ip6_tunnel.c 33.4 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 55 56
#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>

MODULE_AUTHOR("Ville Nuorvala");
57
MODULE_DESCRIPTION("IPv6 tunneling device");
L
Linus Torvalds 已提交
58 59 60 61 62 63 64 65 66 67 68
MODULE_LICENSE("GPL");

#define IPV6_TLV_TEL_DST_SIZE 8

#ifdef IP6_TNL_DEBUG
#define IP6_TNL_TRACE(x...) printk(KERN_DEBUG "%s:" x "\n", __FUNCTION__)
#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

/* the IPv6 tunnel fallback device */
82
static struct net_device *ip6_fb_tnl_dev;
L
Linus Torvalds 已提交
83 84 85 86 87 88 89 90


/* 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 */
91
static DEFINE_RWLOCK(ip6_tnl_lock);
L
Linus Torvalds 已提交
92 93 94 95 96

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

97
	if (dst && dst->obsolete &&
L
Linus Torvalds 已提交
98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121
	    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;
}

/**
122
 * ip6_tnl_lookup - fetch tunnel matching the end-point addresses
123 124
 *   @remote: the address of the tunnel exit-point
 *   @local: the address of the tunnel entry-point
L
Linus Torvalds 已提交
125
 *
126
 * Return:
L
Linus Torvalds 已提交
127
 *   tunnel matching given end-points if found,
128
 *   else fallback tunnel if its device is up,
L
Linus Torvalds 已提交
129 130 131 132
 *   else %NULL
 **/

static struct ip6_tnl *
133
ip6_tnl_lookup(struct in6_addr *remote, struct in6_addr *local)
L
Linus Torvalds 已提交
134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151
{
	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;
}

/**
152
 * ip6_tnl_bucket - get head of list matching given tunnel parameters
153
 *   @p: parameters containing tunnel end-points
L
Linus Torvalds 已提交
154 155
 *
 * Description:
156
 *   ip6_tnl_bucket() returns the head of the list matching the
L
Linus Torvalds 已提交
157 158
 *   &struct in6_addr entries laddr and raddr in @p.
 *
159
 * Return: head of IPv6 tunnel list
L
Linus Torvalds 已提交
160 161 162
 **/

static struct ip6_tnl **
163
ip6_tnl_bucket(struct ip6_tnl_parm *p)
L
Linus Torvalds 已提交
164 165 166 167 168 169 170 171 172 173 174 175 176 177
{
	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];
}

/**
178
 * ip6_tnl_link - add tunnel to hash table
L
Linus Torvalds 已提交
179 180 181 182
 *   @t: tunnel to be added
 **/

static void
183
ip6_tnl_link(struct ip6_tnl *t)
L
Linus Torvalds 已提交
184
{
185
	struct ip6_tnl **tp = ip6_tnl_bucket(&t->parms);
L
Linus Torvalds 已提交
186 187

	t->next = *tp;
188
	write_lock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
189
	*tp = t;
190
	write_unlock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
191 192 193
}

/**
194
 * ip6_tnl_unlink - remove tunnel from hash table
L
Linus Torvalds 已提交
195 196 197 198
 *   @t: tunnel to be removed
 **/

static void
199
ip6_tnl_unlink(struct ip6_tnl *t)
L
Linus Torvalds 已提交
200 201 202
{
	struct ip6_tnl **tp;

203
	for (tp = ip6_tnl_bucket(&t->parms); *tp; tp = &(*tp)->next) {
L
Linus Torvalds 已提交
204
		if (t == *tp) {
205
			write_lock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
206
			*tp = t->next;
207
			write_unlock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
208 209 210 211 212 213 214 215 216 217 218 219
			break;
		}
	}
}

/**
 * ip6_tnl_create() - create a new tunnel
 *   @p: tunnel parameters
 *   @pt: pointer to new tunnel
 *
 * Description:
 *   Create tunnel matching given parameters.
220 221
 *
 * Return:
222
 *   created tunnel or NULL
L
Linus Torvalds 已提交
223 224
 **/

225
static struct ip6_tnl *ip6_tnl_create(struct ip6_tnl_parm *p)
L
Linus Torvalds 已提交
226 227 228 229 230 231 232 233 234 235 236 237
{
	struct net_device *dev;
	struct ip6_tnl *t;
	char name[IFNAMSIZ];
	int err;

	if (p->name[0]) {
		strlcpy(name, p->name, IFNAMSIZ);
	} else {
		int i;
		for (i = 1; i < IP6_TNL_MAX; i++) {
			sprintf(name, "ip6tnl%d", i);
238
			if (__dev_get_by_name(&init_net, name) == NULL)
L
Linus Torvalds 已提交
239 240
				break;
		}
241
		if (i == IP6_TNL_MAX)
242
			goto failed;
L
Linus Torvalds 已提交
243
	}
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
	t = netdev_priv(dev);
249
	dev->init = ip6_tnl_dev_init;
L
Linus Torvalds 已提交
250 251 252 253
	t->parms = *p;

	if ((err = register_netdevice(dev)) < 0) {
		free_netdev(dev);
254
		goto failed;
L
Linus Torvalds 已提交
255 256
	}
	dev_hold(dev);
257
	ip6_tnl_link(t);
258 259 260
	return t;
failed:
	return NULL;
L
Linus Torvalds 已提交
261 262 263
}

/**
264
 * ip6_tnl_locate - find or create tunnel matching given parameters
265
 *   @p: tunnel parameters
L
Linus Torvalds 已提交
266 267 268
 *   @create: != 0 if allowed to create new tunnel if no match found
 *
 * Description:
269
 *   ip6_tnl_locate() first tries to locate an existing tunnel
L
Linus Torvalds 已提交
270 271 272 273
 *   based on @parms. If this is unsuccessful, but @create is set a new
 *   tunnel device is created and registered for use.
 *
 * Return:
274
 *   matching tunnel or NULL
L
Linus Torvalds 已提交
275 276
 **/

277
static struct ip6_tnl *ip6_tnl_locate(struct ip6_tnl_parm *p, int create)
L
Linus Torvalds 已提交
278 279 280 281 282
{
	struct in6_addr *remote = &p->raddr;
	struct in6_addr *local = &p->laddr;
	struct ip6_tnl *t;

283
	for (t = *ip6_tnl_bucket(p); t; t = t->next) {
L
Linus Torvalds 已提交
284
		if (ipv6_addr_equal(local, &t->parms.laddr) &&
285 286
		    ipv6_addr_equal(remote, &t->parms.raddr))
			return t;
L
Linus Torvalds 已提交
287 288
	}
	if (!create)
289 290
		return NULL;
	return ip6_tnl_create(p);
L
Linus Torvalds 已提交
291 292 293
}

/**
294
 * ip6_tnl_dev_uninit - tunnel device uninitializer
L
Linus Torvalds 已提交
295
 *   @dev: the device to be destroyed
296
 *
L
Linus Torvalds 已提交
297
 * Description:
298
 *   ip6_tnl_dev_uninit() removes tunnel from its list
L
Linus Torvalds 已提交
299 300 301
 **/

static void
302
ip6_tnl_dev_uninit(struct net_device *dev)
L
Linus Torvalds 已提交
303
{
304
	struct ip6_tnl *t = netdev_priv(dev);
L
Linus Torvalds 已提交
305

306 307
	if (dev == ip6_fb_tnl_dev) {
		write_lock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
308
		tnls_wc[0] = NULL;
309
		write_unlock_bh(&ip6_tnl_lock);
L
Linus Torvalds 已提交
310
	} else {
311
		ip6_tnl_unlink(t);
L
Linus Torvalds 已提交
312 313 314 315 316 317 318 319 320
	}
	ip6_tnl_dst_reset(t);
	dev_put(dev);
}

/**
 * parse_tvl_tnl_enc_lim - handle encapsulation limit option
 *   @skb: received socket buffer
 *
321 322
 * Return:
 *   0 if none was found,
L
Linus Torvalds 已提交
323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378
 *   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;
}

/**
379
 * ip6_tnl_err - tunnel error handler
L
Linus Torvalds 已提交
380 381
 *
 * Description:
382
 *   ip6_tnl_err() should handle errors in the tunnel according
L
Linus Torvalds 已提交
383 384 385
 *   to the specifications in RFC 2473.
 **/

H
Herbert Xu 已提交
386
static int
387
ip6_tnl_err(struct sk_buff *skb, __u8 ipproto, struct inet6_skb_parm *opt,
A
Al Viro 已提交
388
	    int *type, int *code, int *msg, __u32 *info, int offset)
L
Linus Torvalds 已提交
389 390 391 392 393 394 395 396
{
	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
397
	int err = -ENOENT;
L
Linus Torvalds 已提交
398

399 400
	/* 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
401 402
	   processing of the error. */

403 404
	read_lock(&ip6_tnl_lock);
	if ((t = ip6_tnl_lookup(&ipv6h->daddr, &ipv6h->saddr)) == NULL)
L
Linus Torvalds 已提交
405 406
		goto out;

407 408 409
	if (t->parms.proto != ipproto && t->parms.proto != 0)
		goto out;

H
Herbert Xu 已提交
410 411
	err = 0;

412
	switch (*type) {
L
Linus Torvalds 已提交
413 414 415 416 417 418 419 420 421 422 423
		__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:
424
		if ((*code) == ICMPV6_EXC_HOPLIMIT) {
L
Linus Torvalds 已提交
425 426 427
			if (net_ratelimit())
				printk(KERN_WARNING
				       "%s: Too small hop limit or "
428
				       "routing loop in tunnel!\n",
L
Linus Torvalds 已提交
429 430 431 432 433
				       t->parms.name);
			rel_msg = 1;
		}
		break;
	case ICMPV6_PARAMPROB:
434
		teli = 0;
435
		if ((*code) == ICMPV6_HDR_FIELD)
436
			teli = parse_tlv_tnl_enc_lim(skb, skb->data);
L
Linus Torvalds 已提交
437

A
Al Viro 已提交
438
		if (teli && teli == *info - 2) {
L
Linus Torvalds 已提交
439 440 441 442 443 444 445 446 447
			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;
			}
448 449 450 451
		} else if (net_ratelimit()) {
			printk(KERN_WARNING
			       "%s: Recipient unable to parse tunneled "
			       "packet!\n ", t->parms.name);
L
Linus Torvalds 已提交
452 453 454
		}
		break;
	case ICMPV6_PKT_TOOBIG:
A
Al Viro 已提交
455
		mtu = *info - offset;
L
Linus Torvalds 已提交
456 457 458 459
		if (mtu < IPV6_MIN_MTU)
			mtu = IPV6_MIN_MTU;
		t->dev->mtu = mtu;

A
Al Viro 已提交
460
		if ((len = sizeof (*ipv6h) + ntohs(ipv6h->payload_len)) > mtu) {
L
Linus Torvalds 已提交
461 462 463 464 465 466 467
			rel_type = ICMPV6_PKT_TOOBIG;
			rel_code = 0;
			rel_info = mtu;
			rel_msg = 1;
		}
		break;
	}
468 469 470 471 472 473 474

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

out:
475
	read_unlock(&ip6_tnl_lock);
476 477 478
	return err;
}

479 480
static int
ip4ip6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
A
Al Viro 已提交
481
	   int type, int code, int offset, __be32 info)
482 483 484 485
{
	int rel_msg = 0;
	int rel_type = type;
	int rel_code = code;
A
Al Viro 已提交
486
	__u32 rel_info = ntohl(info);
487 488 489 490 491 492
	int err;
	struct sk_buff *skb2;
	struct iphdr *eiph;
	struct flowi fl;
	struct rtable *rt;

493 494
	err = ip6_tnl_err(skb, IPPROTO_IPIP, opt, &rel_type, &rel_code,
			  &rel_msg, &rel_info, offset);
495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527
	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);
528
	skb_reset_network_header(skb2);
529
	eiph = ip_hdr(skb2);
530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568

	/* 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;
	if (ip_route_output_key(&rt, &fl))
		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;
		if (ip_route_output_key(&rt, &fl) ||
		    rt->u.dst.dev->type != ARPHRD_TUNNEL) {
			ip_rt_put(rt);
			goto out;
		}
	} 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
569
	icmp_send(skb2, rel_type, rel_code, htonl(rel_info));
570 571 572 573 574 575

out:
	kfree_skb(skb2);
	return 0;
}

576 577
static int
ip6ip6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
A
Al Viro 已提交
578
	   int type, int code, int offset, __be32 info)
579 580 581 582
{
	int rel_msg = 0;
	int rel_type = type;
	int rel_code = code;
A
Al Viro 已提交
583
	__u32 rel_info = ntohl(info);
584 585
	int err;

586 587
	err = ip6_tnl_err(skb, IPPROTO_IPV6, opt, &rel_type, &rel_code,
			  &rel_msg, &rel_info, offset);
588 589 590 591
	if (err < 0)
		return err;

	if (rel_msg && pskb_may_pull(skb, offset + sizeof(struct ipv6hdr))) {
L
Linus Torvalds 已提交
592 593
		struct rt6_info *rt;
		struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
594

L
Linus Torvalds 已提交
595
		if (!skb2)
596
			return 0;
L
Linus Torvalds 已提交
597 598 599 600

		dst_release(skb2->dst);
		skb2->dst = NULL;
		skb_pull(skb2, offset);
601
		skb_reset_network_header(skb2);
L
Linus Torvalds 已提交
602 603

		/* Try to guess incoming interface */
604
		rt = rt6_lookup(&ipv6_hdr(skb2)->saddr, NULL, 0, 0);
L
Linus Torvalds 已提交
605 606 607 608 609 610 611 612 613 614 615

		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);
	}
616 617

	return 0;
L
Linus Torvalds 已提交
618 619
}

620 621 622 623 624 625 626
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)
627
		ipv4_change_dsfield(ip_hdr(skb), INET_ECN_MASK, dsfield);
628 629

	if (INET_ECN_is_ce(dsfield))
630
		IP_ECN_set_ce(ip_hdr(skb));
631 632
}

633 634 635
static void ip6ip6_dscp_ecn_decapsulate(struct ip6_tnl *t,
					struct ipv6hdr *ipv6h,
					struct sk_buff *skb)
L
Linus Torvalds 已提交
636
{
637
	if (t->parms.flags & IP6_TNL_F_RCV_DSCP_COPY)
638
		ipv6_copy_dscp(ipv6h, ipv6_hdr(skb));
L
Linus Torvalds 已提交
639

640
	if (INET_ECN_is_ce(ipv6_get_dsfield(ipv6h)))
641
		IP6_ECN_set_ce(ipv6_hdr(skb));
L
Linus Torvalds 已提交
642
}
643

644 645 646 647 648 649
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) {
650
		struct net_device *ldev = NULL;
651 652

		if (p->link)
653
			ldev = dev_get_by_index(&init_net, p->link);
654 655 656 657 658 659 660 661 662 663 664

		if ((ipv6_addr_is_multicast(&p->laddr) ||
		     likely(ipv6_chk_addr(&p->laddr, ldev, 0))) &&
		    likely(!ipv6_chk_addr(&p->raddr, NULL, 0)))
			ret = 1;

		if (ldev)
			dev_put(ldev);
	}
	return ret;
}
L
Linus Torvalds 已提交
665 666

/**
667
 * ip6_tnl_rcv - decapsulate IPv6 packet and retransmit it locally
L
Linus Torvalds 已提交
668
 *   @skb: received socket buffer
669 670
 *   @protocol: ethernet protocol ID
 *   @dscp_ecn_decapsulate: the function to decapsulate DSCP code and ECN
L
Linus Torvalds 已提交
671 672 673 674
 *
 * Return: 0
 **/

675
static int ip6_tnl_rcv(struct sk_buff *skb, __u16 protocol,
676
		       __u8 ipproto,
677 678 679
		       void (*dscp_ecn_decapsulate)(struct ip6_tnl *t,
						    struct ipv6hdr *ipv6h,
						    struct sk_buff *skb))
L
Linus Torvalds 已提交
680 681
{
	struct ip6_tnl *t;
682
	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
L
Linus Torvalds 已提交
683

684
	read_lock(&ip6_tnl_lock);
L
Linus Torvalds 已提交
685

686
	if ((t = ip6_tnl_lookup(&ipv6h->saddr, &ipv6h->daddr)) != NULL) {
687 688 689 690 691
		if (t->parms.proto != ipproto && t->parms.proto != 0) {
			read_unlock(&ip6_tnl_lock);
			goto discard;
		}

L
Linus Torvalds 已提交
692
		if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
693
			read_unlock(&ip6_tnl_lock);
694
			goto discard;
L
Linus Torvalds 已提交
695 696
		}

697
		if (!ip6_tnl_rcv_ctl(t)) {
L
Linus Torvalds 已提交
698
			t->stat.rx_dropped++;
699
			read_unlock(&ip6_tnl_lock);
L
Linus Torvalds 已提交
700 701 702
			goto discard;
		}
		secpath_reset(skb);
703
		skb->mac_header = skb->network_header;
704
		skb_reset_network_header(skb);
705
		skb->protocol = htons(protocol);
L
Linus Torvalds 已提交
706 707 708 709 710
		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;
711
		nf_reset(skb);
712 713 714

		dscp_ecn_decapsulate(t, ipv6h, skb);

L
Linus Torvalds 已提交
715 716 717
		t->stat.rx_packets++;
		t->stat.rx_bytes += skb->len;
		netif_rx(skb);
718
		read_unlock(&ip6_tnl_lock);
L
Linus Torvalds 已提交
719 720
		return 0;
	}
721
	read_unlock(&ip6_tnl_lock);
L
Linus Torvalds 已提交
722
	return 1;
723 724 725 726

discard:
	kfree_skb(skb);
	return 0;
L
Linus Torvalds 已提交
727 728
}

729 730
static int ip4ip6_rcv(struct sk_buff *skb)
{
731 732
	return ip6_tnl_rcv(skb, ETH_P_IP, IPPROTO_IPIP,
			   ip4ip6_dscp_ecn_decapsulate);
733 734
}

735 736
static int ip6ip6_rcv(struct sk_buff *skb)
{
737 738
	return ip6_tnl_rcv(skb, ETH_P_IPV6, IPPROTO_IPV6,
			   ip6ip6_dscp_ecn_decapsulate);
739 740
}

741 742 743 744
struct ipv6_tel_txoption {
	struct ipv6_txoptions ops;
	__u8 dst_opt[8];
};
L
Linus Torvalds 已提交
745

746 747 748
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
749

750 751 752 753 754
	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
755

756 757
	opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
	opt->ops.opt_nflen = 8;
L
Linus Torvalds 已提交
758 759 760
}

/**
761
 * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
L
Linus Torvalds 已提交
762
 *   @t: the outgoing tunnel device
763
 *   @hdr: IPv6 header from the incoming packet
L
Linus Torvalds 已提交
764 765
 *
 * Description:
766
 *   Avoid trivial tunneling loop by checking that tunnel exit-point
L
Linus Torvalds 已提交
767 768
 *   doesn't match source of incoming packet.
 *
769
 * Return:
L
Linus Torvalds 已提交
770 771 772 773 774
 *   1 if conflict,
 *   0 else
 **/

static inline int
775
ip6_tnl_addr_conflict(struct ip6_tnl *t, struct ipv6hdr *hdr)
L
Linus Torvalds 已提交
776 777 778 779
{
	return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
}

780 781 782 783 784
static inline int ip6_tnl_xmit_ctl(struct ip6_tnl *t)
{
	struct ip6_tnl_parm *p = &t->parms;
	int ret = 0;

785
	if (p->flags & IP6_TNL_F_CAP_XMIT) {
786 787 788
		struct net_device *ldev = NULL;

		if (p->link)
789
			ldev = dev_get_by_index(&init_net, p->link);
790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807

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

827 828 829 830 831 832
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
833
{
834
	struct ip6_tnl *t = netdev_priv(dev);
L
Linus Torvalds 已提交
835
	struct net_device_stats *stats = &t->stat;
836
	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
837
	struct ipv6_tel_txoption opt;
L
Linus Torvalds 已提交
838 839 840
	struct dst_entry *dst;
	struct net_device *tdev;
	int mtu;
841
	unsigned int max_headroom = sizeof(struct ipv6hdr);
L
Linus Torvalds 已提交
842
	u8 proto;
843
	int err = -1;
L
Linus Torvalds 已提交
844 845 846 847
	int pkt_len;

	if ((dst = ip6_tnl_dst_check(t)) != NULL)
		dst_hold(dst);
848
	else {
849
		dst = ip6_route_output(NULL, fl);
L
Linus Torvalds 已提交
850

851
		if (dst->error || xfrm_lookup(&dst, fl, NULL, 0) < 0)
852 853
			goto tx_err_link_failure;
	}
L
Linus Torvalds 已提交
854 855 856 857 858 859

	tdev = dst->dev;

	if (tdev == dev) {
		stats->collisions++;
		if (net_ratelimit())
860
			printk(KERN_WARNING
L
Linus Torvalds 已提交
861 862 863 864 865
			       "%s: Local routing loop detected!\n",
			       t->parms.name);
		goto tx_err_dst_release;
	}
	mtu = dst_mtu(dst) - sizeof (*ipv6h);
866
	if (encap_limit >= 0) {
L
Linus Torvalds 已提交
867 868 869 870 871
		max_headroom += 8;
		mtu -= 8;
	}
	if (mtu < IPV6_MIN_MTU)
		mtu = IPV6_MIN_MTU;
872 873
	if (skb->dst)
		skb->dst->ops->update_pmtu(skb->dst, mtu);
L
Linus Torvalds 已提交
874
	if (skb->len > mtu) {
875 876
		*pmtu = mtu;
		err = -EMSGSIZE;
L
Linus Torvalds 已提交
877 878 879 880 881 882 883
		goto tx_err_dst_release;
	}

	/*
	 * Okay, now see if we can stuff it in the buffer as-is.
	 */
	max_headroom += LL_RESERVED_SPACE(tdev);
884

885 886
	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
L
Linus Torvalds 已提交
887
		struct sk_buff *new_skb;
888

L
Linus Torvalds 已提交
889 890 891 892 893 894 895 896 897 898 899
		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);

900
	skb->transport_header = skb->network_header;
L
Linus Torvalds 已提交
901

902
	proto = fl->proto;
903 904 905 906
	if (encap_limit >= 0) {
		init_tel_txopt(&opt, encap_limit);
		ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
	}
907 908
	skb_push(skb, sizeof(struct ipv6hdr));
	skb_reset_network_header(skb);
909
	ipv6h = ipv6_hdr(skb);
910
	*(__be32*)ipv6h = fl->fl6_flowlabel | htonl(0x60000000);
L
Linus Torvalds 已提交
911 912 913 914 915
	dsfield = INET_ECN_encapsulate(0, dsfield);
	ipv6_change_dsfield(ipv6h, ~INET_ECN_MASK, dsfield);
	ipv6h->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
	ipv6h->hop_limit = t->parms.hop_limit;
	ipv6h->nexthdr = proto;
916 917
	ipv6_addr_copy(&ipv6h->saddr, &fl->fl6_src);
	ipv6_addr_copy(&ipv6h->daddr, &fl->fl6_dst);
L
Linus Torvalds 已提交
918 919
	nf_reset(skb);
	pkt_len = skb->len;
920
	err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL,
L
Linus Torvalds 已提交
921 922
		      skb->dst->dev, dst_output);

923
	if (net_xmit_eval(err) == 0) {
L
Linus Torvalds 已提交
924 925 926 927 928 929 930 931 932 933 934 935 936
		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);
937 938 939
	return err;
}

940 941 942 943
static inline int
ip4ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
{
	struct ip6_tnl *t = netdev_priv(dev);
944
	struct iphdr  *iph = ip_hdr(skb);
945 946 947 948 949 950
	int encap_limit = -1;
	struct flowi fl;
	__u8 dsfield;
	__u32 mtu;
	int err;

951 952
	if ((t->parms.proto != IPPROTO_IPIP && t->parms.proto != 0) ||
	    !ip6_tnl_xmit_ctl(t))
953 954 955 956 957 958 959 960 961 962 963
		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
964 965
		fl.fl6_flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
					  & IPV6_TCLASS_MASK;
966 967 968 969 970 971 972 973 974 975 976 977 978

	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;
}

979 980 981 982
static inline int
ip6ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
{
	struct ip6_tnl *t = netdev_priv(dev);
983
	struct ipv6hdr *ipv6h = ipv6_hdr(skb);
984 985 986 987 988 989 990
	int encap_limit = -1;
	__u16 offset;
	struct flowi fl;
	__u8 dsfield;
	__u32 mtu;
	int err;

991 992
	if ((t->parms.proto != IPPROTO_IPV6 && t->parms.proto != 0) ||
	    !ip6_tnl_xmit_ctl(t) || ip6_tnl_addr_conflict(t, ipv6h))
993 994
		return -1;

995 996
	offset = parse_tlv_tnl_enc_lim(skb, skb_network_header(skb));
	if (offset > 0) {
997
		struct ipv6_tlv_tnl_enc_lim *tel;
998
		tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 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
		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) {
1040 1041 1042
	case __constant_htons(ETH_P_IP):
		ret = ip4ip6_tnl_xmit(skb, dev);
		break;
1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055
	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
1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066
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;
1067 1068
	int ltype = ipv6_addr_type(&p->laddr);
	int rtype = ipv6_addr_type(&p->raddr);
L
Linus Torvalds 已提交
1069 1070 1071

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

1072 1073 1074
	if (ltype & (IPV6_ADDR_UNICAST|IPV6_ADDR_MULTICAST) &&
	    rtype & (IPV6_ADDR_UNICAST|IPV6_ADDR_MULTICAST) &&
	    !((ltype|rtype) & IPV6_ADDR_LOOPBACK) &&
1075
	    (!((ltype|rtype) & IPV6_ADDR_LINKLOCAL) || p->link)) {
1076 1077 1078 1079
		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
1080 1081 1082
	}
}

1083
static void ip6_tnl_link_config(struct ip6_tnl *t)
L
Linus Torvalds 已提交
1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112
{
	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) {
1113 1114 1115
		int strict = (ipv6_addr_type(&p->raddr) &
			      (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));

L
Linus Torvalds 已提交
1116
		struct rt6_info *rt = rt6_lookup(&p->raddr, &p->laddr,
1117
						 p->link, strict);
L
Linus Torvalds 已提交
1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135

		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);
	}
}

/**
1136
 * ip6_tnl_change - update the tunnel parameters
L
Linus Torvalds 已提交
1137 1138 1139 1140 1141
 *   @t: tunnel to be changed
 *   @p: tunnel configuration parameters
 *   @active: != 0 if tunnel is ready for use
 *
 * Description:
1142
 *   ip6_tnl_change() updates the tunnel parameters
L
Linus Torvalds 已提交
1143 1144 1145
 **/

static int
1146
ip6_tnl_change(struct ip6_tnl *t, struct ip6_tnl_parm *p)
L
Linus Torvalds 已提交
1147 1148 1149 1150 1151 1152 1153
{
	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;
1154
	t->parms.link = p->link;
1155
	t->parms.proto = p->proto;
1156
	ip6_tnl_dst_reset(t);
1157
	ip6_tnl_link_config(t);
L
Linus Torvalds 已提交
1158 1159 1160 1161
	return 0;
}

/**
1162
 * ip6_tnl_ioctl - configure ipv6 tunnels from userspace
L
Linus Torvalds 已提交
1163 1164 1165 1166 1167
 *   @dev: virtual device associated with tunnel
 *   @ifr: parameters passed from userspace
 *   @cmd: command to be performed
 *
 * Description:
1168
 *   ip6_tnl_ioctl() is used for managing IPv6 tunnels
1169
 *   from userspace.
L
Linus Torvalds 已提交
1170 1171 1172 1173 1174 1175 1176
 *
 *   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
 *
1177
 *   The fallback device "ip6tnl0", created during module
L
Linus Torvalds 已提交
1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189
 *   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
1190
ip6_tnl_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
L
Linus Torvalds 已提交
1191 1192 1193 1194 1195 1196 1197
{
	int err = 0;
	struct ip6_tnl_parm p;
	struct ip6_tnl *t = NULL;

	switch (cmd) {
	case SIOCGETTUNNEL:
1198
		if (dev == ip6_fb_tnl_dev) {
1199
			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof (p))) {
L
Linus Torvalds 已提交
1200 1201 1202
				err = -EFAULT;
				break;
			}
1203
			t = ip6_tnl_locate(&p, 0);
1204 1205
		}
		if (t == NULL)
1206
			t = netdev_priv(dev);
L
Linus Torvalds 已提交
1207 1208 1209 1210 1211 1212 1213 1214 1215 1216
		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;
1217 1218
		err = -EFAULT;
		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof (p)))
L
Linus Torvalds 已提交
1219
			break;
1220
		err = -EINVAL;
1221 1222
		if (p.proto != IPPROTO_IPV6 && p.proto != IPPROTO_IPIP &&
		    p.proto != 0)
L
Linus Torvalds 已提交
1223
			break;
1224 1225
		t = ip6_tnl_locate(&p, cmd == SIOCADDTUNNEL);
		if (dev != ip6_fb_tnl_dev && cmd == SIOCCHGTUNNEL) {
1226 1227 1228 1229 1230 1231 1232 1233
			if (t != NULL) {
				if (t->dev != dev) {
					err = -EEXIST;
					break;
				}
			} else
				t = netdev_priv(dev);

1234 1235 1236
			ip6_tnl_unlink(t);
			err = ip6_tnl_change(t, &p);
			ip6_tnl_link(t);
L
Linus Torvalds 已提交
1237 1238
			netdev_state_change(dev);
		}
1239
		if (t) {
L
Linus Torvalds 已提交
1240
			err = 0;
1241 1242 1243 1244 1245
			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
1246 1247 1248 1249 1250 1251
		break;
	case SIOCDELTUNNEL:
		err = -EPERM;
		if (!capable(CAP_NET_ADMIN))
			break;

1252
		if (dev == ip6_fb_tnl_dev) {
1253 1254
			err = -EFAULT;
			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof (p)))
L
Linus Torvalds 已提交
1255
				break;
1256
			err = -ENOENT;
1257
			if ((t = ip6_tnl_locate(&p, 0)) == NULL)
L
Linus Torvalds 已提交
1258
				break;
1259
			err = -EPERM;
1260
			if (t->dev == ip6_fb_tnl_dev)
L
Linus Torvalds 已提交
1261
				break;
1262
			dev = t->dev;
L
Linus Torvalds 已提交
1263
		}
1264 1265
		err = 0;
		unregister_netdevice(dev);
L
Linus Torvalds 已提交
1266 1267 1268 1269 1270 1271 1272 1273
		break;
	default:
		err = -EINVAL;
	}
	return err;
}

/**
1274
 * ip6_tnl_get_stats - return the stats for tunnel device
L
Linus Torvalds 已提交
1275 1276 1277 1278 1279 1280
 *   @dev: virtual device associated with tunnel
 *
 * Return: stats for device
 **/

static struct net_device_stats *
1281
ip6_tnl_get_stats(struct net_device *dev)
L
Linus Torvalds 已提交
1282
{
1283
	return &(((struct ip6_tnl *)netdev_priv(dev))->stat);
L
Linus Torvalds 已提交
1284 1285 1286
}

/**
1287
 * ip6_tnl_change_mtu - change mtu manually for tunnel device
L
Linus Torvalds 已提交
1288 1289 1290 1291 1292 1293 1294 1295 1296
 *   @dev: virtual device associated with tunnel
 *   @new_mtu: the new mtu
 *
 * Return:
 *   0 on success,
 *   %-EINVAL if mtu too small
 **/

static int
1297
ip6_tnl_change_mtu(struct net_device *dev, int new_mtu)
L
Linus Torvalds 已提交
1298 1299 1300 1301 1302 1303 1304 1305 1306
{
	if (new_mtu < IPV6_MIN_MTU) {
		return -EINVAL;
	}
	dev->mtu = new_mtu;
	return 0;
}

/**
1307
 * ip6_tnl_dev_setup - setup virtual tunnel device
L
Linus Torvalds 已提交
1308 1309 1310 1311 1312 1313
 *   @dev: virtual device associated with tunnel
 *
 * Description:
 *   Initialize function pointers and device parameters
 **/

1314
static void ip6_tnl_dev_setup(struct net_device *dev)
L
Linus Torvalds 已提交
1315
{
1316
	dev->uninit = ip6_tnl_dev_uninit;
L
Linus Torvalds 已提交
1317
	dev->destructor = free_netdev;
1318
	dev->hard_start_xmit = ip6_tnl_xmit;
1319 1320 1321
	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
1322 1323 1324 1325 1326 1327 1328 1329 1330 1331

	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);
}


/**
1332
 * ip6_tnl_dev_init_gen - general initializer for all tunnel devices
L
Linus Torvalds 已提交
1333 1334 1335 1336
 *   @dev: virtual device associated with tunnel
 **/

static inline void
1337
ip6_tnl_dev_init_gen(struct net_device *dev)
L
Linus Torvalds 已提交
1338
{
1339
	struct ip6_tnl *t = netdev_priv(dev);
L
Linus Torvalds 已提交
1340 1341 1342 1343 1344
	t->dev = dev;
	strcpy(t->parms.name, dev->name);
}

/**
1345
 * ip6_tnl_dev_init - initializer for all non fallback tunnel devices
L
Linus Torvalds 已提交
1346 1347 1348 1349
 *   @dev: virtual device associated with tunnel
 **/

static int
1350
ip6_tnl_dev_init(struct net_device *dev)
L
Linus Torvalds 已提交
1351
{
1352
	struct ip6_tnl *t = netdev_priv(dev);
1353 1354
	ip6_tnl_dev_init_gen(dev);
	ip6_tnl_link_config(t);
L
Linus Torvalds 已提交
1355 1356 1357 1358
	return 0;
}

/**
1359
 * ip6_fb_tnl_dev_init - initializer for fallback tunnel device
L
Linus Torvalds 已提交
1360 1361 1362 1363 1364
 *   @dev: fallback device
 *
 * Return: 0
 **/

1365
static int
1366
ip6_fb_tnl_dev_init(struct net_device *dev)
L
Linus Torvalds 已提交
1367
{
1368
	struct ip6_tnl *t = netdev_priv(dev);
1369
	ip6_tnl_dev_init_gen(dev);
1370
	t->parms.proto = IPPROTO_IPV6;
L
Linus Torvalds 已提交
1371 1372 1373 1374 1375
	dev_hold(dev);
	tnls_wc[0] = t;
	return 0;
}

1376 1377 1378 1379 1380 1381
static struct xfrm6_tunnel ip4ip6_handler = {
	.handler	= ip4ip6_rcv,
	.err_handler	= ip4ip6_err,
	.priority	=	1,
};

L
Linus Torvalds 已提交
1382
static struct xfrm6_tunnel ip6ip6_handler = {
1383 1384
	.handler	= ip6ip6_rcv,
	.err_handler	= ip6ip6_err,
H
Herbert Xu 已提交
1385
	.priority	=	1,
L
Linus Torvalds 已提交
1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397
};

/**
 * ip6_tunnel_init - register protocol and reserve needed resources
 *
 * Return: 0 on success
 **/

static int __init ip6_tunnel_init(void)
{
	int  err;

1398
	if (xfrm6_tunnel_register(&ip4ip6_handler, AF_INET)) {
1399
		printk(KERN_ERR "ip6_tunnel init: can't register ip4ip6\n");
1400 1401 1402 1403
		err = -EAGAIN;
		goto out;
	}

1404
	if (xfrm6_tunnel_register(&ip6ip6_handler, AF_INET6)) {
1405
		printk(KERN_ERR "ip6_tunnel init: can't register ip6ip6\n");
1406 1407
		err = -EAGAIN;
		goto unreg_ip4ip6;
L
Linus Torvalds 已提交
1408
	}
1409 1410
	ip6_fb_tnl_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6tnl0",
				      ip6_tnl_dev_setup);
L
Linus Torvalds 已提交
1411

1412
	if (!ip6_fb_tnl_dev) {
L
Linus Torvalds 已提交
1413 1414 1415
		err = -ENOMEM;
		goto fail;
	}
1416
	ip6_fb_tnl_dev->init = ip6_fb_tnl_dev_init;
L
Linus Torvalds 已提交
1417

1418 1419
	if ((err = register_netdev(ip6_fb_tnl_dev))) {
		free_netdev(ip6_fb_tnl_dev);
L
Linus Torvalds 已提交
1420 1421 1422 1423
		goto fail;
	}
	return 0;
fail:
1424
	xfrm6_tunnel_deregister(&ip6ip6_handler, AF_INET6);
1425 1426 1427
unreg_ip4ip6:
	xfrm6_tunnel_deregister(&ip4ip6_handler, AF_INET);
out:
L
Linus Torvalds 已提交
1428 1429 1430
	return err;
}

1431
static void __exit ip6_tnl_destroy_tunnels(void)
1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444
{
	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
1445 1446 1447 1448 1449 1450
/**
 * ip6_tunnel_cleanup - free resources and unregister protocol
 **/

static void __exit ip6_tunnel_cleanup(void)
{
1451
	if (xfrm6_tunnel_deregister(&ip4ip6_handler, AF_INET))
1452
		printk(KERN_INFO "ip6_tunnel close: can't deregister ip4ip6\n");
1453

1454
	if (xfrm6_tunnel_deregister(&ip6ip6_handler, AF_INET6))
1455
		printk(KERN_INFO "ip6_tunnel close: can't deregister ip6ip6\n");
L
Linus Torvalds 已提交
1456

1457
	rtnl_lock();
1458
	ip6_tnl_destroy_tunnels();
1459
	rtnl_unlock();
L
Linus Torvalds 已提交
1460 1461 1462 1463
}

module_init(ip6_tunnel_init);
module_exit(ip6_tunnel_cleanup);