sit.c 28.9 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5
/*
 *	IPv6 over IPv4 tunnel device - Simple Internet Transition (SIT)
 *	Linux INET6 implementation
 *
 *	Authors:
6
 *	Pedro Roque		<roque@di.fc.ul.pt>
L
Linus Torvalds 已提交
7 8 9 10 11 12 13 14 15 16
 *	Alexey Kuznetsov	<kuznet@ms2.inr.ac.ru>
 *
 *	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.
 *
 *	Changes:
 * Roger Venning <r.venning@telstra.com>:	6to4 support
 * Nate Thompson <nate@thebog.net>:		6to4 support
17
 * Fred Templin <fred.l.templin@boeing.com>:	isatap support
L
Linus Torvalds 已提交
18 19 20
 */

#include <linux/module.h>
21
#include <linux/capability.h>
L
Linus Torvalds 已提交
22 23 24 25 26 27 28 29 30
#include <linux/errno.h>
#include <linux/types.h>
#include <linux/socket.h>
#include <linux/sockios.h>
#include <linux/net.h>
#include <linux/in6.h>
#include <linux/netdevice.h>
#include <linux/if_arp.h>
#include <linux/icmp.h>
31
#include <linux/slab.h>
L
Linus Torvalds 已提交
32 33 34
#include <asm/uaccess.h>
#include <linux/init.h>
#include <linux/netfilter_ipv4.h>
35
#include <linux/if_ether.h>
L
Linus Torvalds 已提交
36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53

#include <net/sock.h>
#include <net/snmp.h>

#include <net/ipv6.h>
#include <net/protocol.h>
#include <net/transp_v6.h>
#include <net/ip6_fib.h>
#include <net/ip6_route.h>
#include <net/ndisc.h>
#include <net/addrconf.h>
#include <net/ip.h>
#include <net/udp.h>
#include <net/icmp.h>
#include <net/ipip.h>
#include <net/inet_ecn.h>
#include <net/xfrm.h>
#include <net/dsfield.h>
54 55
#include <net/net_namespace.h>
#include <net/netns/generic.h>
L
Linus Torvalds 已提交
56 57 58 59 60 61 62 63

/*
   This version of net/ipv6/sit.c is cloned of net/ipv4/ip_gre.c

   For comments look at net/ipv4/ip_gre.c --ANK
 */

#define HASH_SIZE  16
A
Al Viro 已提交
64
#define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
L
Linus Torvalds 已提交
65

E
Eric Dumazet 已提交
66
static int ipip6_tunnel_init(struct net_device *dev);
L
Linus Torvalds 已提交
67
static void ipip6_tunnel_setup(struct net_device *dev);
E
Eric Dumazet 已提交
68
static void ipip6_dev_free(struct net_device *dev);
L
Linus Torvalds 已提交
69

70
static int sit_net_id __read_mostly;
71
struct sit_net {
E
Eric Dumazet 已提交
72 73 74 75 76
	struct ip_tunnel __rcu *tunnels_r_l[HASH_SIZE];
	struct ip_tunnel __rcu *tunnels_r[HASH_SIZE];
	struct ip_tunnel __rcu *tunnels_l[HASH_SIZE];
	struct ip_tunnel __rcu *tunnels_wc[1];
	struct ip_tunnel __rcu **tunnels[4];
77

78
	struct net_device *fb_tunnel_dev;
79 80
};

E
Eric Dumazet 已提交
81
/*
E
Eric Dumazet 已提交
82
 * Locking : hash tables are protected by RCU and RTNL
E
Eric Dumazet 已提交
83 84 85 86
 */

#define for_each_ip_tunnel_rcu(start) \
	for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
L
Linus Torvalds 已提交
87

E
Eric Dumazet 已提交
88 89 90 91 92 93
/* often modified stats are per cpu, other are shared (netdev->stats) */
struct pcpu_tstats {
	unsigned long	rx_packets;
	unsigned long	rx_bytes;
	unsigned long	tx_packets;
	unsigned long	tx_bytes;
E
Eric Dumazet 已提交
94
} __attribute__((aligned(4*sizeof(unsigned long))));
E
Eric Dumazet 已提交
95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114

static struct net_device_stats *ipip6_get_stats(struct net_device *dev)
{
	struct pcpu_tstats sum = { 0 };
	int i;

	for_each_possible_cpu(i) {
		const struct pcpu_tstats *tstats = per_cpu_ptr(dev->tstats, i);

		sum.rx_packets += tstats->rx_packets;
		sum.rx_bytes   += tstats->rx_bytes;
		sum.tx_packets += tstats->tx_packets;
		sum.tx_bytes   += tstats->tx_bytes;
	}
	dev->stats.rx_packets = sum.rx_packets;
	dev->stats.rx_bytes   = sum.rx_bytes;
	dev->stats.tx_packets = sum.tx_packets;
	dev->stats.tx_bytes   = sum.tx_bytes;
	return &dev->stats;
}
E
Eric Dumazet 已提交
115 116 117
/*
 * Must be invoked with rcu_read_lock
 */
118
static struct ip_tunnel * ipip6_tunnel_lookup(struct net *net,
119
		struct net_device *dev, __be32 remote, __be32 local)
L
Linus Torvalds 已提交
120
{
E
Eric Dumazet 已提交
121 122
	unsigned int h0 = HASH(remote);
	unsigned int h1 = HASH(local);
L
Linus Torvalds 已提交
123
	struct ip_tunnel *t;
124
	struct sit_net *sitn = net_generic(net, sit_net_id);
L
Linus Torvalds 已提交
125

E
Eric Dumazet 已提交
126
	for_each_ip_tunnel_rcu(sitn->tunnels_r_l[h0 ^ h1]) {
L
Linus Torvalds 已提交
127
		if (local == t->parms.iph.saddr &&
128 129 130
		    remote == t->parms.iph.daddr &&
		    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
		    (t->dev->flags & IFF_UP))
L
Linus Torvalds 已提交
131 132
			return t;
	}
E
Eric Dumazet 已提交
133
	for_each_ip_tunnel_rcu(sitn->tunnels_r[h0]) {
134 135 136
		if (remote == t->parms.iph.daddr &&
		    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
		    (t->dev->flags & IFF_UP))
L
Linus Torvalds 已提交
137 138
			return t;
	}
E
Eric Dumazet 已提交
139
	for_each_ip_tunnel_rcu(sitn->tunnels_l[h1]) {
140 141 142
		if (local == t->parms.iph.saddr &&
		    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
		    (t->dev->flags & IFF_UP))
L
Linus Torvalds 已提交
143 144
			return t;
	}
E
Eric Dumazet 已提交
145
	t = rcu_dereference(sitn->tunnels_wc[0]);
146
	if ((t != NULL) && (t->dev->flags & IFF_UP))
L
Linus Torvalds 已提交
147 148 149 150
		return t;
	return NULL;
}

E
Eric Dumazet 已提交
151
static struct ip_tunnel __rcu **__ipip6_bucket(struct sit_net *sitn,
152
		struct ip_tunnel_parm *parms)
L
Linus Torvalds 已提交
153
{
154 155
	__be32 remote = parms->iph.daddr;
	__be32 local = parms->iph.saddr;
E
Eric Dumazet 已提交
156
	unsigned int h = 0;
L
Linus Torvalds 已提交
157 158 159 160 161 162 163 164 165 166
	int prio = 0;

	if (remote) {
		prio |= 2;
		h ^= HASH(remote);
	}
	if (local) {
		prio |= 1;
		h ^= HASH(local);
	}
167
	return &sitn->tunnels[prio][h];
L
Linus Torvalds 已提交
168 169
}

E
Eric Dumazet 已提交
170
static inline struct ip_tunnel __rcu **ipip6_bucket(struct sit_net *sitn,
171
		struct ip_tunnel *t)
172
{
173
	return __ipip6_bucket(sitn, &t->parms);
174 175
}

176
static void ipip6_tunnel_unlink(struct sit_net *sitn, struct ip_tunnel *t)
L
Linus Torvalds 已提交
177
{
E
Eric Dumazet 已提交
178 179 180 181 182 183 184
	struct ip_tunnel __rcu **tp;
	struct ip_tunnel *iter;

	for (tp = ipip6_bucket(sitn, t);
	     (iter = rtnl_dereference(*tp)) != NULL;
	     tp = &iter->next) {
		if (t == iter) {
185
			RCU_INIT_POINTER(*tp, t->next);
L
Linus Torvalds 已提交
186 187 188 189 190
			break;
		}
	}
}

191
static void ipip6_tunnel_link(struct sit_net *sitn, struct ip_tunnel *t)
L
Linus Torvalds 已提交
192
{
E
Eric Dumazet 已提交
193
	struct ip_tunnel __rcu **tp = ipip6_bucket(sitn, t);
L
Linus Torvalds 已提交
194

195 196
	RCU_INIT_POINTER(t->next, rtnl_dereference(*tp));
	RCU_INIT_POINTER(*tp, t);
L
Linus Torvalds 已提交
197 198
}

199
static void ipip6_tunnel_clone_6rd(struct net_device *dev, struct sit_net *sitn)
200 201
{
#ifdef CONFIG_IPV6_SIT_6RD
202 203
	struct ip_tunnel *t = netdev_priv(dev);

204 205 206 207 208 209 210 211 212 213 214 215
	if (t->dev == sitn->fb_tunnel_dev) {
		ipv6_addr_set(&t->ip6rd.prefix, htonl(0x20020000), 0, 0, 0);
		t->ip6rd.relay_prefix = 0;
		t->ip6rd.prefixlen = 16;
		t->ip6rd.relay_prefixlen = 0;
	} else {
		struct ip_tunnel *t0 = netdev_priv(sitn->fb_tunnel_dev);
		memcpy(&t->ip6rd, &t0->ip6rd, sizeof(t->ip6rd));
	}
#endif
}

E
Eric Dumazet 已提交
216
static struct ip_tunnel *ipip6_tunnel_locate(struct net *net,
217
		struct ip_tunnel_parm *parms, int create)
L
Linus Torvalds 已提交
218
{
A
Al Viro 已提交
219 220
	__be32 remote = parms->iph.daddr;
	__be32 local = parms->iph.saddr;
E
Eric Dumazet 已提交
221 222
	struct ip_tunnel *t, *nt;
	struct ip_tunnel __rcu **tp;
L
Linus Torvalds 已提交
223 224
	struct net_device *dev;
	char name[IFNAMSIZ];
225
	struct sit_net *sitn = net_generic(net, sit_net_id);
L
Linus Torvalds 已提交
226

E
Eric Dumazet 已提交
227 228 229
	for (tp = __ipip6_bucket(sitn, parms);
	    (t = rtnl_dereference(*tp)) != NULL;
	     tp = &t->next) {
230
		if (local == t->parms.iph.saddr &&
231 232
		    remote == t->parms.iph.daddr &&
		    parms->link == t->parms.link) {
233 234 235 236 237
			if (create)
				return NULL;
			else
				return t;
		}
L
Linus Torvalds 已提交
238 239 240 241 242 243
	}
	if (!create)
		goto failed;

	if (parms->name[0])
		strlcpy(name, parms->name, IFNAMSIZ);
244
	else
E
Eric Dumazet 已提交
245
		strcpy(name, "sit%d");
L
Linus Torvalds 已提交
246 247 248 249 250

	dev = alloc_netdev(sizeof(*t), name, ipip6_tunnel_setup);
	if (dev == NULL)
		return NULL;

251 252
	dev_net_set(dev, net);

253
	nt = netdev_priv(dev);
254

L
Linus Torvalds 已提交
255
	nt->parms = *parms;
E
Eric Dumazet 已提交
256 257
	if (ipip6_tunnel_init(dev) < 0)
		goto failed_free;
258
	ipip6_tunnel_clone_6rd(dev, sitn);
L
Linus Torvalds 已提交
259

F
Fred L. Templin 已提交
260 261 262
	if (parms->i_flags & SIT_ISATAP)
		dev->priv_flags |= IFF_ISATAP;

263 264
	if (register_netdevice(dev) < 0)
		goto failed_free;
L
Linus Torvalds 已提交
265 266 267

	dev_hold(dev);

268
	ipip6_tunnel_link(sitn, nt);
L
Linus Torvalds 已提交
269 270
	return nt;

271
failed_free:
E
Eric Dumazet 已提交
272
	ipip6_dev_free(dev);
L
Linus Torvalds 已提交
273 274 275 276
failed:
	return NULL;
}

E
Eric Dumazet 已提交
277 278 279 280 281
#define for_each_prl_rcu(start)			\
	for (prl = rcu_dereference(start);	\
	     prl;				\
	     prl = rcu_dereference(prl->next))

282
static struct ip_tunnel_prl_entry *
283
__ipip6_tunnel_locate_prl(struct ip_tunnel *t, __be32 addr)
284
{
E
Eric Dumazet 已提交
285
	struct ip_tunnel_prl_entry *prl;
286

E
Eric Dumazet 已提交
287 288
	for_each_prl_rcu(t->prl)
		if (prl->addr == addr)
289
			break;
E
Eric Dumazet 已提交
290
	return prl;
291 292 293

}

294 295
static int ipip6_tunnel_get_prl(struct ip_tunnel *t,
				struct ip_tunnel_prl __user *a)
296
{
297
	struct ip_tunnel_prl kprl, *kp;
298 299 300 301
	struct ip_tunnel_prl_entry *prl;
	unsigned int cmax, c = 0, ca, len;
	int ret = 0;

302 303 304 305
	if (copy_from_user(&kprl, a, sizeof(kprl)))
		return -EFAULT;
	cmax = kprl.datalen / sizeof(kprl);
	if (cmax > 1 && kprl.addr != htonl(INADDR_ANY))
306 307 308 309 310 311 312 313 314
		cmax = 1;

	/* For simple GET or for root users,
	 * we try harder to allocate.
	 */
	kp = (cmax <= 1 || capable(CAP_NET_ADMIN)) ?
		kcalloc(cmax, sizeof(*kp), GFP_KERNEL) :
		NULL;

E
Eric Dumazet 已提交
315
	rcu_read_lock();
316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332

	ca = t->prl_count < cmax ? t->prl_count : cmax;

	if (!kp) {
		/* We don't try hard to allocate much memory for
		 * non-root users.
		 * For root users, retry allocating enough memory for
		 * the answer.
		 */
		kp = kcalloc(ca, sizeof(*kp), GFP_ATOMIC);
		if (!kp) {
			ret = -ENOMEM;
			goto out;
		}
	}

	c = 0;
E
Eric Dumazet 已提交
333
	for_each_prl_rcu(t->prl) {
334
		if (c >= cmax)
335
			break;
336
		if (kprl.addr != htonl(INADDR_ANY) && prl->addr != kprl.addr)
337 338 339 340
			continue;
		kp[c].addr = prl->addr;
		kp[c].flags = prl->flags;
		c++;
341
		if (kprl.addr != htonl(INADDR_ANY))
342 343 344
			break;
	}
out:
E
Eric Dumazet 已提交
345
	rcu_read_unlock();
346 347

	len = sizeof(*kp) * c;
348 349 350
	ret = 0;
	if ((len && copy_to_user(a + 1, kp, len)) || put_user(len, &a->datalen))
		ret = -EFAULT;
351 352 353

	kfree(kp);

354
	return ret;
355 356
}

357 358 359 360
static int
ipip6_tunnel_add_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a, int chg)
{
	struct ip_tunnel_prl_entry *p;
361 362
	int err = 0;

363 364 365
	if (a->addr == htonl(INADDR_ANY))
		return -EINVAL;

E
Eric Dumazet 已提交
366
	ASSERT_RTNL();
367

E
Eric Dumazet 已提交
368
	for (p = rtnl_dereference(t->prl); p; p = rtnl_dereference(p->next)) {
369
		if (p->addr == a->addr) {
E
Eric Dumazet 已提交
370 371 372 373
			if (chg) {
				p->flags = a->flags;
				goto out;
			}
374 375
			err = -EEXIST;
			goto out;
376 377 378
		}
	}

379 380 381 382
	if (chg) {
		err = -ENXIO;
		goto out;
	}
383 384

	p = kzalloc(sizeof(struct ip_tunnel_prl_entry), GFP_KERNEL);
385 386 387 388
	if (!p) {
		err = -ENOBUFS;
		goto out;
	}
389 390

	p->next = t->prl;
391 392
	p->addr = a->addr;
	p->flags = a->flags;
E
Eric Dumazet 已提交
393
	t->prl_count++;
394
	RCU_INIT_POINTER(t->prl, p);
395 396
out:
	return err;
397 398
}

E
Eric Dumazet 已提交
399 400 401 402 403 404
static void prl_list_destroy_rcu(struct rcu_head *head)
{
	struct ip_tunnel_prl_entry *p, *n;

	p = container_of(head, struct ip_tunnel_prl_entry, rcu_head);
	do {
405
		n = rcu_dereference_protected(p->next, 1);
E
Eric Dumazet 已提交
406 407 408 409 410
		kfree(p);
		p = n;
	} while (p);
}

411 412 413
static int
ipip6_tunnel_del_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a)
{
414 415
	struct ip_tunnel_prl_entry *x;
	struct ip_tunnel_prl_entry __rcu **p;
416 417
	int err = 0;

E
Eric Dumazet 已提交
418
	ASSERT_RTNL();
419

420
	if (a && a->addr != htonl(INADDR_ANY)) {
421 422 423 424
		for (p = &t->prl;
		     (x = rtnl_dereference(*p)) != NULL;
		     p = &x->next) {
			if (x->addr == a->addr) {
425
				*p = x->next;
426
				kfree_rcu(x, rcu_head);
427
				t->prl_count--;
428
				goto out;
429 430
			}
		}
431
		err = -ENXIO;
432
	} else {
433 434
		x = rtnl_dereference(t->prl);
		if (x) {
E
Eric Dumazet 已提交
435 436 437
			t->prl_count = 0;
			call_rcu(&x->rcu_head, prl_list_destroy_rcu);
			t->prl = NULL;
438 439
		}
	}
440
out:
441
	return err;
442 443 444
}

static int
445
isatap_chksrc(struct sk_buff *skb, const struct iphdr *iph, struct ip_tunnel *t)
446
{
447
	struct ip_tunnel_prl_entry *p;
448 449
	int ok = 1;

E
Eric Dumazet 已提交
450
	rcu_read_lock();
451
	p = __ipip6_tunnel_locate_prl(t, iph->saddr);
452
	if (p) {
453
		if (p->flags & PRL_DEFAULT)
454 455 456 457
			skb->ndisc_nodetype = NDISC_NODETYPE_DEFAULT;
		else
			skb->ndisc_nodetype = NDISC_NODETYPE_NODEFAULT;
	} else {
458 459
		const struct in6_addr *addr6 = &ipv6_hdr(skb)->saddr;

460 461
		if (ipv6_addr_is_isatap(addr6) &&
		    (addr6->s6_addr32[3] == iph->saddr) &&
462
		    ipv6_chk_prefix(addr6, t->dev))
463 464 465 466
			skb->ndisc_nodetype = NDISC_NODETYPE_HOST;
		else
			ok = 0;
	}
E
Eric Dumazet 已提交
467
	rcu_read_unlock();
468 469 470
	return ok;
}

L
Linus Torvalds 已提交
471 472
static void ipip6_tunnel_uninit(struct net_device *dev)
{
473 474 475
	struct net *net = dev_net(dev);
	struct sit_net *sitn = net_generic(net, sit_net_id);

476
	if (dev == sitn->fb_tunnel_dev) {
477
		RCU_INIT_POINTER(sitn->tunnels_wc[0], NULL);
L
Linus Torvalds 已提交
478
	} else {
479
		ipip6_tunnel_unlink(sitn, netdev_priv(dev));
480
		ipip6_tunnel_del_prl(netdev_priv(dev), NULL);
L
Linus Torvalds 已提交
481
	}
E
Eric Dumazet 已提交
482
	dev_put(dev);
L
Linus Torvalds 已提交
483 484 485
}


486
static int ipip6_err(struct sk_buff *skb, u32 info)
L
Linus Torvalds 已提交
487 488
{

489
/* All the routers (except for Linux) return only
L
Linus Torvalds 已提交
490 491 492
   8 bytes of packet payload. It means, that precise relaying of
   ICMP in the real Internet is absolutely infeasible.
 */
493
	const struct iphdr *iph = (const struct iphdr *)skb->data;
494 495
	const int type = icmp_hdr(skb)->type;
	const int code = icmp_hdr(skb)->code;
L
Linus Torvalds 已提交
496
	struct ip_tunnel *t;
497
	int err;
L
Linus Torvalds 已提交
498 499 500 501

	switch (type) {
	default:
	case ICMP_PARAMETERPROB:
502
		return 0;
L
Linus Torvalds 已提交
503 504 505 506 507 508

	case ICMP_DEST_UNREACH:
		switch (code) {
		case ICMP_SR_FAILED:
		case ICMP_PORT_UNREACH:
			/* Impossible event. */
509
			return 0;
L
Linus Torvalds 已提交
510 511
		case ICMP_FRAG_NEEDED:
			/* Soft state for pmtu is maintained by IP core. */
512
			return 0;
L
Linus Torvalds 已提交
513 514 515 516 517 518 519 520 521 522
		default:
			/* All others are translated to HOST_UNREACH.
			   rfc2003 contains "deep thoughts" about NET_UNREACH,
			   I believe they are just ether pollution. --ANK
			 */
			break;
		}
		break;
	case ICMP_TIME_EXCEEDED:
		if (code != ICMP_EXC_TTL)
523
			return 0;
L
Linus Torvalds 已提交
524 525 526
		break;
	}

527 528
	err = -ENOENT;

E
Eric Dumazet 已提交
529
	rcu_read_lock();
530 531 532 533
	t = ipip6_tunnel_lookup(dev_net(skb->dev),
				skb->dev,
				iph->daddr,
				iph->saddr);
L
Linus Torvalds 已提交
534 535
	if (t == NULL || t->parms.iph.daddr == 0)
		goto out;
536 537

	err = 0;
L
Linus Torvalds 已提交
538 539 540
	if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
		goto out;

541
	if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
L
Linus Torvalds 已提交
542 543 544 545 546
		t->err_count++;
	else
		t->err_count = 1;
	t->err_time = jiffies;
out:
E
Eric Dumazet 已提交
547
	rcu_read_unlock();
548
	return err;
L
Linus Torvalds 已提交
549 550
}

551
static inline void ipip6_ecn_decapsulate(const struct iphdr *iph, struct sk_buff *skb)
L
Linus Torvalds 已提交
552 553
{
	if (INET_ECN_is_ce(iph->tos))
554
		IP6_ECN_set_ce(ipv6_hdr(skb));
L
Linus Torvalds 已提交
555 556 557 558
}

static int ipip6_rcv(struct sk_buff *skb)
{
559
	const struct iphdr *iph;
L
Linus Torvalds 已提交
560 561 562 563 564
	struct ip_tunnel *tunnel;

	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
		goto out;

565
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
566

E
Eric Dumazet 已提交
567
	rcu_read_lock();
568 569 570
	tunnel = ipip6_tunnel_lookup(dev_net(skb->dev), skb->dev,
				     iph->saddr, iph->daddr);
	if (tunnel != NULL) {
E
Eric Dumazet 已提交
571 572
		struct pcpu_tstats *tstats;

L
Linus Torvalds 已提交
573
		secpath_reset(skb);
574
		skb->mac_header = skb->network_header;
575
		skb_reset_network_header(skb);
576
		IPCB(skb)->flags = 0;
L
Linus Torvalds 已提交
577 578
		skb->protocol = htons(ETH_P_IPV6);
		skb->pkt_type = PACKET_HOST;
F
Fred L. Templin 已提交
579 580

		if ((tunnel->dev->priv_flags & IFF_ISATAP) &&
581
		    !isatap_chksrc(skb, iph, tunnel)) {
582
			tunnel->dev->stats.rx_errors++;
E
Eric Dumazet 已提交
583
			rcu_read_unlock();
F
Fred L. Templin 已提交
584 585 586
			kfree_skb(skb);
			return 0;
		}
587

E
Eric Dumazet 已提交
588 589 590 591 592
		tstats = this_cpu_ptr(tunnel->dev->tstats);
		tstats->rx_packets++;
		tstats->rx_bytes += skb->len;

		__skb_tunnel_rx(skb, tunnel->dev);
593

L
Linus Torvalds 已提交
594
		ipip6_ecn_decapsulate(iph, skb);
E
Eric Dumazet 已提交
595

596
		netif_rx(skb);
E
Eric Dumazet 已提交
597

E
Eric Dumazet 已提交
598
		rcu_read_unlock();
L
Linus Torvalds 已提交
599 600 601
		return 0;
	}

602
	/* no tunnel matched,  let upstream know, ipsec may handle it */
E
Eric Dumazet 已提交
603
	rcu_read_unlock();
604
	return 1;
L
Linus Torvalds 已提交
605
out:
606
	kfree_skb(skb);
L
Linus Torvalds 已提交
607 608 609
	return 0;
}

610 611 612 613 614
/*
 * Returns the embedded IPv4 address if the IPv6 address
 * comes from 6rd / 6to4 (RFC 3056) addr space.
 */
static inline
615
__be32 try_6rd(const struct in6_addr *v6dst, struct ip_tunnel *tunnel)
L
Linus Torvalds 已提交
616
{
A
Al Viro 已提交
617
	__be32 dst = 0;
L
Linus Torvalds 已提交
618

619 620 621
#ifdef CONFIG_IPV6_SIT_6RD
	if (ipv6_prefix_equal(v6dst, &tunnel->ip6rd.prefix,
			      tunnel->ip6rd.prefixlen)) {
E
Eric Dumazet 已提交
622
		unsigned int pbw0, pbi0;
623 624 625 626 627 628
		int pbi1;
		u32 d;

		pbw0 = tunnel->ip6rd.prefixlen >> 5;
		pbi0 = tunnel->ip6rd.prefixlen & 0x1f;

629
		d = (ntohl(v6dst->s6_addr32[pbw0]) << pbi0) >>
630 631 632 633
		    tunnel->ip6rd.relay_prefixlen;

		pbi1 = pbi0 - tunnel->ip6rd.relay_prefixlen;
		if (pbi1 > 0)
634
			d |= ntohl(v6dst->s6_addr32[pbw0 + 1]) >>
635 636 637 638 639
			     (32 - pbi1);

		dst = tunnel->ip6rd.relay_prefix | htonl(d);
	}
#else
L
Linus Torvalds 已提交
640
	if (v6dst->s6_addr16[0] == htons(0x2002)) {
641
		/* 6to4 v6 addr has 16 bits prefix, 32 v4addr, 16 SLA, ... */
L
Linus Torvalds 已提交
642 643
		memcpy(&dst, &v6dst->s6_addr16[1], 4);
	}
644
#endif
L
Linus Torvalds 已提交
645 646 647 648 649 650 651 652
	return dst;
}

/*
 *	This function assumes it is being called from dev_queue_xmit()
 *	and that skb is filled properly by that function.
 */

653 654
static netdev_tx_t ipip6_tunnel_xmit(struct sk_buff *skb,
				     struct net_device *dev)
L
Linus Torvalds 已提交
655
{
656
	struct ip_tunnel *tunnel = netdev_priv(dev);
E
Eric Dumazet 已提交
657
	struct pcpu_tstats *tstats;
658 659
	const struct iphdr  *tiph = &tunnel->parms.iph;
	const struct ipv6hdr *iph6 = ipv6_hdr(skb);
L
Linus Torvalds 已提交
660
	u8     tos = tunnel->parms.iph.tos;
661
	__be16 df = tiph->frag_off;
L
Linus Torvalds 已提交
662
	struct rtable *rt;     			/* Route to the other host */
E
Eric Dumazet 已提交
663
	struct net_device *tdev;		/* Device to other host */
L
Linus Torvalds 已提交
664
	struct iphdr  *iph;			/* Our new IP header */
665
	unsigned int max_headroom;		/* The extra header space needed */
A
Al Viro 已提交
666
	__be32 dst = tiph->daddr;
667
	struct flowi4 fl4;
L
Linus Torvalds 已提交
668
	int    mtu;
669
	const struct in6_addr *addr6;
L
Linus Torvalds 已提交
670 671 672 673 674
	int addr_type;

	if (skb->protocol != htons(ETH_P_IPV6))
		goto tx_error;

675 676 677
	if (tos == 1)
		tos = ipv6_get_dsfield(iph6);

F
Fred L. Templin 已提交
678 679 680 681
	/* ISATAP (RFC4214) - must come before 6to4 */
	if (dev->priv_flags & IFF_ISATAP) {
		struct neighbour *neigh = NULL;

E
Eric Dumazet 已提交
682
		if (skb_dst(skb))
683
			neigh = dst_get_neighbour(skb_dst(skb));
F
Fred L. Templin 已提交
684 685 686 687 688 689 690

		if (neigh == NULL) {
			if (net_ratelimit())
				printk(KERN_DEBUG "sit: nexthop == NULL\n");
			goto tx_error;
		}

691
		addr6 = (const struct in6_addr*)&neigh->primary_key;
F
Fred L. Templin 已提交
692 693 694 695 696 697 698 699 700
		addr_type = ipv6_addr_type(addr6);

		if ((addr_type & IPV6_ADDR_UNICAST) &&
		     ipv6_addr_is_isatap(addr6))
			dst = addr6->s6_addr32[3];
		else
			goto tx_error;
	}

L
Linus Torvalds 已提交
701
	if (!dst)
702
		dst = try_6rd(&iph6->daddr, tunnel);
L
Linus Torvalds 已提交
703 704 705 706

	if (!dst) {
		struct neighbour *neigh = NULL;

E
Eric Dumazet 已提交
707
		if (skb_dst(skb))
708
			neigh = dst_get_neighbour(skb_dst(skb));
L
Linus Torvalds 已提交
709 710 711 712 713 714 715

		if (neigh == NULL) {
			if (net_ratelimit())
				printk(KERN_DEBUG "sit: nexthop == NULL\n");
			goto tx_error;
		}

716
		addr6 = (const struct in6_addr*)&neigh->primary_key;
L
Linus Torvalds 已提交
717 718 719
		addr_type = ipv6_addr_type(addr6);

		if (addr_type == IPV6_ADDR_ANY) {
720
			addr6 = &ipv6_hdr(skb)->daddr;
L
Linus Torvalds 已提交
721 722 723 724 725 726 727 728 729
			addr_type = ipv6_addr_type(addr6);
		}

		if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
			goto tx_error_icmp;

		dst = addr6->s6_addr32[3];
	}

730
	rt = ip_route_output_ports(dev_net(dev), &fl4, NULL,
731 732 733 734 735 736 737
				   dst, tiph->saddr,
				   0, 0,
				   IPPROTO_IPV6, RT_TOS(tos),
				   tunnel->parms.link);
	if (IS_ERR(rt)) {
		dev->stats.tx_carrier_errors++;
		goto tx_error_icmp;
L
Linus Torvalds 已提交
738 739 740
	}
	if (rt->rt_type != RTN_UNICAST) {
		ip_rt_put(rt);
E
Eric Dumazet 已提交
741
		dev->stats.tx_carrier_errors++;
L
Linus Torvalds 已提交
742 743
		goto tx_error_icmp;
	}
744
	tdev = rt->dst.dev;
L
Linus Torvalds 已提交
745 746 747

	if (tdev == dev) {
		ip_rt_put(rt);
E
Eric Dumazet 已提交
748
		dev->stats.collisions++;
L
Linus Torvalds 已提交
749 750 751
		goto tx_error;
	}

752
	if (df) {
753
		mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
L
Linus Torvalds 已提交
754

755
		if (mtu < 68) {
E
Eric Dumazet 已提交
756
			dev->stats.collisions++;
757 758 759
			ip_rt_put(rt);
			goto tx_error;
		}
L
Linus Torvalds 已提交
760

761 762 763 764 765 766 767 768 769
		if (mtu < IPV6_MIN_MTU) {
			mtu = IPV6_MIN_MTU;
			df = 0;
		}

		if (tunnel->parms.iph.daddr && skb_dst(skb))
			skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);

		if (skb->len > mtu) {
770
			icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
771 772 773
			ip_rt_put(rt);
			goto tx_error;
		}
L
Linus Torvalds 已提交
774 775 776
	}

	if (tunnel->err_count > 0) {
777 778
		if (time_before(jiffies,
				tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
L
Linus Torvalds 已提交
779 780 781 782 783 784 785 786 787 788 789
			tunnel->err_count--;
			dst_link_failure(skb);
		} else
			tunnel->err_count = 0;
	}

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

790 791
	if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
	    (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
L
Linus Torvalds 已提交
792 793 794
		struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
		if (!new_skb) {
			ip_rt_put(rt);
E
Eric Dumazet 已提交
795
			dev->stats.tx_dropped++;
L
Linus Torvalds 已提交
796
			dev_kfree_skb(skb);
797
			return NETDEV_TX_OK;
L
Linus Torvalds 已提交
798 799 800 801 802
		}
		if (skb->sk)
			skb_set_owner_w(new_skb, skb->sk);
		dev_kfree_skb(skb);
		skb = new_skb;
803
		iph6 = ipv6_hdr(skb);
L
Linus Torvalds 已提交
804 805
	}

806
	skb->transport_header = skb->network_header;
807 808
	skb_push(skb, sizeof(struct iphdr));
	skb_reset_network_header(skb);
L
Linus Torvalds 已提交
809
	memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
810
	IPCB(skb)->flags = 0;
E
Eric Dumazet 已提交
811
	skb_dst_drop(skb);
812
	skb_dst_set(skb, &rt->dst);
L
Linus Torvalds 已提交
813 814 815 816 817

	/*
	 *	Push down and install the IPIP header.
	 */

818
	iph 			=	ip_hdr(skb);
L
Linus Torvalds 已提交
819 820
	iph->version		=	4;
	iph->ihl		=	sizeof(struct iphdr)>>2;
821
	iph->frag_off		=	df;
L
Linus Torvalds 已提交
822 823
	iph->protocol		=	IPPROTO_IPV6;
	iph->tos		=	INET_ECN_encapsulate(tos, ipv6_get_dsfield(iph6));
824 825
	iph->daddr		=	fl4.daddr;
	iph->saddr		=	fl4.saddr;
L
Linus Torvalds 已提交
826 827 828 829 830

	if ((iph->ttl = tiph->ttl) == 0)
		iph->ttl	=	iph6->hop_limit;

	nf_reset(skb);
E
Eric Dumazet 已提交
831 832
	tstats = this_cpu_ptr(dev->tstats);
	__IPTUNNEL_XMIT(tstats, &dev->stats);
833
	return NETDEV_TX_OK;
L
Linus Torvalds 已提交
834 835 836 837

tx_error_icmp:
	dst_link_failure(skb);
tx_error:
E
Eric Dumazet 已提交
838
	dev->stats.tx_errors++;
L
Linus Torvalds 已提交
839
	dev_kfree_skb(skb);
840
	return NETDEV_TX_OK;
L
Linus Torvalds 已提交
841 842
}

843 844 845 846
static void ipip6_tunnel_bind_dev(struct net_device *dev)
{
	struct net_device *tdev = NULL;
	struct ip_tunnel *tunnel;
847
	const struct iphdr *iph;
848
	struct flowi4 fl4;
849 850 851 852 853

	tunnel = netdev_priv(dev);
	iph = &tunnel->parms.iph;

	if (iph->daddr) {
854
		struct rtable *rt = ip_route_output_ports(dev_net(dev), &fl4, NULL,
855 856 857 858 859
							  iph->daddr, iph->saddr,
							  0, 0,
							  IPPROTO_IPV6,
							  RT_TOS(iph->tos),
							  tunnel->parms.link);
860 861

		if (!IS_ERR(rt)) {
862
			tdev = rt->dst.dev;
863 864 865 866 867 868
			ip_rt_put(rt);
		}
		dev->flags |= IFF_POINTOPOINT;
	}

	if (!tdev && tunnel->parms.link)
869
		tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
870 871 872 873 874 875 876 877 878 879

	if (tdev) {
		dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
		dev->mtu = tdev->mtu - sizeof(struct iphdr);
		if (dev->mtu < IPV6_MIN_MTU)
			dev->mtu = IPV6_MIN_MTU;
	}
	dev->iflink = tunnel->parms.link;
}

L
Linus Torvalds 已提交
880 881 882 883 884
static int
ipip6_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
{
	int err = 0;
	struct ip_tunnel_parm p;
885
	struct ip_tunnel_prl prl;
L
Linus Torvalds 已提交
886
	struct ip_tunnel *t;
887 888
	struct net *net = dev_net(dev);
	struct sit_net *sitn = net_generic(net, sit_net_id);
889 890 891
#ifdef CONFIG_IPV6_SIT_6RD
	struct ip_tunnel_6rd ip6rd;
#endif
L
Linus Torvalds 已提交
892 893 894

	switch (cmd) {
	case SIOCGETTUNNEL:
895 896 897
#ifdef CONFIG_IPV6_SIT_6RD
	case SIOCGET6RD:
#endif
L
Linus Torvalds 已提交
898
		t = NULL;
899
		if (dev == sitn->fb_tunnel_dev) {
L
Linus Torvalds 已提交
900 901 902 903
			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
				err = -EFAULT;
				break;
			}
904
			t = ipip6_tunnel_locate(net, &p, 0);
L
Linus Torvalds 已提交
905 906
		}
		if (t == NULL)
907
			t = netdev_priv(dev);
908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926

		err = -EFAULT;
		if (cmd == SIOCGETTUNNEL) {
			memcpy(&p, &t->parms, sizeof(p));
			if (copy_to_user(ifr->ifr_ifru.ifru_data, &p,
					 sizeof(p)))
				goto done;
#ifdef CONFIG_IPV6_SIT_6RD
		} else {
			ipv6_addr_copy(&ip6rd.prefix, &t->ip6rd.prefix);
			ip6rd.relay_prefix = t->ip6rd.relay_prefix;
			ip6rd.prefixlen = t->ip6rd.prefixlen;
			ip6rd.relay_prefixlen = t->ip6rd.relay_prefixlen;
			if (copy_to_user(ifr->ifr_ifru.ifru_data, &ip6rd,
					 sizeof(ip6rd)))
				goto done;
#endif
		}
		err = 0;
L
Linus Torvalds 已提交
927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945
		break;

	case SIOCADDTUNNEL:
	case SIOCCHGTUNNEL:
		err = -EPERM;
		if (!capable(CAP_NET_ADMIN))
			goto done;

		err = -EFAULT;
		if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
			goto done;

		err = -EINVAL;
		if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPV6 ||
		    p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
			goto done;
		if (p.iph.ttl)
			p.iph.frag_off |= htons(IP_DF);

946
		t = ipip6_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
L
Linus Torvalds 已提交
947

948
		if (dev != sitn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
L
Linus Torvalds 已提交
949 950 951 952 953 954 955 956 957 958 959
			if (t != NULL) {
				if (t->dev != dev) {
					err = -EEXIST;
					break;
				}
			} else {
				if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
				    (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
					err = -EINVAL;
					break;
				}
960
				t = netdev_priv(dev);
961
				ipip6_tunnel_unlink(sitn, t);
962
				synchronize_net();
L
Linus Torvalds 已提交
963 964 965 966
				t->parms.iph.saddr = p.iph.saddr;
				t->parms.iph.daddr = p.iph.daddr;
				memcpy(dev->dev_addr, &p.iph.saddr, 4);
				memcpy(dev->broadcast, &p.iph.daddr, 4);
967
				ipip6_tunnel_link(sitn, t);
L
Linus Torvalds 已提交
968 969 970 971 972 973 974 975 976
				netdev_state_change(dev);
			}
		}

		if (t) {
			err = 0;
			if (cmd == SIOCCHGTUNNEL) {
				t->parms.iph.ttl = p.iph.ttl;
				t->parms.iph.tos = p.iph.tos;
977 978 979 980 981
				if (t->parms.link != p.link) {
					t->parms.link = p.link;
					ipip6_tunnel_bind_dev(dev);
					netdev_state_change(dev);
				}
L
Linus Torvalds 已提交
982 983 984 985 986 987 988 989 990 991 992 993
			}
			if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
				err = -EFAULT;
		} else
			err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
		break;

	case SIOCDELTUNNEL:
		err = -EPERM;
		if (!capable(CAP_NET_ADMIN))
			goto done;

994
		if (dev == sitn->fb_tunnel_dev) {
L
Linus Torvalds 已提交
995 996 997 998
			err = -EFAULT;
			if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
				goto done;
			err = -ENOENT;
999
			if ((t = ipip6_tunnel_locate(net, &p, 0)) == NULL)
L
Linus Torvalds 已提交
1000 1001
				goto done;
			err = -EPERM;
1002
			if (t == netdev_priv(sitn->fb_tunnel_dev))
L
Linus Torvalds 已提交
1003 1004 1005
				goto done;
			dev = t->dev;
		}
1006 1007
		unregister_netdevice(dev);
		err = 0;
L
Linus Torvalds 已提交
1008 1009
		break;

1010
	case SIOCGETPRL:
1011 1012 1013 1014 1015 1016 1017 1018 1019
		err = -EINVAL;
		if (dev == sitn->fb_tunnel_dev)
			goto done;
		err = -ENOENT;
		if (!(t = netdev_priv(dev)))
			goto done;
		err = ipip6_tunnel_get_prl(t, ifr->ifr_ifru.ifru_data);
		break;

1020 1021 1022 1023
	case SIOCADDPRL:
	case SIOCDELPRL:
	case SIOCCHGPRL:
		err = -EPERM;
1024
		if (!capable(CAP_NET_ADMIN))
1025 1026
			goto done;
		err = -EINVAL;
1027
		if (dev == sitn->fb_tunnel_dev)
1028 1029 1030 1031 1032 1033 1034 1035
			goto done;
		err = -EFAULT;
		if (copy_from_user(&prl, ifr->ifr_ifru.ifru_data, sizeof(prl)))
			goto done;
		err = -ENOENT;
		if (!(t = netdev_priv(dev)))
			goto done;

1036 1037
		switch (cmd) {
		case SIOCDELPRL:
1038
			err = ipip6_tunnel_del_prl(t, &prl);
1039 1040 1041
			break;
		case SIOCADDPRL:
		case SIOCCHGPRL:
1042
			err = ipip6_tunnel_add_prl(t, &prl, cmd == SIOCCHGPRL);
1043 1044
			break;
		}
1045
		netdev_state_change(dev);
1046 1047
		break;

1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075
#ifdef CONFIG_IPV6_SIT_6RD
	case SIOCADD6RD:
	case SIOCCHG6RD:
	case SIOCDEL6RD:
		err = -EPERM;
		if (!capable(CAP_NET_ADMIN))
			goto done;

		err = -EFAULT;
		if (copy_from_user(&ip6rd, ifr->ifr_ifru.ifru_data,
				   sizeof(ip6rd)))
			goto done;

		t = netdev_priv(dev);

		if (cmd != SIOCDEL6RD) {
			struct in6_addr prefix;
			__be32 relay_prefix;

			err = -EINVAL;
			if (ip6rd.relay_prefixlen > 32 ||
			    ip6rd.prefixlen + (32 - ip6rd.relay_prefixlen) > 64)
				goto done;

			ipv6_addr_prefix(&prefix, &ip6rd.prefix,
					 ip6rd.prefixlen);
			if (!ipv6_addr_equal(&prefix, &ip6rd.prefix))
				goto done;
1076 1077 1078 1079 1080 1081
			if (ip6rd.relay_prefixlen)
				relay_prefix = ip6rd.relay_prefix &
					       htonl(0xffffffffUL <<
						     (32 - ip6rd.relay_prefixlen));
			else
				relay_prefix = 0;
1082 1083 1084 1085 1086 1087 1088 1089
			if (relay_prefix != ip6rd.relay_prefix)
				goto done;

			ipv6_addr_copy(&t->ip6rd.prefix, &prefix);
			t->ip6rd.relay_prefix = relay_prefix;
			t->ip6rd.prefixlen = ip6rd.prefixlen;
			t->ip6rd.relay_prefixlen = ip6rd.relay_prefixlen;
		} else
1090
			ipip6_tunnel_clone_6rd(dev, sitn);
1091 1092 1093 1094 1095

		err = 0;
		break;
#endif

L
Linus Torvalds 已提交
1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111
	default:
		err = -EINVAL;
	}

done:
	return err;
}

static int ipip6_tunnel_change_mtu(struct net_device *dev, int new_mtu)
{
	if (new_mtu < IPV6_MIN_MTU || new_mtu > 0xFFF8 - sizeof(struct iphdr))
		return -EINVAL;
	dev->mtu = new_mtu;
	return 0;
}

1112 1113 1114 1115 1116
static const struct net_device_ops ipip6_netdev_ops = {
	.ndo_uninit	= ipip6_tunnel_uninit,
	.ndo_start_xmit	= ipip6_tunnel_xmit,
	.ndo_do_ioctl	= ipip6_tunnel_ioctl,
	.ndo_change_mtu	= ipip6_tunnel_change_mtu,
E
Eric Dumazet 已提交
1117
	.ndo_get_stats	= ipip6_get_stats,
1118 1119
};

E
Eric Dumazet 已提交
1120 1121 1122 1123 1124 1125
static void ipip6_dev_free(struct net_device *dev)
{
	free_percpu(dev->tstats);
	free_netdev(dev);
}

L
Linus Torvalds 已提交
1126 1127
static void ipip6_tunnel_setup(struct net_device *dev)
{
1128
	dev->netdev_ops		= &ipip6_netdev_ops;
E
Eric Dumazet 已提交
1129
	dev->destructor 	= ipip6_dev_free;
L
Linus Torvalds 已提交
1130 1131 1132

	dev->type		= ARPHRD_SIT;
	dev->hard_header_len 	= LL_MAX_HEADER + sizeof(struct iphdr);
1133
	dev->mtu		= ETH_DATA_LEN - sizeof(struct iphdr);
L
Linus Torvalds 已提交
1134
	dev->flags		= IFF_NOARP;
1135
	dev->priv_flags	       &= ~IFF_XMIT_DST_RELEASE;
L
Linus Torvalds 已提交
1136 1137
	dev->iflink		= 0;
	dev->addr_len		= 4;
1138
	dev->features		|= NETIF_F_NETNS_LOCAL;
E
Eric Dumazet 已提交
1139
	dev->features		|= NETIF_F_LLTX;
L
Linus Torvalds 已提交
1140 1141
}

E
Eric Dumazet 已提交
1142
static int ipip6_tunnel_init(struct net_device *dev)
L
Linus Torvalds 已提交
1143
{
1144
	struct ip_tunnel *tunnel = netdev_priv(dev);
L
Linus Torvalds 已提交
1145 1146 1147 1148 1149 1150 1151

	tunnel->dev = dev;
	strcpy(tunnel->parms.name, dev->name);

	memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
	memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);

1152
	ipip6_tunnel_bind_dev(dev);
E
Eric Dumazet 已提交
1153 1154 1155 1156 1157
	dev->tstats = alloc_percpu(struct pcpu_tstats);
	if (!dev->tstats)
		return -ENOMEM;

	return 0;
L
Linus Torvalds 已提交
1158 1159
}

E
Eric Dumazet 已提交
1160
static int __net_init ipip6_fb_tunnel_init(struct net_device *dev)
L
Linus Torvalds 已提交
1161
{
1162
	struct ip_tunnel *tunnel = netdev_priv(dev);
L
Linus Torvalds 已提交
1163
	struct iphdr *iph = &tunnel->parms.iph;
1164 1165
	struct net *net = dev_net(dev);
	struct sit_net *sitn = net_generic(net, sit_net_id);
L
Linus Torvalds 已提交
1166 1167 1168 1169 1170 1171 1172 1173 1174

	tunnel->dev = dev;
	strcpy(tunnel->parms.name, dev->name);

	iph->version		= 4;
	iph->protocol		= IPPROTO_IPV6;
	iph->ihl		= 5;
	iph->ttl		= 64;

E
Eric Dumazet 已提交
1175 1176 1177
	dev->tstats = alloc_percpu(struct pcpu_tstats);
	if (!dev->tstats)
		return -ENOMEM;
L
Linus Torvalds 已提交
1178
	dev_hold(dev);
1179
	RCU_INIT_POINTER(sitn->tunnels_wc[0], tunnel);
E
Eric Dumazet 已提交
1180
	return 0;
L
Linus Torvalds 已提交
1181 1182
}

1183
static struct xfrm_tunnel sit_handler __read_mostly = {
L
Linus Torvalds 已提交
1184 1185
	.handler	=	ipip6_rcv,
	.err_handler	=	ipip6_err,
1186
	.priority	=	1,
L
Linus Torvalds 已提交
1187 1188
};

1189
static void __net_exit sit_destroy_tunnels(struct sit_net *sitn, struct list_head *head)
1190 1191 1192 1193 1194 1195
{
	int prio;

	for (prio = 1; prio < 4; prio++) {
		int h;
		for (h = 0; h < HASH_SIZE; h++) {
1196
			struct ip_tunnel *t;
1197

1198
			t = rtnl_dereference(sitn->tunnels[prio][h]);
1199 1200
			while (t != NULL) {
				unregister_netdevice_queue(t->dev, head);
1201
				t = rtnl_dereference(t->next);
1202
			}
1203 1204 1205 1206
		}
	}
}

1207
static int __net_init sit_init_net(struct net *net)
1208
{
1209
	struct sit_net *sitn = net_generic(net, sit_net_id);
1210 1211
	int err;

1212 1213 1214 1215 1216
	sitn->tunnels[0] = sitn->tunnels_wc;
	sitn->tunnels[1] = sitn->tunnels_l;
	sitn->tunnels[2] = sitn->tunnels_r;
	sitn->tunnels[3] = sitn->tunnels_r_l;

1217 1218 1219 1220 1221 1222
	sitn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "sit0",
					   ipip6_tunnel_setup);
	if (!sitn->fb_tunnel_dev) {
		err = -ENOMEM;
		goto err_alloc_dev;
	}
1223
	dev_net_set(sitn->fb_tunnel_dev, net);
1224

E
Eric Dumazet 已提交
1225 1226 1227 1228
	err = ipip6_fb_tunnel_init(sitn->fb_tunnel_dev);
	if (err)
		goto err_dev_free;

1229
	ipip6_tunnel_clone_6rd(sitn->fb_tunnel_dev, sitn);
1230 1231 1232 1233

	if ((err = register_netdev(sitn->fb_tunnel_dev)))
		goto err_reg_dev;

1234 1235
	return 0;

1236
err_reg_dev:
1237
	dev_put(sitn->fb_tunnel_dev);
E
Eric Dumazet 已提交
1238 1239
err_dev_free:
	ipip6_dev_free(sitn->fb_tunnel_dev);
1240
err_alloc_dev:
1241 1242 1243
	return err;
}

1244
static void __net_exit sit_exit_net(struct net *net)
1245
{
1246
	struct sit_net *sitn = net_generic(net, sit_net_id);
1247
	LIST_HEAD(list);
1248

1249
	rtnl_lock();
1250 1251 1252
	sit_destroy_tunnels(sitn, &list);
	unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
	unregister_netdevice_many(&list);
1253
	rtnl_unlock();
1254 1255 1256 1257 1258
}

static struct pernet_operations sit_net_ops = {
	.init = sit_init_net,
	.exit = sit_exit_net,
1259 1260
	.id   = &sit_net_id,
	.size = sizeof(struct sit_net),
1261 1262
};

1263
static void __exit sit_cleanup(void)
L
Linus Torvalds 已提交
1264
{
1265
	xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1266

1267
	unregister_pernet_device(&sit_net_ops);
E
Eric Dumazet 已提交
1268
	rcu_barrier(); /* Wait for completion of call_rcu()'s */
L
Linus Torvalds 已提交
1269 1270
}

1271
static int __init sit_init(void)
L
Linus Torvalds 已提交
1272 1273 1274 1275 1276
{
	int err;

	printk(KERN_INFO "IPv6 over IPv4 tunneling driver\n");

1277
	err = register_pernet_device(&sit_net_ops);
1278
	if (err < 0)
1279 1280 1281 1282 1283 1284
		return err;
	err = xfrm4_tunnel_register(&sit_handler, AF_INET6);
	if (err < 0) {
		unregister_pernet_device(&sit_net_ops);
		printk(KERN_INFO "sit init: Can't add protocol\n");
	}
L
Linus Torvalds 已提交
1285 1286
	return err;
}
1287 1288 1289

module_init(sit_init);
module_exit(sit_cleanup);
1290
MODULE_LICENSE("GPL");
1291
MODULE_ALIAS_NETDEV("sit0");