732
		read_unlock(&ip6_tnl_lock);
L
Linus Torvalds 已提交
733 734
		return 0;
	}
735
	read_unlock(&ip6_tnl_lock);
L
Linus Torvalds 已提交
736
	return 1;
737 738 739 740

discard:
	kfree_skb(skb);
	return 0;
L
Linus Torvalds 已提交
741 742
}

743 744
static int ip4ip6_rcv(struct sk_buff *skb)
{
745 746
	return ip6_tnl_rcv(skb, ETH_P_IP, IPPROTO_IPIP,
			   ip4ip6_dscp_ecn_decapsulate);
747 748
}

749 750
static int ip6ip6_rcv(struct sk_buff *skb)
{
751 752
	return ip6_tnl_rcv(skb, ETH_P_IPV6, IPPROTO_IPV6,
			   ip6ip6_dscp_ecn_decapsulate);
753 754
}

755 756 757 758
struct ipv6_tel_txoption {
	struct ipv6_txoptions ops;
	__u8 dst_opt[8];
};
L
Linus Torvalds 已提交
759

760 761 762
static void init_tel_txopt(struct ipv6_tel_txoption *opt, __u8 encap_limit)
{
	memset(opt, 0, sizeof(struct ipv6_tel_txoption));
L
Linus Torvalds 已提交
763

764 765 766 767 768
	opt->dst_opt[2] = IPV6_TLV_TNL_ENCAP_LIMIT;
	opt->dst_opt[3] = 1;
	opt->dst_opt[4] = encap_limit;
	opt->dst_opt[5] = IPV6_TLV_PADN;
	opt->dst_opt[6] = 1;
L
Linus Torvalds 已提交
769

770 771
	opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
	opt->ops.opt_nflen = 8;
L
Linus Torvalds 已提交
772 773 774
}

/**
775
 * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
L
Linus Torvalds 已提交
776
 *   @t: the outgoing tunnel device
777
 *   @hdr: IPv6 header from the incoming packet
L
Linus Torvalds 已提交
778 779
 *
 * Description:
780
 *   Avoid trivial tunneling loop by checking that tunnel exit-point
L
Linus Torvalds 已提交
781 782
 *   doesn't match source of incoming packet.
 *
783
 * Return:
L
Linus Torvalds 已提交
784 785 786 787 788
 *   1 if conflict,
 *   0 else
 **/

static inline int
789
ip6_tnl_addr_conflict(struct ip6_tnl *t, struct ipv6hdr *hdr)
L
Linus Torvalds 已提交
790 791 792 793
{
	return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
}

794 795 796 797 798
static inline int ip6_tnl_xmit_ctl(struct ip6_tnl *t)
{
	struct ip6_tnl_parm *p = &t->parms;
	int ret = 0;

799
	if (p->flags & IP6_TNL_F_CAP_XMIT) {
800 801 802
		struct net_device *ldev = NULL;

		if (p->link)
803
			ldev = dev_get_by_index(&init_net, p->link);
804

805
		if (unlikely(!ipv6_chk_addr(&init_net, &p->laddr, ldev, 0)))
806 807 808 809
			printk(KERN_WARNING
			       "%s xmit: Local address not yet configured!\n",
			       p->name);
		else if (!ipv6_addr_is_multicast(&p->raddr) &&
810
			 unlikely(ipv6_chk_addr(&init_net, &p->raddr, NULL, 0)))
811 812 813 814 815 816 817 818 819 820 821
			printk(KERN_WARNING
			       "%s xmit: Routing loop! "
			       "Remote address found on this node!\n",
			       p->name);
		else
			ret = 1;
		if (ldev)
			dev_put(ldev);
	}
	return ret;
}
L
Linus Torvalds 已提交
822
/**
823
 * ip6_tnl_xmit2 - encapsulate packet and send
L
Linus Torvalds 已提交
824
 *   @skb: the outgoing socket buffer
825
 *   @dev: the outgoing tunnel device
826 827 828 829
 *   @dsfield: dscp code for outer header
 *   @fl: flow of tunneled packet
 *   @encap_limit: encapsulation limit
 *   @pmtu: Path MTU is stored if packet is too big
L
Linus Torvalds 已提交
830 831 832 833 834
 *
 * Description:
 *   Build new header and do some sanity checks on the packet before sending
 *   it.
 *
835
 * Return:
836
 *   0 on success
837 838
 *   -1 fail
 *   %-EMSGSIZE message too big. return mtu in this case.
L
Linus Torvalds 已提交
839 840
 **/

841 842 843 844 845 846
static int ip6_tnl_xmit2(struct sk_buff *skb,
			 struct net_device *dev,
			 __u8 dsfield,
			 struct flowi *fl,
			 int encap_limit,
			 __u32 *pmtu)
L
Linus Torvalds 已提交
847
{
848
	struct ip6_tnl *t = netdev_priv(dev);
L
Linus Torvalds 已提交
849
	struct net_device_stats *stats = &t->stat;
850
	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
851
	struct ipv6_tel_txoption opt;
L
Linus Torvalds 已提交
852 853 854
	struct dst_entry *dst;
	struct net_device *tdev;
	int mtu;
855
	unsigned int max_headroom = sizeof(struct ipv6hdr);
L
Linus Torvalds 已提交
856
	u8 proto;
857
	int err = -1;
L
Linus Torvalds 已提交
858 859 860 861
	int pkt_len;

	if ((dst = ip6_tnl_dst_check(t)) != NULL)
		dst_hold(dst);
862
	else {
863
		dst = ip6_route_output(&init_net, NULL, fl);
L
Linus Torvalds 已提交
864

865
		if (dst->error || xfrm_lookup(&dst, fl, NULL, 0) < 0)
866 867
			goto tx_err_link_failure;
	}
L
Linus Torvalds 已提交
868 869 870 871 872 873

	tdev = dst->dev;

	if (tdev == dev) {
		stats->collisions++;
		if (net_ratelimit())
874
			printk(KERN_WARNING
L
Linus Torvalds 已提交
875 876 877 878 879
			       "%s: Local routing loop detected!\n",
			       t->parms.name);
		goto tx_err_dst_release;
	}
	mtu = dst_mtu(dst) - sizeof (*ipv6h);
880
	if (encap_limit >= 0) {
L
Linus Torvalds 已提交
881 882 883 884 885
		max_headroom += 8;
		mtu -= 8;
	}
	if (mtu < IPV6_MIN_MTU)
		mtu = IPV6_MIN_MTU;
886 887
	if (skb->dst)
		skb->dst->ops->update_pmtu(skb->dst, mtu);
L
Linus Torvalds 已提交
888
	if (skb->len > mtu) {
889 890
		*pmtu = mtu;
		err = -EMSGSIZE;
L
Linus Torvalds 已提交
891 892 893 894 895 896 897
		goto tx_err_dst_release;
	}

	/*
	 * Okay, now see if we can stuff it in the buffer as-is.
	 */
	max_headroom += LL_RESERVED_SPACE(tdev);
898

899 900
	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
L
Linus Torvalds 已提交
901
		struct sk_buff *new_skb;
902

L
Linus Torvalds 已提交
903 904 905 906 907 908 909 910 911 912 913
		if (!(new_skb = skb_realloc_headroom(skb, max_headroom)))
			goto tx_err_dst_release;

		if (skb->sk)
			skb_set_owner_w(new_skb, skb->sk);
		kfree_skb(skb);
		skb = new_skb;
	}
	dst_release(skb->dst);
	skb->dst = dst_clone(dst);

914
	skb->transport_header = skb->network_header;
L
Linus Torvalds 已提交
915

916
	proto = fl->proto;
917 918 919 920
	if (encap_limit >= 0) {
		init_tel_txopt(&opt, encap_limit);
		ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
	}
921 922
	skb_push(skb, sizeof(struct ipv6hdr));
	skb_reset_network_header(skb);
923
	ipv6h = ipv6_hdr(skb);
924
	*(__be32*)ipv6h = fl->fl6_flowlabel | htonl(0x60000000);
L
Linus Torvalds 已提交
925 926 927 928
	dsfield = INET_ECN_encapsulate(0, dsfield);
	ipv6_change_dsfield(ipv6h, ~INET_ECN_MASK, dsfield);
	ipv6h->hop_limit = t->parms.hop_limit;
	ipv6h->nexthdr = proto;
929 930
	ipv6_addr_copy(&ipv6h->saddr, &fl->fl6_src);
	ipv6_addr_copy(&ipv6h->daddr, &fl->fl6_dst);
L
Linus Torvalds 已提交
931 932
	nf_reset(skb);
	pkt_len = skb->len;
H
Herbert Xu 已提交
933
	err = ip6_local_out(skb);
L
Linus Torvalds 已提交
934

935
	if (net_xmit_eval(err) == 0) {
L
Linus Torvalds 已提交
936 937 938 939 940 941 942 943 944 945 946 947 948
		stats->tx_bytes += pkt_len;
		stats->tx_packets++;
	} else {
		stats->tx_errors++;
		stats->tx_aborted_errors++;
	}
	ip6_tnl_dst_store(t, dst);
	return 0;
tx_err_link_failure:
	stats->tx_carrier_errors++;
	dst_link_failure(skb);
tx_err_dst_release:
	dst_release(dst);
949 950 951
	return err;
}

952 953 954 955
static inline int
ip4ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
{
	struct ip6_tnl *t = netdev_priv(dev);
956
	struct iphdr  *iph = ip_hdr(skb);
957 958 959 960 961 962
	int encap_limit = -1;
	struct flowi fl;
	__u8 dsfield;
	__u32 mtu;
	int err;

963 964
	if ((t->parms.proto != IPPROTO_IPIP && t->parms.proto != 0) ||
	    !ip6_tnl_xmit_ctl(t))
965 966 967 968 969 970 971 972 973 974 975
		return -1;

	if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
		encap_limit = t->parms.encap_limit;

	memcpy(&fl, &t->fl, sizeof (fl));
	fl.proto = IPPROTO_IPIP;

	dsfield = ipv4_get_dsfield(iph);

	if ((t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS))
A
Al Viro 已提交
976 977
		fl.fl6_flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
					  & IPV6_TCLASS_MASK;
978 979 980 981 982 983 984 985 986 987 988 989 990

	err = ip6_tnl_xmit2(skb, dev, dsfield, &fl, encap_limit, &mtu);
	if (err != 0) {
		/* XXX: send ICMP error even if DF is not set. */
		if (err == -EMSGSIZE)
			icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
				  htonl(mtu));
		return -1;
	}

	return 0;
}

991 992 993 994
static inline int
ip6ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
{
	struct ip6_tnl *t = netdev_priv(dev);
995
	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
996 997 998 999 1000 1001 1002
	int encap_limit = -1;
	__u16 offset;
	struct flowi fl;
	__u8 dsfield;
	__u32 mtu;
	int err;

1003 1004
	if ((t->parms.proto != IPPROTO_IPV6 && t->parms.proto != 0) ||
	    !ip6_tnl_xmit_ctl(t) || ip6_tnl_addr_conflict(t, ipv6h))
1005 1006
		return -1;

1007 1008
	offset = parse_tlv_tnl_enc_lim(skb, skb_network_header(skb));
	if (offset > 0) {
1009
		struct ipv6_tlv_tnl_enc_lim *tel;
1010
		tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051
		if (tel->encap_limit == 0) {
			icmpv6_send(skb, ICMPV6_PARAMPROB,
				    ICMPV6_HDR_FIELD, offset + 2, skb->dev);
			return -1;
		}
		encap_limit = tel->encap_limit - 1;
	} else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
		encap_limit = t->parms.encap_limit;

	memcpy(&fl, &t->fl, sizeof (fl));
	fl.proto = IPPROTO_IPV6;

	dsfield = ipv6_get_dsfield(ipv6h);
	if ((t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS))
		fl.fl6_flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
	if ((t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL))
		fl.fl6_flowlabel |= (*(__be32 *) ipv6h & IPV6_FLOWLABEL_MASK);

	err = ip6_tnl_xmit2(skb, dev, dsfield, &fl, encap_limit, &mtu);
	if (err != 0) {
		if (err == -EMSGSIZE)
			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, dev);
		return -1;
	}

	return 0;
}

static int
ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
{
	struct ip6_tnl *t = netdev_priv(dev);
	struct net_device_stats *stats = &t->stat;
	int ret;

	if (t->recursion++) {
		t->stat.collisions++;
		goto tx_err;
	}

	switch (skb->protocol) {
1052 1053 1054
	case __constant_htons(ETH_P_IP):
		ret = ip4ip6_tnl_xmit(skb, dev);
		break;
1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067
	case __constant_htons(ETH_P_IPV6):
		ret = ip6ip6_tnl_xmit(skb, dev);
		break;
	default:
		goto tx_err;
	}

	if (ret < 0)
		goto tx_err;

	t->recursion--;
	return 0;

L
Linus Torvalds 已提交
1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078
tx_err:
	stats->tx_errors++;
	stats->tx_dropped++;
	kfree_skb(skb);
	t->recursion--;
	return 0;
}

static void ip6_tnl_set_cap(struct ip6_tnl *t)
{
	struct ip6_tnl_parm *p = &t->parms;
1079 1080
	int ltype = ipv6_addr_type(&p->laddr);
	int rtype = ipv6_addr_type(&p->raddr);
L
Linus Torvalds 已提交
1081 1082 1083

	p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV);

1084 1085 1086
	if (ltype & (IPV6_ADDR_UNICAST|IPV6_ADDR_MULTICAST) &&
	    rtype & (IPV6_ADDR_UNICAST|IPV6_ADDR_MULTICAST) &&
	    !((ltype|rtype) & IPV6_ADDR_LOOPBACK) &&
1087
	    (!((ltype|rtype) & IPV6_ADDR_LINKLOCAL) || p->link)) {
1088 1089 1090 1091
		if (ltype&IPV6_ADDR_UNICAST)
			p->flags |= IP6_TNL_F_CAP_XMIT;
		if (rtype&IPV6_ADDR_UNICAST)
			p->flags |= IP6_TNL_F_CAP_RCV;
L
Linus Torvalds 已提交
1092 1093 1094
	}
}

1095
static void ip6_tnl_link_config(struct ip6_tnl *t)
L
Linus Torvalds 已提交
1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124
{
	struct net_device *dev = t->dev;
	struct ip6_tnl_parm *p = &t->parms;
	struct flowi *fl = &t->fl;

	memcpy(&dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
	memcpy(&dev->broadcast, &p->raddr, sizeof(struct in6_addr));

	/* Set up flowi template */
	ipv6_addr_copy(&fl->fl6_src, &p->laddr);
	ipv6_addr_copy(&fl->fl6_dst, &p->raddr);
	fl->oif = p->link;
	fl->fl6_flowlabel = 0;

	if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
		fl->fl6_flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
	if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
		fl->fl6_flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;

	ip6_tnl_set_cap(t);

	if (p->flags&IP6_TNL_F_CAP_XMIT && p->flags&IP6_TNL_F_CAP_RCV)
		dev->flags |= IFF_POINTOPOINT;
	else
		dev->flags &= ~IFF_POINTOPOINT;

	dev->iflink = p->link;

	if (p->flags & IP6_TNL_F_CAP_XMIT) {
1125 1126 1127
		int strict = (ipv6_addr_type(&p->raddr) &
			      (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));

1128
		struct rt6_info *rt = rt6_lookup(&init_net, &p->raddr, &p->laddr,
1129
						 p->link, strict);
L
Linus Torvalds 已提交
1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147

		if (rt == NULL)
			return;

		if (rt->rt6i_dev) {
			dev->hard_header_len = rt->rt6i_dev->hard_header_len +
				sizeof (struct ipv6hdr);

			dev->mtu = rt->rt6i_dev->mtu - sizeof (struct ipv6hdr);

			if (dev->mtu < IPV6_MIN_MTU)
				dev->mtu = IPV6_MIN_MTU;
		}
		dst_release(&rt->u.dst);
	}
}

/**
1148
 * ip6_tnl_change - update the tunnel parameters
L
Linus Torvalds 已提交
1149 1150 1151 1152 1153
 *   @t: tunnel to be changed
 *   @p: tunnel configuration parameters
 *   @active: != 0 if tunnel is ready for use
 *
 * Description:
1154
 *   ip6_tnl_change() updates the tunnel parameters
L
Linus Torvalds 已提交
1155 1156 1157
 **/

static int
1158
ip6_tnl_change(struct ip6_tnl *t, struct ip6_tnl_parm *p)
L
Linus Torvalds 已提交
1159 1160 1161 1162 1163 1164 1165
{
	ipv6_addr_copy(&t->parms.laddr, &p->laddr);
	ipv6_addr_copy(&t->parms.raddr, &p->raddr);
	t->parms.flags = p->flags;
	t->parms.hop_limit = p->hop_limit;
	t->parms.encap_limit = p->encap_limit;
	t->parms.flowinfo = p->flowinfo;
1166
	t->parms.link = p->link;
1167
	t->parms.proto = p->proto;
1168
	ip6_tnl_dst_reset(t);
1169
	ip6_tnl_link_config(t);
L
Linus Torvalds 已提交
1170 1171 1172 1173
	return 0;
}

/**
1174
 * ip6_tnl_ioctl - configure ipv6 tunnels from userspace
L
Linus Torvalds 已提交
1175 1176 1177 1178 1179
 *   @dev: virtual device associated with tunnel
 *   @ifr: parameters passed from userspace
 *   @cmd: command to be performed
 *
 * Description:
1180
 *   ip6_tnl_ioctl() is used for managing IPv6 tunnels
1181
 *   from userspace.
L
Linus Torvalds 已提交
1182 1183 1184 1185 1186 1187 1188
 *
 *   The possible commands are the following:
 *     %SIOCGETTUNNEL: get tunnel parameters for device
 *     %SIOCADDTUNNEL: add tunnel matching given tunnel parameters
 *     %SIOCCHGTUNNEL: change tunnel parameters to those given
 *     %SIOCDELTUNNEL: delete tunnel
 *
1189
 *   The fallback device "ip6tnl0", created during module
L
Linus Torvalds 已提交
1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201
 *   initialization, can be used for creating other tunnel devices.
 *
 * Return:
 *   0 on success,
 *   %-EFAULT if unable to copy data to or from userspace,
 *   %-EPERM if current process hasn't %CAP_NET_ADMIN set
 *   %-EINVAL if passed tunnel parameters are invalid,
 *   %-EEXIST if changing a tunnel's parameters would cause a conflict
 *   %-ENODEV if attempting to change or delete a nonexisting device
 **/

static int
1202
ip6_tnl_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
L
Linus Torvalds 已提交
1203 1204 1205 1206
{
	int err = 0;
	struct ip6_tnl_parm p;
	struct ip6_tnl *t = NULL;
1207 1208
	struct net *net = dev_net(dev);
	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
L
Linus Torvalds 已提交
1209 1210 1211

	switch (cmd) {
	case SIOCGETTUNNEL:
1212
		if (dev == ip6_fb_tnl_dev) {
1213
			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof (p))) {
L
Linus Torvalds 已提交
1214 1215 1216
				err = -EFAULT;
				break;
			}
1217
			t = ip6_tnl_locate(net, &p, 0);
1218 1219
		}
		if (t == NULL)
1220
			t = netdev_priv(dev);
L
Linus Torvalds 已提交
1221 1222 1223 1224 1225 1226 1227 1228 1229 1230
		memcpy(&p, &t->parms, sizeof (p));
		if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof (p))) {
			err = -EFAULT;
		}
		break;
	case SIOCADDTUNNEL:
	case SIOCCHGTUNNEL:
		err = -EPERM;
		if (!capable(CAP_NET_ADMIN))
			break;
1231 1232
		err = -EFAULT;
		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof (p)))
L
Linus Torvalds 已提交
1233
			break;
1234
		err = -EINVAL;
1235 1236
		if (p.proto != IPPROTO_IPV6 && p.proto != IPPROTO_IPIP &&
		    p.proto != 0)
L
Linus Torvalds 已提交
1237
			break;
1238
		t = ip6_tnl_locate(net, &p, cmd == SIOCADDTUNNEL);
1239
		if (dev != ip6_fb_tnl_dev && cmd == SIOCCHGTUNNEL) {
1240 1241 1242 1243 1244 1245 1246 1247
			if (t != NULL) {
				if (t->dev != dev) {
					err = -EEXIST;
					break;
				}
			} else
				t = netdev_priv(dev);

1248
			ip6_tnl_unlink(ip6n, t);
1249
			err = ip6_tnl_change(t, &p);
1250
			ip6_tnl_link(ip6n, t);
L
Linus Torvalds 已提交
1251 1252
			netdev_state_change(dev);
		}
1253
		if (t) {
L
Linus Torvalds 已提交
1254
			err = 0;
1255 1256 1257 1258 1259
			if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof (p)))
				err = -EFAULT;

		} else
			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
L
Linus Torvalds 已提交
1260 1261 1262 1263 1264 1265
		break;
	case SIOCDELTUNNEL:
		err = -EPERM;
		if (!capable(CAP_NET_ADMIN))
			break;

1266
		if (dev == ip6_fb_tnl_dev) {
1267 1268
			err = -EFAULT;
			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof (p)))
L
Linus Torvalds 已提交
1269
				break;
1270
			err = -ENOENT;
1271
			if ((t = ip6_tnl_locate(net, &p, 0)) == NULL)
L
Linus Torvalds 已提交
1272
				break;
1273
			err = -EPERM;
1274
			if (t->dev == ip6_fb_tnl_dev)
L
Linus Torvalds 已提交
1275
				break;
1276
			dev = t->dev;
L
Linus Torvalds 已提交
1277
		}
1278 1279
		err = 0;
		unregister_netdevice(dev);
L
Linus Torvalds 已提交
1280 1281 1282 1283 1284 1285 1286 1287
		break;
	default:
		err = -EINVAL;
	}
	return err;
}

/**
1288
 * ip6_tnl_get_stats - return the stats for tunnel device
L
Linus Torvalds 已提交
1289 1290 1291 1292 1293 1294
 *   @dev: virtual device associated with tunnel
 *
 * Return: stats for device
 **/

static struct net_device_stats *
1295
ip6_tnl_get_stats(struct net_device *dev)
L
Linus Torvalds 已提交
1296
{
1297
	return &(((struct ip6_tnl *)netdev_priv(dev))->stat);
L
Linus Torvalds 已提交
1298 1299 1300
}

/**
1301
 * ip6_tnl_change_mtu - change mtu manually for tunnel device
L
Linus Torvalds 已提交
1302 1303 1304 1305 1306 1307 1308 1309 1310
 *   @dev: virtual device associated with tunnel
 *   @new_mtu: the new mtu
 *
 * Return:
 *   0 on success,
 *   %-EINVAL if mtu too small
 **/

static int
1311
ip6_tnl_change_mtu(struct net_device *dev, int new_mtu)
L
Linus Torvalds 已提交
1312 1313 1314 1315 1316 1317 1318 1319 1320
{
	if (new_mtu < IPV6_MIN_MTU) {
		return -EINVAL;
	}
	dev->mtu = new_mtu;
	return 0;
}

/**
1321
 * ip6_tnl_dev_setup - setup virtual tunnel device
L
Linus Torvalds 已提交
1322 1323 1324 1325 1326 1327
 *   @dev: virtual device associated with tunnel
 *
 * Description:
 *   Initialize function pointers and device parameters
 **/

1328
static void ip6_tnl_dev_setup(struct net_device *dev)
L
Linus Torvalds 已提交
1329
{
1330
	dev->uninit = ip6_tnl_dev_uninit;
L
Linus Torvalds 已提交
1331
	dev->destructor = free_netdev;
1332
	dev->hard_start_xmit = ip6_tnl_xmit;
1333 1334 1335
	dev->get_stats = ip6_tnl_get_stats;
	dev->do_ioctl = ip6_tnl_ioctl;
	dev->change_mtu = ip6_tnl_change_mtu;
L
Linus Torvalds 已提交
1336 1337 1338 1339 1340 1341 1342 1343 1344 1345

	dev->type = ARPHRD_TUNNEL6;
	dev->hard_header_len = LL_MAX_HEADER + sizeof (struct ipv6hdr);
	dev->mtu = ETH_DATA_LEN - sizeof (struct ipv6hdr);
	dev->flags |= IFF_NOARP;
	dev->addr_len = sizeof(struct in6_addr);
}


/**
1346
 * ip6_tnl_dev_init_gen - general initializer for all tunnel devices
L
Linus Torvalds 已提交
1347 1348 1349 1350
 *   @dev: virtual device associated with tunnel
 **/

static inline void
1351
ip6_tnl_dev_init_gen(struct net_device *dev)
L
Linus Torvalds 已提交
1352
{
1353
	struct ip6_tnl *t = netdev_priv(dev);
L
Linus Torvalds 已提交
1354 1355 1356 1357 1358
	t->dev = dev;
	strcpy(t->parms.name, dev->name);
}

/**
1359
 * ip6_tnl_dev_init - initializer for all non fallback tunnel devices
L
Linus Torvalds 已提交
1360 1361 1362 1363
 *   @dev: virtual device associated with tunnel
 **/

static int
1364
ip6_tnl_dev_init(struct net_device *dev)
L
Linus Torvalds 已提交
1365
{
1366
	struct ip6_tnl *t = netdev_priv(dev);
1367 1368
	ip6_tnl_dev_init_gen(dev);
	ip6_tnl_link_config(t);
L
Linus Torvalds 已提交
1369 1370 1371 1372
	return 0;
}

/**
1373
 * ip6_fb_tnl_dev_init - initializer for fallback tunnel device
L
Linus Torvalds 已提交
1374 1375 1376 1377 1378
 *   @dev: fallback device
 *
 * Return: 0
 **/

1379
static int
1380
ip6_fb_tnl_dev_init(struct net_device *dev)
L
Linus Torvalds 已提交
1381
{
1382
	struct ip6_tnl *t = netdev_priv(dev);
1383
	ip6_tnl_dev_init_gen(dev);
1384
	t->parms.proto = IPPROTO_IPV6;
L
Linus Torvalds 已提交
1385 1386 1387 1388 1389
	dev_hold(dev);
	tnls_wc[0] = t;
	return 0;
}

1390 1391 1392 1393 1394 1395
static struct xfrm6_tunnel ip4ip6_handler = {
	.handler	= ip4ip6_rcv,
	.err_handler	= ip4ip6_err,
	.priority	=	1,
};

L
Linus Torvalds 已提交
1396
static struct xfrm6_tunnel ip6ip6_handler = {
1397 1398
	.handler	= ip6ip6_rcv,
	.err_handler	= ip6ip6_err,
H
Herbert Xu 已提交
1399
	.priority	=	1,
L
Linus Torvalds 已提交
1400 1401
};

1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436
static int ip6_tnl_init_net(struct net *net)
{
	int err;
	struct ip6_tnl_net *ip6n;

	err = -ENOMEM;
	ip6n = kmalloc(sizeof(struct ip6_tnl_net), GFP_KERNEL);
	if (ip6n == NULL)
		goto err_alloc;

	err = net_assign_generic(net, ip6_tnl_net_id, ip6n);
	if (err < 0)
		goto err_assign;

	return 0;

err_assign:
	kfree(ip6n);
err_alloc:
	return err;
}

static void ip6_tnl_exit_net(struct net *net)
{
	struct ip6_tnl_net *ip6n;

	ip6n = net_generic(net, ip6_tnl_net_id);
	kfree(ip6n);
}

static struct pernet_operations ip6_tnl_net_ops = {
	.init = ip6_tnl_init_net,
	.exit = ip6_tnl_exit_net,
};

L
Linus Torvalds 已提交
1437 1438 1439 1440 1441 1442 1443 1444 1445 1446
/**
 * ip6_tunnel_init - register protocol and reserve needed resources
 *
 * Return: 0 on success
 **/

static int __init ip6_tunnel_init(void)
{
	int  err;

1447
	if (xfrm6_tunnel_register(&ip4ip6_handler, AF_INET)) {
1448
		printk(KERN_ERR "ip6_tunnel init: can't register ip4ip6\n");
1449 1450 1451 1452
		err = -EAGAIN;
		goto out;
	}

1453
	if (xfrm6_tunnel_register(&ip6ip6_handler, AF_INET6)) {
1454
		printk(KERN_ERR "ip6_tunnel init: can't register ip6ip6\n");
1455 1456
		err = -EAGAIN;
		goto unreg_ip4ip6;
L
Linus Torvalds 已提交
1457
	}
1458 1459
	ip6_fb_tnl_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6tnl0",
				      ip6_tnl_dev_setup);
L
Linus Torvalds 已提交
1460

1461
	if (!ip6_fb_tnl_dev) {
L
Linus Torvalds 已提交
1462 1463 1464
		err = -ENOMEM;
		goto fail;
	}
1465
	ip6_fb_tnl_dev->init = ip6_fb_tnl_dev_init;
L
Linus Torvalds 已提交
1466

1467 1468
	if ((err = register_netdev(ip6_fb_tnl_dev))) {
		free_netdev(ip6_fb_tnl_dev);
L
Linus Torvalds 已提交
1469 1470
		goto fail;
	}
1471 1472 1473 1474

	err = register_pernet_gen_device(&ip6_tnl_net_id, &ip6_tnl_net_ops);
	if (err < 0)
		goto err_pernet;
L
Linus Torvalds 已提交
1475
	return 0;
1476 1477
err_pernet:
	unregister_netdevice(ip6_fb_tnl_dev);
L
Linus Torvalds 已提交
1478
fail:
1479
	xfrm6_tunnel_deregister(&ip6ip6_handler, AF_INET6);
1480 1481 1482
unreg_ip4ip6:
	xfrm6_tunnel_deregister(&ip4ip6_handler, AF_INET);
out:
L
Linus Torvalds 已提交
1483 1484 1485
	return err;
}

1486
static void __exit ip6_tnl_destroy_tunnels(void)
1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499
{
	int h;
	struct ip6_tnl *t;

	for (h = 0; h < HASH_SIZE; h++) {
		while ((t = tnls_r_l[h]) != NULL)
			unregister_netdevice(t->dev);
	}

	t = tnls_wc[0];
	unregister_netdevice(t->dev);
}

L
Linus Torvalds 已提交
1500 1501 1502 1503 1504 1505
/**
 * ip6_tunnel_cleanup - free resources and unregister protocol
 **/

static void __exit ip6_tunnel_cleanup(void)
{
1506
	if (xfrm6_tunnel_deregister(&ip4ip6_handler, AF_INET))
1507
		printk(KERN_INFO "ip6_tunnel close: can't deregister ip4ip6\n");
1508

1509
	if (xfrm6_tunnel_deregister(&ip6ip6_handler, AF_INET6))
1510
		printk(KERN_INFO "ip6_tunnel close: can't deregister ip6ip6\n");
L
Linus Torvalds 已提交
1511

1512
	rtnl_lock();
1513
	ip6_tnl_destroy_tunnels();
1514
	rtnl_unlock();
1515 1516

	unregister_pernet_gen_device(ip6_tnl_net_id, &ip6_tnl_net_ops);
L
Linus Torvalds 已提交
1517 1518 1519 1520
}

module_init(ip6_tunnel_init);
module_exit(ip6_tunnel_cleanup);