ip6_tunnel.c 35.0 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
 *
 *      Based on:
10
 *      linux/net/ipv6/sit.c and linux/net/ipv4/ipip.c
L
Linus Torvalds 已提交
11 12 13 14 15 16 17 18 19 20 21
 *
 *      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>
22
#include <linux/capability.h>
L
Linus Torvalds 已提交
23 24 25
#include <linux/errno.h>
#include <linux/types.h>
#include <linux/sockios.h>
26
#include <linux/icmp.h>
L
Linus Torvalds 已提交
27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43
#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>

44
#include <net/icmp.h>
L
Linus Torvalds 已提交
45 46 47 48 49 50 51 52
#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>
53 54
#include <net/net_namespace.h>
#include <net/netns/generic.h>
L
Linus Torvalds 已提交
55 56

MODULE_AUTHOR("Ville Nuorvala");
57
MODULE_DESCRIPTION("IPv6 tunneling device");
L
Linus Torvalds 已提交
58 59 60 61 62
MODULE_LICENSE("GPL");

#define IPV6_TLV_TEL_DST_SIZE 8

#ifdef IP6_TNL_DEBUG
63
#define IP6_TNL_TRACE(x...) printk(KERN_DEBUG "%s:" x "\n", __func__)
L
Linus Torvalds 已提交
64 65 66 67 68
#else
#define IP6_TNL_TRACE(x...) do {;} while(0)
#endif

#define IPV6_TCLASS_MASK (IPV6_FLOWINFO_MASK & ~IPV6_FLOWLABEL_MASK)
69
#define IPV6_TCLASS_SHIFT 20
L
Linus Torvalds 已提交
70 71 72

#define HASH_SIZE  32

A
Al Viro 已提交
73
#define HASH(addr) ((__force u32)((addr)->s6_addr32[0] ^ (addr)->s6_addr32[1] ^ \
74 75
		     (addr)->s6_addr32[2] ^ (addr)->s6_addr32[3]) & \
		    (HASH_SIZE - 1))
L
Linus Torvalds 已提交
76

77 78 79
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
80

81 82
static int ip6_tnl_net_id;
struct ip6_tnl_net {
83 84
	/* the IPv6 tunnel fallback device */
	struct net_device *fb_tnl_dev;
85 86 87 88
	/* lists for storing tunnels in use */
	struct ip6_tnl *tnls_r_l[HASH_SIZE];
	struct ip6_tnl *tnls_wc[1];
	struct ip6_tnl **tnls[2];
89 90
};

L
Linus Torvalds 已提交
91
/* lock for the tunnel lists */
92
static DEFINE_RWLOCK(ip6_tnl_lock);
L
Linus Torvalds 已提交
93 94 95 96 97

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

98
	if (dst && dst->obsolete &&
L
Linus Torvalds 已提交
99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122
	    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;
}

/**
123
 * ip6_tnl_lookup - fetch tunnel matching the end-point addresses
124 125
 *   @remote: the address of the tunnel exit-point
 *   @local: the address of the tunnel entry-point
L
Linus Torvalds 已提交
126
 *
127
 * Return:
L
Linus Torvalds 已提交
128
 *   tunnel matching given end-points if found,
129
 *   else fallback tunnel if its device is up,
L
Linus Torvalds 已提交
130 131 132 133
 *   else %NULL
 **/

static struct ip6_tnl *
134
ip6_tnl_lookup(struct net *net, struct in6_addr *remote, struct in6_addr *local)
L
Linus Torvalds 已提交
135 136 137 138
{
	unsigned h0 = HASH(remote);
	unsigned h1 = HASH(local);
	struct ip6_tnl *t;
139
	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
L
Linus Torvalds 已提交
140

141
	for (t = ip6n->tnls_r_l[h0 ^ h1]; t; t = t->next) {
L
Linus Torvalds 已提交
142 143 144 145 146
		if (ipv6_addr_equal(local, &t->parms.laddr) &&
		    ipv6_addr_equal(remote, &t->parms.raddr) &&
		    (t->dev->flags & IFF_UP))
			return t;
	}
147
	if ((t = ip6n->tnls_wc[0]) != NULL && (t->dev->flags & IFF_UP))
L
Linus Torvalds 已提交
148 149 150 151 152 153
		return t;

	return NULL;
}

/**
154
 * ip6_tnl_bucket - get head of list matching given tunnel parameters
155
 *   @p: parameters containing tunnel end-points
L
Linus Torvalds 已提交
156 157
 *
 * Description:
158
 *   ip6_tnl_bucket() returns the head of the list matching the
L
Linus Torvalds 已提交
159 160
 *   &struct in6_addr entries laddr and raddr in @p.
 *
161
 * Return: head of IPv6 tunnel list
L
Linus Torvalds 已提交
162 163 164
 **/

static struct ip6_tnl **
165
ip6_tnl_bucket(struct ip6_tnl_net *ip6n, struct ip6_tnl_parm *p)
L
Linus Torvalds 已提交
166 167 168 169 170 171 172 173 174 175
{
	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);
	}
176
	return &ip6n->tnls[prio][h];
L
Linus Torvalds 已提交
177 178 179
}

/**
180
 * ip6_tnl_link - add tunnel to hash table
L
Linus Torvalds 已提交
181 182 183 184
 *   @t: tunnel to be added
 **/

static void
185
ip6_tnl_link(struct ip6_tnl_net *ip6n, struct ip6_tnl *t)
L
Linus Torvalds 已提交
186
{
187
	struct ip6_tnl **tp = ip6_tnl_bucket(ip6n, &t->parms);
L
Linus Torvalds 已提交
188 189

	t->next = *tp;
190
	write_lock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
191
	*tp = t;
192
	write_unlock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
193 194 195
}

/**
196
 * ip6_tnl_unlink - remove tunnel from hash table
L
Linus Torvalds 已提交
197 198 199 200
 *   @t: tunnel to be removed
 **/

static void
201
ip6_tnl_unlink(struct ip6_tnl_net *ip6n, struct ip6_tnl *t)
L
Linus Torvalds 已提交
202 203 204
{
	struct ip6_tnl **tp;

205
	for (tp = ip6_tnl_bucket(ip6n, &t->parms); *tp; tp = &(*tp)->next) {
L
Linus Torvalds 已提交
206
		if (t == *tp) {
207
			write_lock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
208
			*tp = t->next;
209
			write_unlock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
210 211 212 213 214 215 216 217 218 219 220 221
			break;
		}
	}
}

/**
 * ip6_tnl_create() - create a new tunnel
 *   @p: tunnel parameters
 *   @pt: pointer to new tunnel
 *
 * Description:
 *   Create tunnel matching given parameters.
222 223
 *
 * Return:
224
 *   created tunnel or NULL
L
Linus Torvalds 已提交
225 226
 **/

227
static struct ip6_tnl *ip6_tnl_create(struct net *net, struct ip6_tnl_parm *p)
L
Linus Torvalds 已提交
228 229 230 231 232
{
	struct net_device *dev;
	struct ip6_tnl *t;
	char name[IFNAMSIZ];
	int err;
233
	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
L
Linus Torvalds 已提交
234

235
	if (p->name[0])
L
Linus Torvalds 已提交
236
		strlcpy(name, p->name, IFNAMSIZ);
237 238 239
	else
		sprintf(name, "ip6tnl%%d");

240
	dev = alloc_netdev(sizeof (*t), name, ip6_tnl_dev_setup);
L
Linus Torvalds 已提交
241
	if (dev == NULL)
242
		goto failed;
L
Linus Torvalds 已提交
243

244 245
	dev_net_set(dev, net);

246 247 248 249 250
	if (strchr(name, '%')) {
		if (dev_alloc_name(dev, name) < 0)
			goto failed_free;
	}

251
	t = netdev_priv(dev);
252
	dev->init = ip6_tnl_dev_init;
L
Linus Torvalds 已提交
253 254
	t->parms = *p;

255 256 257
	if ((err = register_netdevice(dev)) < 0)
		goto failed_free;

L
Linus Torvalds 已提交
258
	dev_hold(dev);
259
	ip6_tnl_link(ip6n, t);
260
	return t;
261 262 263

failed_free:
	free_netdev(dev);
264 265
failed:
	return NULL;
L
Linus Torvalds 已提交
266 267 268
}

/**
269
 * ip6_tnl_locate - find or create tunnel matching given parameters
270
 *   @p: tunnel parameters
L
Linus Torvalds 已提交
271 272 273
 *   @create: != 0 if allowed to create new tunnel if no match found
 *
 * Description:
274
 *   ip6_tnl_locate() first tries to locate an existing tunnel
L
Linus Torvalds 已提交
275 276 277 278
 *   based on @parms. If this is unsuccessful, but @create is set a new
 *   tunnel device is created and registered for use.
 *
 * Return:
279
 *   matching tunnel or NULL
L
Linus Torvalds 已提交
280 281
 **/

282 283
static struct ip6_tnl *ip6_tnl_locate(struct net *net,
		struct ip6_tnl_parm *p, int create)
L
Linus Torvalds 已提交
284 285 286 287
{
	struct in6_addr *remote = &p->raddr;
	struct in6_addr *local = &p->laddr;
	struct ip6_tnl *t;
288
	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
L
Linus Torvalds 已提交
289

290
	for (t = *ip6_tnl_bucket(ip6n, p); t; t = t->next) {
L
Linus Torvalds 已提交
291
		if (ipv6_addr_equal(local, &t->parms.laddr) &&
292 293
		    ipv6_addr_equal(remote, &t->parms.raddr))
			return t;
L
Linus Torvalds 已提交
294 295
	}
	if (!create)
296
		return NULL;
297
	return ip6_tnl_create(net, p);
L
Linus Torvalds 已提交
298 299 300
}

/**
301
 * ip6_tnl_dev_uninit - tunnel device uninitializer
L
Linus Torvalds 已提交
302
 *   @dev: the device to be destroyed
303
 *
L
Linus Torvalds 已提交
304
 * Description:
305
 *   ip6_tnl_dev_uninit() removes tunnel from its list
L
Linus Torvalds 已提交
306 307 308
 **/

static void
309
ip6_tnl_dev_uninit(struct net_device *dev)
L
Linus Torvalds 已提交
310
{
311
	struct ip6_tnl *t = netdev_priv(dev);
312 313
	struct net *net = dev_net(dev);
	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
L
Linus Torvalds 已提交
314

315
	if (dev == ip6n->fb_tnl_dev) {
316
		write_lock_bh(&ip6_tnl_lock);
317
		ip6n->tnls_wc[0] = NULL;
318
		write_unlock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
319
	} else {
320
		ip6_tnl_unlink(ip6n, t);
L
Linus Torvalds 已提交
321 322 323 324 325 326 327 328 329
	}
	ip6_tnl_dst_reset(t);
	dev_put(dev);
}

/**
 * parse_tvl_tnl_enc_lim - handle encapsulation limit option
 *   @skb: received socket buffer
 *
330 331
 * Return:
 *   0 if none was found,
L
Linus Torvalds 已提交
332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387
 *   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;
}

/**
388
 * ip6_tnl_err - tunnel error handler
L
Linus Torvalds 已提交
389 390
 *
 * Description:
391
 *   ip6_tnl_err() should handle errors in the tunnel according
L
Linus Torvalds 已提交
392 393 394
 *   to the specifications in RFC 2473.
 **/

H
Herbert Xu 已提交
395
static int
396
ip6_tnl_err(struct sk_buff *skb, __u8 ipproto, struct inet6_skb_parm *opt,
A
Al Viro 已提交
397
	    int *type, int *code, int *msg, __u32 *info, int offset)
L
Linus Torvalds 已提交
398 399 400 401 402 403 404 405
{
	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
406
	int err = -ENOENT;
L
Linus Torvalds 已提交
407

408 409
	/* 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
410 411
	   processing of the error. */

412
	read_lock(&ip6_tnl_lock);
413
	if ((t = ip6_tnl_lookup(dev_net(skb->dev), &ipv6h->daddr,
414
					&ipv6h->saddr)) == NULL)
L
Linus Torvalds 已提交
415 416
		goto out;

417 418 419
	if (t->parms.proto != ipproto && t->parms.proto != 0)
		goto out;

H
Herbert Xu 已提交
420 421
	err = 0;

422
	switch (*type) {
L
Linus Torvalds 已提交
423 424 425 426 427 428 429 430 431 432 433
		__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:
434
		if ((*code) == ICMPV6_EXC_HOPLIMIT) {
L
Linus Torvalds 已提交
435 436 437
			if (net_ratelimit())
				printk(KERN_WARNING
				       "%s: Too small hop limit or "
438
				       "routing loop in tunnel!\n",
L
Linus Torvalds 已提交
439 440 441 442 443
				       t->parms.name);
			rel_msg = 1;
		}
		break;
	case ICMPV6_PARAMPROB:
444
		teli = 0;
445
		if ((*code) == ICMPV6_HDR_FIELD)
446
			teli = parse_tlv_tnl_enc_lim(skb, skb->data);
L
Linus Torvalds 已提交
447

A
Al Viro 已提交
448
		if (teli && teli == *info - 2) {
L
Linus Torvalds 已提交
449 450 451 452 453 454 455 456 457
			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;
			}
458 459 460 461
		} else if (net_ratelimit()) {
			printk(KERN_WARNING
			       "%s: Recipient unable to parse tunneled "
			       "packet!\n ", t->parms.name);
L
Linus Torvalds 已提交
462 463 464
		}
		break;
	case ICMPV6_PKT_TOOBIG:
A
Al Viro 已提交
465
		mtu = *info - offset;
L
Linus Torvalds 已提交
466 467 468 469
		if (mtu < IPV6_MIN_MTU)
			mtu = IPV6_MIN_MTU;
		t->dev->mtu = mtu;

A
Al Viro 已提交
470
		if ((len = sizeof (*ipv6h) + ntohs(ipv6h->payload_len)) > mtu) {
L
Linus Torvalds 已提交
471 472 473 474 475 476 477
			rel_type = ICMPV6_PKT_TOOBIG;
			rel_code = 0;
			rel_info = mtu;
			rel_msg = 1;
		}
		break;
	}
478 479 480 481 482 483 484

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

out:
485
	read_unlock(&ip6_tnl_lock);
486 487 488
	return err;
}

489 490
static int
ip4ip6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
A
Al Viro 已提交
491
	   int type, int code, int offset, __be32 info)
492 493 494 495
{
	int rel_msg = 0;
	int rel_type = type;
	int rel_code = code;
A
Al Viro 已提交
496
	__u32 rel_info = ntohl(info);
497 498 499 500 501 502
	int err;
	struct sk_buff *skb2;
	struct iphdr *eiph;
	struct flowi fl;
	struct rtable *rt;

503 504
	err = ip6_tnl_err(skb, IPPROTO_IPIP, opt, &rel_type, &rel_code,
			  &rel_msg, &rel_info, offset);
505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537
	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);
538
	skb_reset_network_header(skb2);
539
	eiph = ip_hdr(skb2);
540 541 542 543 544 545

	/* 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;
546
	if (ip_route_output_key(dev_net(skb->dev), &rt, &fl))
547 548 549 550 551 552 553 554 555 556 557
		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;
558
		if (ip_route_output_key(dev_net(skb->dev), &rt, &fl) ||
559 560 561 562
		    rt->u.dst.dev->type != ARPHRD_TUNNEL) {
			ip_rt_put(rt);
			goto out;
		}
563
		skb2->dst = (struct dst_entry *)rt;
564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579
	} 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
580
	icmp_send(skb2, rel_type, rel_code, htonl(rel_info));
581 582 583 584 585 586

out:
	kfree_skb(skb2);
	return 0;
}

587 588
static int
ip6ip6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
A
Al Viro 已提交
589
	   int type, int code, int offset, __be32 info)
590 591 592 593
{
	int rel_msg = 0;
	int rel_type = type;
	int rel_code = code;
A
Al Viro 已提交
594
	__u32 rel_info = ntohl(info);
595 596
	int err;

597 598
	err = ip6_tnl_err(skb, IPPROTO_IPV6, opt, &rel_type, &rel_code,
			  &rel_msg, &rel_info, offset);
599 600 601 602
	if (err < 0)
		return err;

	if (rel_msg && pskb_may_pull(skb, offset + sizeof(struct ipv6hdr))) {
L
Linus Torvalds 已提交
603 604
		struct rt6_info *rt;
		struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
605

L
Linus Torvalds 已提交
606
		if (!skb2)
607
			return 0;
L
Linus Torvalds 已提交
608 609 610 611

		dst_release(skb2->dst);
		skb2->dst = NULL;
		skb_pull(skb2, offset);
612
		skb_reset_network_header(skb2);
L
Linus Torvalds 已提交
613 614

		/* Try to guess incoming interface */
615 616
		rt = rt6_lookup(dev_net(skb->dev), &ipv6_hdr(skb2)->saddr,
				NULL, 0, 0);
L
Linus Torvalds 已提交
617 618 619 620 621 622 623 624 625 626 627

		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);
	}
628 629

	return 0;
L
Linus Torvalds 已提交
630 631
}

632 633 634 635 636 637 638
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)
639
		ipv4_change_dsfield(ip_hdr(skb), INET_ECN_MASK, dsfield);
640 641

	if (INET_ECN_is_ce(dsfield))
642
		IP_ECN_set_ce(ip_hdr(skb));
643 644
}

645 646 647
static void ip6ip6_dscp_ecn_decapsulate(struct ip6_tnl *t,
					struct ipv6hdr *ipv6h,
					struct sk_buff *skb)
L
Linus Torvalds 已提交
648
{
649
	if (t->parms.flags & IP6_TNL_F_RCV_DSCP_COPY)
650
		ipv6_copy_dscp(ipv6_get_dsfield(ipv6h), ipv6_hdr(skb));
L
Linus Torvalds 已提交
651

652
	if (INET_ECN_is_ce(ipv6_get_dsfield(ipv6h)))
653
		IP6_ECN_set_ce(ipv6_hdr(skb));
L
Linus Torvalds 已提交
654
}
655

656 657 658 659
static inline int ip6_tnl_rcv_ctl(struct ip6_tnl *t)
{
	struct ip6_tnl_parm *p = &t->parms;
	int ret = 0;
660
	struct net *net = dev_net(t->dev);
661 662

	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(net, p->link);
667 668

		if ((ipv6_addr_is_multicast(&p->laddr) ||
669 670
		     likely(ipv6_chk_addr(net, &p->laddr, ldev, 0))) &&
		    likely(!ipv6_chk_addr(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)) {
712
			t->dev->stats.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);

729 730
		t->dev->stats.rx_packets++;
		t->dev->stats.rx_bytes += skb->len;
L
Linus Torvalds 已提交
731
		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
static inline int ip6_tnl_xmit_ctl(struct ip6_tnl *t)
{
	struct ip6_tnl_parm *p = &t->parms;
	int ret = 0;
798
	struct net *net = dev_net(t->dev);
799

800
	if (p->flags & IP6_TNL_F_CAP_XMIT) {
801 802 803
		struct net_device *ldev = NULL;

		if (p->link)
804
			ldev = dev_get_by_index(net, p->link);
805

806
		if (unlikely(!ipv6_chk_addr(net, &p->laddr, ldev, 0)))
807 808 809 810
			printk(KERN_WARNING
			       "%s xmit: Local address not yet configured!\n",
			       p->name);
		else if (!ipv6_addr_is_multicast(&p->raddr) &&
811
			 unlikely(ipv6_chk_addr(net, &p->raddr, NULL, 0)))
812 813 814 815 816 817 818 819 820 821 822
			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
823
/**
824
 * ip6_tnl_xmit2 - encapsulate packet and send
L
Linus Torvalds 已提交
825
 *   @skb: the outgoing socket buffer
826
 *   @dev: the outgoing tunnel device
827 828 829 830
 *   @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
831 832 833 834 835
 *
 * Description:
 *   Build new header and do some sanity checks on the packet before sending
 *   it.
 *
836
 * Return:
837
 *   0 on success
838 839
 *   -1 fail
 *   %-EMSGSIZE message too big. return mtu in this case.
L
Linus Torvalds 已提交
840 841
 **/

842 843 844 845 846 847
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
848
{
849
	struct ip6_tnl *t = netdev_priv(dev);
850
	struct net_device_stats *stats = &t->dev->stats;
851
	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
852
	struct ipv6_tel_txoption opt;
L
Linus Torvalds 已提交
853 854 855
	struct dst_entry *dst;
	struct net_device *tdev;
	int mtu;
856
	unsigned int max_headroom = sizeof(struct ipv6hdr);
L
Linus Torvalds 已提交
857
	u8 proto;
858
	int err = -1;
L
Linus Torvalds 已提交
859 860 861 862
	int pkt_len;

	if ((dst = ip6_tnl_dst_check(t)) != NULL)
		dst_hold(dst);
863
	else {
864
		dst = ip6_route_output(dev_net(dev), NULL, fl);
L
Linus Torvalds 已提交
865

866
		if (dst->error || xfrm_lookup(&dst, fl, NULL, 0) < 0)
867 868
			goto tx_err_link_failure;
	}
L
Linus Torvalds 已提交
869 870 871 872 873 874

	tdev = dst->dev;

	if (tdev == dev) {
		stats->collisions++;
		if (net_ratelimit())
875
			printk(KERN_WARNING
L
Linus Torvalds 已提交
876 877 878 879 880
			       "%s: Local routing loop detected!\n",
			       t->parms.name);
		goto tx_err_dst_release;
	}
	mtu = dst_mtu(dst) - sizeof (*ipv6h);
881
	if (encap_limit >= 0) {
L
Linus Torvalds 已提交
882 883 884 885 886
		max_headroom += 8;
		mtu -= 8;
	}
	if (mtu < IPV6_MIN_MTU)
		mtu = IPV6_MIN_MTU;
887 888
	if (skb->dst)
		skb->dst->ops->update_pmtu(skb->dst, mtu);
L
Linus Torvalds 已提交
889
	if (skb->len > mtu) {
890 891
		*pmtu = mtu;
		err = -EMSGSIZE;
L
Linus Torvalds 已提交
892 893 894 895 896 897 898
		goto tx_err_dst_release;
	}

	/*
	 * Okay, now see if we can stuff it in the buffer as-is.
	 */
	max_headroom += LL_RESERVED_SPACE(tdev);
899

900 901
	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
L
Linus Torvalds 已提交
902
		struct sk_buff *new_skb;
903

L
Linus Torvalds 已提交
904 905 906 907 908 909 910 911 912 913 914
		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);

915
	skb->transport_header = skb->network_header;
L
Linus Torvalds 已提交
916

917
	proto = fl->proto;
918 919 920 921
	if (encap_limit >= 0) {
		init_tel_txopt(&opt, encap_limit);
		ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
	}
922 923
	skb_push(skb, sizeof(struct ipv6hdr));
	skb_reset_network_header(skb);
924
	ipv6h = ipv6_hdr(skb);
925
	*(__be32*)ipv6h = fl->fl6_flowlabel | htonl(0x60000000);
L
Linus Torvalds 已提交
926 927 928 929
	dsfield = INET_ECN_encapsulate(0, dsfield);
	ipv6_change_dsfield(ipv6h, ~INET_ECN_MASK, dsfield);
	ipv6h->hop_limit = t->parms.hop_limit;
	ipv6h->nexthdr = proto;
930 931
	ipv6_addr_copy(&ipv6h->saddr, &fl->fl6_src);
	ipv6_addr_copy(&ipv6h->daddr, &fl->fl6_dst);
L
Linus Torvalds 已提交
932 933
	nf_reset(skb);
	pkt_len = skb->len;
H
Herbert Xu 已提交
934
	err = ip6_local_out(skb);
L
Linus Torvalds 已提交
935

936
	if (net_xmit_eval(err) == 0) {
L
Linus Torvalds 已提交
937 938 939 940 941 942 943 944 945 946 947 948 949
		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);
950 951 952
	return err;
}

953 954 955 956
static inline int
ip4ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
{
	struct ip6_tnl *t = netdev_priv(dev);
957
	struct iphdr  *iph = ip_hdr(skb);
958 959 960 961 962 963
	int encap_limit = -1;
	struct flowi fl;
	__u8 dsfield;
	__u32 mtu;
	int err;

964 965
	if ((t->parms.proto != IPPROTO_IPIP && t->parms.proto != 0) ||
	    !ip6_tnl_xmit_ctl(t))
966 967 968 969 970 971 972 973 974 975 976
		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
977 978
		fl.fl6_flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
					  & IPV6_TCLASS_MASK;
979 980 981 982 983 984 985 986 987 988 989 990 991

	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;
}

992 993 994 995
static inline int
ip6ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
{
	struct ip6_tnl *t = netdev_priv(dev);
996
	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
997 998 999 1000 1001 1002 1003
	int encap_limit = -1;
	__u16 offset;
	struct flowi fl;
	__u8 dsfield;
	__u32 mtu;
	int err;

1004 1005
	if ((t->parms.proto != IPPROTO_IPV6 && t->parms.proto != 0) ||
	    !ip6_tnl_xmit_ctl(t) || ip6_tnl_addr_conflict(t, ipv6h))
1006 1007
		return -1;

1008 1009
	offset = parse_tlv_tnl_enc_lim(skb, skb_network_header(skb));
	if (offset > 0) {
1010
		struct ipv6_tlv_tnl_enc_lim *tel;
1011
		tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
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
		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);
1044
	struct net_device_stats *stats = &t->dev->stats;
1045 1046 1047
	int ret;

	if (t->recursion++) {
1048
		stats->collisions++;
1049 1050 1051 1052
		goto tx_err;
	}

	switch (skb->protocol) {
1053 1054 1055
	case __constant_htons(ETH_P_IP):
		ret = ip4ip6_tnl_xmit(skb, dev);
		break;
1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068
	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
1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079
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;
1080 1081
	int ltype = ipv6_addr_type(&p->laddr);
	int rtype = ipv6_addr_type(&p->raddr);
L
Linus Torvalds 已提交
1082 1083 1084

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

1085 1086 1087
	if (ltype & (IPV6_ADDR_UNICAST|IPV6_ADDR_MULTICAST) &&
	    rtype & (IPV6_ADDR_UNICAST|IPV6_ADDR_MULTICAST) &&
	    !((ltype|rtype) & IPV6_ADDR_LOOPBACK) &&
1088
	    (!((ltype|rtype) & IPV6_ADDR_LINKLOCAL) || p->link)) {
1089 1090 1091 1092
		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
1093 1094 1095
	}
}

1096
static void ip6_tnl_link_config(struct ip6_tnl *t)
L
Linus Torvalds 已提交
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 1125
{
	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) {
1126 1127 1128
		int strict = (ipv6_addr_type(&p->raddr) &
			      (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));

1129 1130
		struct rt6_info *rt = rt6_lookup(dev_net(dev),
						 &p->raddr, &p->laddr,
1131
						 p->link, strict);
L
Linus Torvalds 已提交
1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149

		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);
	}
}

/**
1150
 * ip6_tnl_change - update the tunnel parameters
L
Linus Torvalds 已提交
1151 1152 1153 1154 1155
 *   @t: tunnel to be changed
 *   @p: tunnel configuration parameters
 *   @active: != 0 if tunnel is ready for use
 *
 * Description:
1156
 *   ip6_tnl_change() updates the tunnel parameters
L
Linus Torvalds 已提交
1157 1158 1159
 **/

static int
1160
ip6_tnl_change(struct ip6_tnl *t, struct ip6_tnl_parm *p)
L
Linus Torvalds 已提交
1161 1162 1163 1164 1165 1166 1167
{
	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;
1168
	t->parms.link = p->link;
1169
	t->parms.proto = p->proto;
1170
	ip6_tnl_dst_reset(t);
1171
	ip6_tnl_link_config(t);
L
Linus Torvalds 已提交
1172 1173 1174 1175
	return 0;
}

/**
1176
 * ip6_tnl_ioctl - configure ipv6 tunnels from userspace
L
Linus Torvalds 已提交
1177 1178 1179 1180 1181
 *   @dev: virtual device associated with tunnel
 *   @ifr: parameters passed from userspace
 *   @cmd: command to be performed
 *
 * Description:
1182
 *   ip6_tnl_ioctl() is used for managing IPv6 tunnels
1183
 *   from userspace.
L
Linus Torvalds 已提交
1184 1185 1186 1187 1188 1189 1190
 *
 *   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
 *
1191
 *   The fallback device "ip6tnl0", created during module
L
Linus Torvalds 已提交
1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203
 *   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
1204
ip6_tnl_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
L
Linus Torvalds 已提交
1205 1206 1207 1208
{
	int err = 0;
	struct ip6_tnl_parm p;
	struct ip6_tnl *t = NULL;
1209 1210
	struct net *net = dev_net(dev);
	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
L
Linus Torvalds 已提交
1211 1212 1213

	switch (cmd) {
	case SIOCGETTUNNEL:
1214
		if (dev == ip6n->fb_tnl_dev) {
1215
			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof (p))) {
L
Linus Torvalds 已提交
1216 1217 1218
				err = -EFAULT;
				break;
			}
1219
			t = ip6_tnl_locate(net, &p, 0);
1220 1221
		}
		if (t == NULL)
1222
			t = netdev_priv(dev);
L
Linus Torvalds 已提交
1223 1224 1225 1226 1227 1228 1229 1230 1231 1232
		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;
1233 1234
		err = -EFAULT;
		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof (p)))
L
Linus Torvalds 已提交
1235
			break;
1236
		err = -EINVAL;
1237 1238
		if (p.proto != IPPROTO_IPV6 && p.proto != IPPROTO_IPIP &&
		    p.proto != 0)
L
Linus Torvalds 已提交
1239
			break;
1240
		t = ip6_tnl_locate(net, &p, cmd == SIOCADDTUNNEL);
1241
		if (dev != ip6n->fb_tnl_dev && cmd == SIOCCHGTUNNEL) {
1242 1243 1244 1245 1246 1247 1248 1249
			if (t != NULL) {
				if (t->dev != dev) {
					err = -EEXIST;
					break;
				}
			} else
				t = netdev_priv(dev);

1250
			ip6_tnl_unlink(ip6n, t);
1251
			err = ip6_tnl_change(t, &p);
1252
			ip6_tnl_link(ip6n, t);
L
Linus Torvalds 已提交
1253 1254
			netdev_state_change(dev);
		}
1255
		if (t) {
L
Linus Torvalds 已提交
1256
			err = 0;
1257 1258 1259 1260 1261
			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
1262 1263 1264 1265 1266 1267
		break;
	case SIOCDELTUNNEL:
		err = -EPERM;
		if (!capable(CAP_NET_ADMIN))
			break;

1268
		if (dev == ip6n->fb_tnl_dev) {
1269 1270
			err = -EFAULT;
			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof (p)))
L
Linus Torvalds 已提交
1271
				break;
1272
			err = -ENOENT;
1273
			if ((t = ip6_tnl_locate(net, &p, 0)) == NULL)
L
Linus Torvalds 已提交
1274
				break;
1275
			err = -EPERM;
1276
			if (t->dev == ip6n->fb_tnl_dev)
L
Linus Torvalds 已提交
1277
				break;
1278
			dev = t->dev;
L
Linus Torvalds 已提交
1279
		}
1280 1281
		err = 0;
		unregister_netdevice(dev);
L
Linus Torvalds 已提交
1282 1283 1284 1285 1286 1287 1288 1289
		break;
	default:
		err = -EINVAL;
	}
	return err;
}

/**
1290
 * ip6_tnl_change_mtu - change mtu manually for tunnel device
L
Linus Torvalds 已提交
1291 1292 1293 1294 1295 1296 1297 1298 1299
 *   @dev: virtual device associated with tunnel
 *   @new_mtu: the new mtu
 *
 * Return:
 *   0 on success,
 *   %-EINVAL if mtu too small
 **/

static int
1300
ip6_tnl_change_mtu(struct net_device *dev, int new_mtu)
L
Linus Torvalds 已提交
1301 1302 1303 1304 1305 1306 1307 1308 1309
{
	if (new_mtu < IPV6_MIN_MTU) {
		return -EINVAL;
	}
	dev->mtu = new_mtu;
	return 0;
}

/**
1310
 * ip6_tnl_dev_setup - setup virtual tunnel device
L
Linus Torvalds 已提交
1311 1312 1313 1314 1315 1316
 *   @dev: virtual device associated with tunnel
 *
 * Description:
 *   Initialize function pointers and device parameters
 **/

1317
static void ip6_tnl_dev_setup(struct net_device *dev)
L
Linus Torvalds 已提交
1318
{
1319
	dev->uninit = ip6_tnl_dev_uninit;
L
Linus Torvalds 已提交
1320
	dev->destructor = free_netdev;
1321
	dev->hard_start_xmit = ip6_tnl_xmit;
1322 1323
	dev->do_ioctl = ip6_tnl_ioctl;
	dev->change_mtu = ip6_tnl_change_mtu;
L
Linus Torvalds 已提交
1324 1325 1326 1327 1328 1329

	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);
1330
	dev->features |= NETIF_F_NETNS_LOCAL;
L
Linus Torvalds 已提交
1331 1332 1333 1334
}


/**
1335
 * ip6_tnl_dev_init_gen - general initializer for all tunnel devices
L
Linus Torvalds 已提交
1336 1337 1338 1339
 *   @dev: virtual device associated with tunnel
 **/

static inline void
1340
ip6_tnl_dev_init_gen(struct net_device *dev)
L
Linus Torvalds 已提交
1341
{
1342
	struct ip6_tnl *t = netdev_priv(dev);
L
Linus Torvalds 已提交
1343 1344 1345 1346 1347
	t->dev = dev;
	strcpy(t->parms.name, dev->name);
}

/**
1348
 * ip6_tnl_dev_init - initializer for all non fallback tunnel devices
L
Linus Torvalds 已提交
1349 1350 1351 1352
 *   @dev: virtual device associated with tunnel
 **/

static int
1353
ip6_tnl_dev_init(struct net_device *dev)
L
Linus Torvalds 已提交
1354
{
1355
	struct ip6_tnl *t = netdev_priv(dev);
1356 1357
	ip6_tnl_dev_init_gen(dev);
	ip6_tnl_link_config(t);
L
Linus Torvalds 已提交
1358 1359 1360 1361
	return 0;
}

/**
1362
 * ip6_fb_tnl_dev_init - initializer for fallback tunnel device
L
Linus Torvalds 已提交
1363 1364 1365 1366 1367
 *   @dev: fallback device
 *
 * Return: 0
 **/

1368
static int
1369
ip6_fb_tnl_dev_init(struct net_device *dev)
L
Linus Torvalds 已提交
1370
{
1371
	struct ip6_tnl *t = netdev_priv(dev);
1372 1373 1374
	struct net *net = dev_net(dev);
	struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);

1375
	ip6_tnl_dev_init_gen(dev);
1376
	t->parms.proto = IPPROTO_IPV6;
L
Linus Torvalds 已提交
1377
	dev_hold(dev);
1378
	ip6n->tnls_wc[0] = t;
L
Linus Torvalds 已提交
1379 1380 1381
	return 0;
}

1382 1383 1384 1385 1386 1387
static struct xfrm6_tunnel ip4ip6_handler = {
	.handler	= ip4ip6_rcv,
	.err_handler	= ip4ip6_err,
	.priority	=	1,
};

L
Linus Torvalds 已提交
1388
static struct xfrm6_tunnel ip6ip6_handler = {
1389 1390
	.handler	= ip6ip6_rcv,
	.err_handler	= ip6ip6_err,
H
Herbert Xu 已提交
1391
	.priority	=	1,
L
Linus Torvalds 已提交
1392 1393
};

1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407
static void ip6_tnl_destroy_tunnels(struct ip6_tnl_net *ip6n)
{
	int h;
	struct ip6_tnl *t;

	for (h = 0; h < HASH_SIZE; h++) {
		while ((t = ip6n->tnls_r_l[h]) != NULL)
			unregister_netdevice(t->dev);
	}

	t = ip6n->tnls_wc[0];
	unregister_netdevice(t->dev);
}

1408 1409 1410 1411 1412 1413
static int ip6_tnl_init_net(struct net *net)
{
	int err;
	struct ip6_tnl_net *ip6n;

	err = -ENOMEM;
1414
	ip6n = kzalloc(sizeof(struct ip6_tnl_net), GFP_KERNEL);
1415 1416 1417 1418 1419 1420 1421
	if (ip6n == NULL)
		goto err_alloc;

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

1422 1423 1424
	ip6n->tnls[0] = ip6n->tnls_wc;
	ip6n->tnls[1] = ip6n->tnls_r_l;

1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437
	err = -ENOMEM;
	ip6n->fb_tnl_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6tnl0",
				      ip6_tnl_dev_setup);

	if (!ip6n->fb_tnl_dev)
		goto err_alloc_dev;

	ip6n->fb_tnl_dev->init = ip6_fb_tnl_dev_init;
	dev_net_set(ip6n->fb_tnl_dev, net);

	err = register_netdev(ip6n->fb_tnl_dev);
	if (err < 0)
		goto err_register;
1438 1439
	return 0;

1440 1441 1442 1443
err_register:
	free_netdev(ip6n->fb_tnl_dev);
err_alloc_dev:
	/* nothing */
1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454
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);
1455 1456 1457
	rtnl_lock();
	ip6_tnl_destroy_tunnels(ip6n);
	rtnl_unlock();
1458 1459 1460 1461 1462 1463 1464 1465
	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
1466 1467 1468 1469 1470 1471 1472 1473 1474 1475
/**
 * ip6_tunnel_init - register protocol and reserve needed resources
 *
 * Return: 0 on success
 **/

static int __init ip6_tunnel_init(void)
{
	int  err;

1476
	if (xfrm6_tunnel_register(&ip4ip6_handler, AF_INET)) {
1477
		printk(KERN_ERR "ip6_tunnel init: can't register ip4ip6\n");
1478 1479 1480 1481
		err = -EAGAIN;
		goto out;
	}

1482
	if (xfrm6_tunnel_register(&ip6ip6_handler, AF_INET6)) {
1483
		printk(KERN_ERR "ip6_tunnel init: can't register ip6ip6\n");
1484 1485
		err = -EAGAIN;
		goto unreg_ip4ip6;
L
Linus Torvalds 已提交
1486
	}
1487 1488 1489 1490

	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
1491
	return 0;
1492
err_pernet:
1493
	xfrm6_tunnel_deregister(&ip6ip6_handler, AF_INET6);
1494 1495 1496
unreg_ip4ip6:
	xfrm6_tunnel_deregister(&ip4ip6_handler, AF_INET);
out:
L
Linus Torvalds 已提交
1497 1498 1499 1500 1501 1502 1503 1504 1505
	return err;
}

/**
 * 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
	unregister_pernet_gen_device(ip6_tnl_net_id, &ip6_tnl_net_ops);
L
Linus Torvalds 已提交
1513 1514 1515 1516
}

module_init(ip6_tunnel_init);
module_exit(ip6_tunnel_cleanup);