af_inet6.c 24.6 KB
Newer Older
L
Linus Torvalds 已提交
1 2
/*
 *	PF_INET6 socket protocol family
3
 *	Linux INET6 implementation
L
Linus Torvalds 已提交
4 5
 *
 *	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 17 18 19 20
 *
 *	Adapted from linux/net/ipv4/af_inet.c
 *
 * 	Fixes:
 *	piggy, Karl Knutson	:	Socket protocol table
 * 	Hideaki YOSHIFUJI	:	sin6_scope_id support
 * 	Arnaldo Melo		: 	check proc_net_create return, cleanups
 *
 *	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.
 */

21
#define pr_fmt(fmt) "IPv6: " fmt
L
Linus Torvalds 已提交
22 23

#include <linux/module.h>
24
#include <linux/capability.h>
L
Linus Torvalds 已提交
25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
#include <linux/errno.h>
#include <linux/types.h>
#include <linux/socket.h>
#include <linux/in.h>
#include <linux/kernel.h>
#include <linux/timer.h>
#include <linux/string.h>
#include <linux/sockios.h>
#include <linux/net.h>
#include <linux/fcntl.h>
#include <linux/mm.h>
#include <linux/interrupt.h>
#include <linux/proc_fs.h>
#include <linux/stat.h>
#include <linux/init.h>
40
#include <linux/slab.h>
L
Linus Torvalds 已提交
41 42 43 44

#include <linux/inet.h>
#include <linux/netdevice.h>
#include <linux/icmpv6.h>
45
#include <linux/netfilter_ipv6.h>
L
Linus Torvalds 已提交
46 47 48 49

#include <net/ip.h>
#include <net/ipv6.h>
#include <net/udp.h>
50
#include <net/udplite.h>
L
Linus Torvalds 已提交
51 52 53 54
#include <net/tcp.h>
#include <net/ipip.h>
#include <net/protocol.h>
#include <net/inet_common.h>
55
#include <net/route.h>
L
Linus Torvalds 已提交
56 57 58 59 60 61 62 63
#include <net/transp_v6.h>
#include <net/ip6_route.h>
#include <net/addrconf.h>
#ifdef CONFIG_IPV6_TUNNEL
#include <net/ip6_tunnel.h>
#endif

#include <asm/uaccess.h>
64
#include <linux/mroute6.h>
L
Linus Torvalds 已提交
65 66 67 68 69

MODULE_AUTHOR("Cast of dozens");
MODULE_DESCRIPTION("IPv6 protocol stack for Linux");
MODULE_LICENSE("GPL");

70
/* The inetsw6 table contains everything that inet6_create needs to
L
Linus Torvalds 已提交
71 72 73 74 75
 * build a new socket.
 */
static struct list_head inetsw6[SOCK_MAX];
static DEFINE_SPINLOCK(inetsw6_lock);

76 77 78 79 80
struct ipv6_params ipv6_defaults = {
	.disable_ipv6 = 0,
	.autoconf = 1,
};

81
static int disable_ipv6_mod;
82 83 84 85 86 87 88 89 90

module_param_named(disable, disable_ipv6_mod, int, 0444);
MODULE_PARM_DESC(disable, "Disable IPv6 module such that it is non-functional");

module_param_named(disable_ipv6, ipv6_defaults.disable_ipv6, int, 0444);
MODULE_PARM_DESC(disable_ipv6, "Disable IPv6 on all interfaces");

module_param_named(autoconf, ipv6_defaults.autoconf, int, 0444);
MODULE_PARM_DESC(autoconf, "Enable IPv6 address autoconfiguration on all interfaces");
91

L
Linus Torvalds 已提交
92 93 94 95 96 97 98
static __inline__ struct ipv6_pinfo *inet6_sk_generic(struct sock *sk)
{
	const int offset = sk->sk_prot->obj_size - sizeof(struct ipv6_pinfo);

	return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
}

99 100
static int inet6_create(struct net *net, struct socket *sock, int protocol,
			int kern)
L
Linus Torvalds 已提交
101 102 103 104 105 106 107 108
{
	struct inet_sock *inet;
	struct ipv6_pinfo *np;
	struct sock *sk;
	struct inet_protosw *answer;
	struct proto *answer_prot;
	unsigned char answer_flags;
	char answer_no_check;
109 110
	int try_loading_module = 0;
	int err;
L
Linus Torvalds 已提交
111

112 113 114 115 116
	if (sock->type != SOCK_RAW &&
	    sock->type != SOCK_DGRAM &&
	    !inet_ehash_secret)
		build_ehash_secret();

L
Linus Torvalds 已提交
117
	/* Look for the requested type/protocol pair. */
118 119
lookup_protocol:
	err = -ESOCKTNOSUPPORT;
L
Linus Torvalds 已提交
120
	rcu_read_lock();
121
	list_for_each_entry_rcu(answer, &inetsw6[sock->type], list) {
L
Linus Torvalds 已提交
122

123
		err = 0;
L
Linus Torvalds 已提交
124 125 126 127 128 129 130 131 132 133 134 135 136
		/* Check the non-wild match. */
		if (protocol == answer->protocol) {
			if (protocol != IPPROTO_IP)
				break;
		} else {
			/* Check for the two wild cases. */
			if (IPPROTO_IP == protocol) {
				protocol = answer->protocol;
				break;
			}
			if (IPPROTO_IP == answer->protocol)
				break;
		}
137
		err = -EPROTONOSUPPORT;
L
Linus Torvalds 已提交
138 139
	}

140
	if (err) {
141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162
		if (try_loading_module < 2) {
			rcu_read_unlock();
			/*
			 * Be more specific, e.g. net-pf-10-proto-132-type-1
			 * (net-pf-PF_INET6-proto-IPPROTO_SCTP-type-SOCK_STREAM)
			 */
			if (++try_loading_module == 1)
				request_module("net-pf-%d-proto-%d-type-%d",
						PF_INET6, protocol, sock->type);
			/*
			 * Fall back to generic, e.g. net-pf-10-proto-132
			 * (net-pf-PF_INET6-proto-IPPROTO_SCTP)
			 */
			else
				request_module("net-pf-%d-proto-%d",
						PF_INET6, protocol);
			goto lookup_protocol;
		} else
			goto out_rcu_unlock;
	}

	err = -EPERM;
163 164
	if (sock->type == SOCK_RAW && !kern &&
	    !ns_capable(net->user_ns, CAP_NET_RAW))
L
Linus Torvalds 已提交
165 166 167 168 169 170 171 172
		goto out_rcu_unlock;

	sock->ops = answer->ops;
	answer_prot = answer->prot;
	answer_no_check = answer->no_check;
	answer_flags = answer->flags;
	rcu_read_unlock();

173
	WARN_ON(answer_prot->slab == NULL);
L
Linus Torvalds 已提交
174

175
	err = -ENOBUFS;
176
	sk = sk_alloc(net, PF_INET6, GFP_KERNEL, answer_prot);
L
Linus Torvalds 已提交
177 178 179 180 181
	if (sk == NULL)
		goto out;

	sock_init_data(sock, sk);

182
	err = 0;
L
Linus Torvalds 已提交
183 184
	sk->sk_no_check = answer_no_check;
	if (INET_PROTOSW_REUSE & answer_flags)
185
		sk->sk_reuse = SK_CAN_REUSE;
L
Linus Torvalds 已提交
186 187

	inet = inet_sk(sk);
188
	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
L
Linus Torvalds 已提交
189 190

	if (SOCK_RAW == sock->type) {
E
Eric Dumazet 已提交
191
		inet->inet_num = protocol;
L
Linus Torvalds 已提交
192 193 194 195
		if (IPPROTO_RAW == protocol)
			inet->hdrincl = 1;
	}

196
	sk->sk_destruct		= inet_sock_destruct;
L
Linus Torvalds 已提交
197 198 199 200 201 202 203
	sk->sk_family		= PF_INET6;
	sk->sk_protocol		= protocol;

	sk->sk_backlog_rcv	= answer->prot->backlog_rcv;

	inet_sk(sk)->pinet6 = np = inet6_sk_generic(sk);
	np->hop_limit	= -1;
204
	np->mcast_hops	= IPV6_DEFAULT_MCASTHOPS;
L
Linus Torvalds 已提交
205 206
	np->mc_loop	= 1;
	np->pmtudisc	= IPV6_PMTUDISC_WANT;
207
	np->ipv6only	= net->ipv6.sysctl.bindv6only;
208

L
Linus Torvalds 已提交
209 210 211 212 213 214 215 216 217
	/* Init the ipv4 part of the socket since we can have sockets
	 * using v6 API for ipv4.
	 */
	inet->uc_ttl	= -1;

	inet->mc_loop	= 1;
	inet->mc_ttl	= 1;
	inet->mc_index	= 0;
	inet->mc_list	= NULL;
218
	inet->rcv_tos	= 0;
L
Linus Torvalds 已提交
219 220 221 222 223

	if (ipv4_config.no_pmtu_disc)
		inet->pmtudisc = IP_PMTUDISC_DONT;
	else
		inet->pmtudisc = IP_PMTUDISC_WANT;
224
	/*
225 226 227 228 229 230 231 232 233
	 * Increment only the relevant sk_prot->socks debug field, this changes
	 * the previous behaviour of incrementing both the equivalent to
	 * answer->prot->socks (inet6_sock_nr) and inet_sock_nr.
	 *
	 * This allows better debug granularity as we'll know exactly how many
	 * UDPv6, TCPv6, etc socks were allocated, not the sum of all IPv6
	 * transport protocol socks. -acme
	 */
	sk_refcnt_debug_inc(sk);
L
Linus Torvalds 已提交
234

E
Eric Dumazet 已提交
235
	if (inet->inet_num) {
L
Linus Torvalds 已提交
236 237 238 239
		/* It assumes that any protocol which allows
		 * the user to assign a number at socket
		 * creation time automatically shares.
		 */
E
Eric Dumazet 已提交
240
		inet->inet_sport = htons(inet->inet_num);
L
Linus Torvalds 已提交
241 242 243
		sk->sk_prot->hash(sk);
	}
	if (sk->sk_prot->init) {
244 245
		err = sk->sk_prot->init(sk);
		if (err) {
L
Linus Torvalds 已提交
246 247 248 249 250
			sk_common_release(sk);
			goto out;
		}
	}
out:
251
	return err;
L
Linus Torvalds 已提交
252 253 254 255 256 257 258 259 260
out_rcu_unlock:
	rcu_read_unlock();
	goto out;
}


/* bind for INET6 API */
int inet6_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
{
261
	struct sockaddr_in6 *addr = (struct sockaddr_in6 *)uaddr;
L
Linus Torvalds 已提交
262 263 264
	struct sock *sk = sock->sk;
	struct inet_sock *inet = inet_sk(sk);
	struct ipv6_pinfo *np = inet6_sk(sk);
265
	struct net *net = sock_net(sk);
A
Al Viro 已提交
266
	__be32 v4addr = 0;
L
Linus Torvalds 已提交
267 268 269 270 271 272 273 274 275 276
	unsigned short snum;
	int addr_type = 0;
	int err = 0;

	/* If the socket has its own bind function then use it. */
	if (sk->sk_prot->bind)
		return sk->sk_prot->bind(sk, uaddr, addr_len);

	if (addr_len < SIN6_LEN_RFC2133)
		return -EINVAL;
277 278

	if (addr->sin6_family != AF_INET6)
279
		return -EAFNOSUPPORT;
280

L
Linus Torvalds 已提交
281 282 283 284 285 286 287 288 289 290 291
	addr_type = ipv6_addr_type(&addr->sin6_addr);
	if ((addr_type & IPV6_ADDR_MULTICAST) && sock->type == SOCK_STREAM)
		return -EINVAL;

	snum = ntohs(addr->sin6_port);
	if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
		return -EACCES;

	lock_sock(sk);

	/* Check these errors (active socket, double bind). */
E
Eric Dumazet 已提交
292
	if (sk->sk_state != TCP_CLOSE || inet->inet_num) {
L
Linus Torvalds 已提交
293 294 295 296 297 298
		err = -EINVAL;
		goto out;
	}

	/* Check if the address belongs to the host. */
	if (addr_type == IPV6_ADDR_MAPPED) {
299 300
		int chk_addr_ret;

301 302 303 304 305 306 307
		/* Binding to v4-mapped address on a v6-only socket
		 * makes no sense
		 */
		if (np->ipv6only) {
			err = -EINVAL;
			goto out;
		}
308

309
		/* Reproduce AF_INET checks to make the bindings consistent */
L
Linus Torvalds 已提交
310
		v4addr = addr->sin6_addr.s6_addr32[3];
311 312 313 314 315 316
		chk_addr_ret = inet_addr_type(net, v4addr);
		if (!sysctl_ip_nonlocal_bind &&
		    !(inet->freebind || inet->transparent) &&
		    v4addr != htonl(INADDR_ANY) &&
		    chk_addr_ret != RTN_LOCAL &&
		    chk_addr_ret != RTN_MULTICAST &&
317 318
		    chk_addr_ret != RTN_BROADCAST) {
			err = -EADDRNOTAVAIL;
L
Linus Torvalds 已提交
319
			goto out;
320
		}
L
Linus Torvalds 已提交
321 322 323 324
	} else {
		if (addr_type != IPV6_ADDR_ANY) {
			struct net_device *dev = NULL;

325
			rcu_read_lock();
L
Linus Torvalds 已提交
326 327 328 329 330 331 332 333
			if (addr_type & IPV6_ADDR_LINKLOCAL) {
				if (addr_len >= sizeof(struct sockaddr_in6) &&
				    addr->sin6_scope_id) {
					/* Override any existing binding, if another one
					 * is supplied by user.
					 */
					sk->sk_bound_dev_if = addr->sin6_scope_id;
				}
334

L
Linus Torvalds 已提交
335 336 337
				/* Binding to link-local address requires an interface */
				if (!sk->sk_bound_dev_if) {
					err = -EINVAL;
338
					goto out_unlock;
L
Linus Torvalds 已提交
339
				}
340
				dev = dev_get_by_index_rcu(net, sk->sk_bound_dev_if);
L
Linus Torvalds 已提交
341 342
				if (!dev) {
					err = -ENODEV;
343
					goto out_unlock;
L
Linus Torvalds 已提交
344 345 346 347 348 349 350 351
				}
			}

			/* ipv4 addr of the socket is invalid.  Only the
			 * unspecified and mapped address have a v4 equivalent.
			 */
			v4addr = LOOPBACK4_IPV6;
			if (!(addr_type & IPV6_ADDR_MULTICAST))	{
352
				if (!(inet->freebind || inet->transparent) &&
353
				    !ipv6_chk_addr(net, &addr->sin6_addr,
354
						   dev, 0)) {
L
Linus Torvalds 已提交
355
					err = -EADDRNOTAVAIL;
356
					goto out_unlock;
L
Linus Torvalds 已提交
357 358
				}
			}
359
			rcu_read_unlock();
L
Linus Torvalds 已提交
360 361 362
		}
	}

E
Eric Dumazet 已提交
363 364
	inet->inet_rcv_saddr = v4addr;
	inet->inet_saddr = v4addr;
L
Linus Torvalds 已提交
365

A
Alexey Dobriyan 已提交
366
	np->rcv_saddr = addr->sin6_addr;
367

L
Linus Torvalds 已提交
368
	if (!(addr_type & IPV6_ADDR_MULTICAST))
A
Alexey Dobriyan 已提交
369
		np->saddr = addr->sin6_addr;
L
Linus Torvalds 已提交
370 371 372 373 374 375 376 377

	/* Make sure we are allowed to bind here. */
	if (sk->sk_prot->get_port(sk, snum)) {
		inet_reset_saddr(sk);
		err = -EADDRINUSE;
		goto out;
	}

378
	if (addr_type != IPV6_ADDR_ANY) {
L
Linus Torvalds 已提交
379
		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
380 381 382
		if (addr_type != IPV6_ADDR_MAPPED)
			np->ipv6only = 1;
	}
L
Linus Torvalds 已提交
383 384
	if (snum)
		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
E
Eric Dumazet 已提交
385 386 387
	inet->inet_sport = htons(inet->inet_num);
	inet->inet_dport = 0;
	inet->inet_daddr = 0;
L
Linus Torvalds 已提交
388 389 390
out:
	release_sock(sk);
	return err;
391 392 393
out_unlock:
	rcu_read_unlock();
	goto out;
L
Linus Torvalds 已提交
394
}
395 396
EXPORT_SYMBOL(inet6_bind);

L
Linus Torvalds 已提交
397 398 399 400 401 402 403 404 405 406 407 408 409 410 411
int inet6_release(struct socket *sock)
{
	struct sock *sk = sock->sk;

	if (sk == NULL)
		return -EINVAL;

	/* Free mc lists */
	ipv6_sock_mc_close(sk);

	/* Free ac lists */
	ipv6_sock_ac_close(sk);

	return inet_release(sock);
}
412 413
EXPORT_SYMBOL(inet6_release);

414
void inet6_destroy_sock(struct sock *sk)
L
Linus Torvalds 已提交
415 416 417 418 419 420 421
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct sk_buff *skb;
	struct ipv6_txoptions *opt;

	/* Release rx options */

422 423
	skb = xchg(&np->pktoptions, NULL);
	if (skb != NULL)
L
Linus Torvalds 已提交
424
		kfree_skb(skb);
425

426 427
	skb = xchg(&np->rxpmtu, NULL);
	if (skb != NULL)
428
		kfree_skb(skb);
L
Linus Torvalds 已提交
429 430 431 432 433 434

	/* Free flowlabels */
	fl6_free_socklist(sk);

	/* Free tx options */

435 436
	opt = xchg(&np->opt, NULL);
	if (opt != NULL)
L
Linus Torvalds 已提交
437 438
		sock_kfree_s(sk, opt, opt->tot_len);
}
439 440
EXPORT_SYMBOL_GPL(inet6_destroy_sock);

L
Linus Torvalds 已提交
441 442 443
/*
 *	This does both peername and sockname.
 */
444

L
Linus Torvalds 已提交
445 446 447
int inet6_getname(struct socket *sock, struct sockaddr *uaddr,
		 int *uaddr_len, int peer)
{
448
	struct sockaddr_in6 *sin = (struct sockaddr_in6 *)uaddr;
L
Linus Torvalds 已提交
449 450 451
	struct sock *sk = sock->sk;
	struct inet_sock *inet = inet_sk(sk);
	struct ipv6_pinfo *np = inet6_sk(sk);
452

L
Linus Torvalds 已提交
453 454 455 456
	sin->sin6_family = AF_INET6;
	sin->sin6_flowinfo = 0;
	sin->sin6_scope_id = 0;
	if (peer) {
E
Eric Dumazet 已提交
457
		if (!inet->inet_dport)
L
Linus Torvalds 已提交
458 459 460 461
			return -ENOTCONN;
		if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
		    peer == 1)
			return -ENOTCONN;
E
Eric Dumazet 已提交
462
		sin->sin6_port = inet->inet_dport;
A
Alexey Dobriyan 已提交
463
		sin->sin6_addr = np->daddr;
L
Linus Torvalds 已提交
464 465 466 467
		if (np->sndflow)
			sin->sin6_flowinfo = np->flow_label;
	} else {
		if (ipv6_addr_any(&np->rcv_saddr))
A
Alexey Dobriyan 已提交
468
			sin->sin6_addr = np->saddr;
L
Linus Torvalds 已提交
469
		else
A
Alexey Dobriyan 已提交
470
			sin->sin6_addr = np->rcv_saddr;
L
Linus Torvalds 已提交
471

E
Eric Dumazet 已提交
472
		sin->sin6_port = inet->inet_sport;
L
Linus Torvalds 已提交
473 474 475 476
	}
	if (ipv6_addr_type(&sin->sin6_addr) & IPV6_ADDR_LINKLOCAL)
		sin->sin6_scope_id = sk->sk_bound_dev_if;
	*uaddr_len = sizeof(*sin);
E
Eric Dumazet 已提交
477
	return 0;
L
Linus Torvalds 已提交
478
}
479 480
EXPORT_SYMBOL(inet6_getname);

L
Linus Torvalds 已提交
481 482 483
int inet6_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
{
	struct sock *sk = sock->sk;
484
	struct net *net = sock_net(sk);
L
Linus Torvalds 已提交
485

486
	switch (cmd) {
L
Linus Torvalds 已提交
487 488 489
	case SIOCGSTAMP:
		return sock_get_timestamp(sk, (struct timeval __user *)arg);

490 491 492
	case SIOCGSTAMPNS:
		return sock_get_timestampns(sk, (struct timespec __user *)arg);

L
Linus Torvalds 已提交
493 494
	case SIOCADDRT:
	case SIOCDELRT:
495

E
Eric Dumazet 已提交
496
		return ipv6_route_ioctl(net, cmd, (void __user *)arg);
L
Linus Torvalds 已提交
497 498

	case SIOCSIFADDR:
499
		return addrconf_add_ifaddr(net, (void __user *) arg);
L
Linus Torvalds 已提交
500
	case SIOCDIFADDR:
501
		return addrconf_del_ifaddr(net, (void __user *) arg);
L
Linus Torvalds 已提交
502
	case SIOCSIFDSTADDR:
503
		return addrconf_set_dstaddr(net, (void __user *) arg);
L
Linus Torvalds 已提交
504
	default:
505 506 507
		if (!sk->sk_prot->ioctl)
			return -ENOIOCTLCMD;
		return sk->sk_prot->ioctl(sk, cmd, arg);
L
Linus Torvalds 已提交
508 509
	}
	/*NOTREACHED*/
E
Eric Dumazet 已提交
510
	return 0;
L
Linus Torvalds 已提交
511
}
512 513
EXPORT_SYMBOL(inet6_ioctl);

514
const struct proto_ops inet6_stream_ops = {
515 516 517 518 519 520 521 522 523 524 525 526 527 528
	.family		   = PF_INET6,
	.owner		   = THIS_MODULE,
	.release	   = inet6_release,
	.bind		   = inet6_bind,
	.connect	   = inet_stream_connect,	/* ok		*/
	.socketpair	   = sock_no_socketpair,	/* a do nothing	*/
	.accept		   = inet_accept,		/* ok		*/
	.getname	   = inet6_getname,
	.poll		   = tcp_poll,			/* ok		*/
	.ioctl		   = inet6_ioctl,		/* must change  */
	.listen		   = inet_listen,		/* ok		*/
	.shutdown	   = inet_shutdown,		/* ok		*/
	.setsockopt	   = sock_common_setsockopt,	/* ok		*/
	.getsockopt	   = sock_common_getsockopt,	/* ok		*/
529 530
	.sendmsg	   = inet_sendmsg,		/* ok		*/
	.recvmsg	   = inet_recvmsg,		/* ok		*/
531
	.mmap		   = sock_no_mmap,
532
	.sendpage	   = inet_sendpage,
533
	.splice_read	   = tcp_splice_read,
534
#ifdef CONFIG_COMPAT
535 536
	.compat_setsockopt = compat_sock_common_setsockopt,
	.compat_getsockopt = compat_sock_common_getsockopt,
537
#endif
L
Linus Torvalds 已提交
538 539
};

540
const struct proto_ops inet6_dgram_ops = {
541 542 543 544 545 546 547 548 549 550 551 552 553 554 555
	.family		   = PF_INET6,
	.owner		   = THIS_MODULE,
	.release	   = inet6_release,
	.bind		   = inet6_bind,
	.connect	   = inet_dgram_connect,	/* ok		*/
	.socketpair	   = sock_no_socketpair,	/* a do nothing	*/
	.accept		   = sock_no_accept,		/* a do nothing	*/
	.getname	   = inet6_getname,
	.poll		   = udp_poll,			/* ok		*/
	.ioctl		   = inet6_ioctl,		/* must change  */
	.listen		   = sock_no_listen,		/* ok		*/
	.shutdown	   = inet_shutdown,		/* ok		*/
	.setsockopt	   = sock_common_setsockopt,	/* ok		*/
	.getsockopt	   = sock_common_getsockopt,	/* ok		*/
	.sendmsg	   = inet_sendmsg,		/* ok		*/
556
	.recvmsg	   = inet_recvmsg,		/* ok		*/
557 558
	.mmap		   = sock_no_mmap,
	.sendpage	   = sock_no_sendpage,
559
#ifdef CONFIG_COMPAT
560 561
	.compat_setsockopt = compat_sock_common_setsockopt,
	.compat_getsockopt = compat_sock_common_getsockopt,
562
#endif
L
Linus Torvalds 已提交
563 564
};

565
static const struct net_proto_family inet6_family_ops = {
L
Linus Torvalds 已提交
566 567 568 569 570
	.family = PF_INET6,
	.create = inet6_create,
	.owner	= THIS_MODULE,
};

571
int inet6_register_protosw(struct inet_protosw *p)
L
Linus Torvalds 已提交
572 573 574 575
{
	struct list_head *lh;
	struct inet_protosw *answer;
	struct list_head *last_perm;
576 577
	int protocol = p->protocol;
	int ret;
L
Linus Torvalds 已提交
578 579 580

	spin_lock_bh(&inetsw6_lock);

581
	ret = -EINVAL;
L
Linus Torvalds 已提交
582 583 584 585 586
	if (p->type >= SOCK_MAX)
		goto out_illegal;

	/* If we are trying to override a permanent protocol, bail. */
	answer = NULL;
587
	ret = -EPERM;
L
Linus Torvalds 已提交
588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606
	last_perm = &inetsw6[p->type];
	list_for_each(lh, &inetsw6[p->type]) {
		answer = list_entry(lh, struct inet_protosw, list);

		/* Check only the non-wild match. */
		if (INET_PROTOSW_PERMANENT & answer->flags) {
			if (protocol == answer->protocol)
				break;
			last_perm = lh;
		}

		answer = NULL;
	}
	if (answer)
		goto out_permanent;

	/* Add the new entry after the last permanent entry if any, so that
	 * the new entry does not override a permanent entry when matched with
	 * a wild-card protocol. But it is allowed to override any existing
607
	 * non-permanent entry.  This means that when we remove this entry, the
L
Linus Torvalds 已提交
608 609 610
	 * system automatically returns to the old behavior.
	 */
	list_add_rcu(&p->list, last_perm);
611
	ret = 0;
L
Linus Torvalds 已提交
612 613
out:
	spin_unlock_bh(&inetsw6_lock);
614
	return ret;
L
Linus Torvalds 已提交
615 616

out_permanent:
617
	pr_err("Attempt to override permanent protocol %d\n", protocol);
L
Linus Torvalds 已提交
618 619 620
	goto out;

out_illegal:
621
	pr_err("Ignoring attempt to register invalid socket type %d\n",
L
Linus Torvalds 已提交
622 623 624
	       p->type);
	goto out;
}
625 626
EXPORT_SYMBOL(inet6_register_protosw);

L
Linus Torvalds 已提交
627 628 629 630
void
inet6_unregister_protosw(struct inet_protosw *p)
{
	if (INET_PROTOSW_PERMANENT & p->flags) {
631
		pr_err("Attempt to unregister permanent protocol %d\n",
L
Linus Torvalds 已提交
632 633 634 635 636 637 638 639 640
		       p->protocol);
	} else {
		spin_lock_bh(&inetsw6_lock);
		list_del_rcu(&p->list);
		spin_unlock_bh(&inetsw6_lock);

		synchronize_net();
	}
}
641 642
EXPORT_SYMBOL(inet6_unregister_protosw);

643 644 645
int inet6_sk_rebuild_header(struct sock *sk)
{
	struct ipv6_pinfo *np = inet6_sk(sk);
646
	struct dst_entry *dst;
647 648 649 650 651

	dst = __sk_dst_check(sk, np->dst_cookie);

	if (dst == NULL) {
		struct inet_sock *inet = inet_sk(sk);
652
		struct in6_addr *final_p, final;
653 654 655 656
		struct flowi6 fl6;

		memset(&fl6, 0, sizeof(fl6));
		fl6.flowi6_proto = sk->sk_protocol;
A
Alexey Dobriyan 已提交
657 658
		fl6.daddr = np->daddr;
		fl6.saddr = np->saddr;
659 660 661
		fl6.flowlabel = np->flow_label;
		fl6.flowi6_oif = sk->sk_bound_dev_if;
		fl6.flowi6_mark = sk->sk_mark;
662 663
		fl6.fl6_dport = inet->inet_dport;
		fl6.fl6_sport = inet->inet_sport;
664 665 666 667 668
		security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));

		final_p = fl6_update_dst(&fl6, np->opt, &final);

		dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false);
669
		if (IS_ERR(dst)) {
670
			sk->sk_route_caps = 0;
671 672
			sk->sk_err_soft = -PTR_ERR(dst);
			return PTR_ERR(dst);
673 674
		}

675
		__ip6_dst_store(sk, dst, NULL, NULL);
676 677 678 679 680 681
	}

	return 0;
}
EXPORT_SYMBOL_GPL(inet6_sk_rebuild_header);

E
Eric Dumazet 已提交
682
bool ipv6_opt_accepted(const struct sock *sk, const struct sk_buff *skb)
683
{
E
Eric Dumazet 已提交
684 685
	const struct ipv6_pinfo *np = inet6_sk(sk);
	const struct inet6_skb_parm *opt = IP6CB(skb);
686 687 688 689

	if (np->rxopt.all) {
		if ((opt->hop && (np->rxopt.bits.hopopts ||
				  np->rxopt.bits.ohopopts)) ||
690 691
		    ((IPV6_FLOWINFO_MASK &
		      *(__be32 *)skb_network_header(skb)) &&
692 693 694 695 696
		     np->rxopt.bits.rxflow) ||
		    (opt->srcrt && (np->rxopt.bits.srcrt ||
		     np->rxopt.bits.osrcrt)) ||
		    ((opt->dst1 || opt->dst0) &&
		     (np->rxopt.bits.dstopts || np->rxopt.bits.odstopts)))
E
Eric Dumazet 已提交
697
			return true;
698
	}
E
Eric Dumazet 已提交
699
	return false;
700 701 702
}
EXPORT_SYMBOL_GPL(ipv6_opt_accepted);

703
static struct packet_type ipv6_packet_type __read_mostly = {
704
	.type = cpu_to_be16(ETH_P_IPV6),
705
	.func = ipv6_rcv,
706 707
};

708 709 710 711 712 713 714 715 716 717 718
static int __init ipv6_packet_init(void)
{
	dev_add_pack(&ipv6_packet_type);
	return 0;
}

static void ipv6_packet_cleanup(void)
{
	dev_remove_pack(&ipv6_packet_type);
}

719 720
static int __net_init ipv6_init_mibs(struct net *net)
{
721
	if (snmp_mib_init((void __percpu **)net->mib.udp_stats_in6,
722 723
			  sizeof(struct udp_mib),
			  __alignof__(struct udp_mib)) < 0)
724
		return -ENOMEM;
725
	if (snmp_mib_init((void __percpu **)net->mib.udplite_stats_in6,
726 727
			  sizeof(struct udp_mib),
			  __alignof__(struct udp_mib)) < 0)
728
		goto err_udplite_mib;
729
	if (snmp_mib_init((void __percpu **)net->mib.ipv6_statistics,
730 731
			  sizeof(struct ipstats_mib),
			  __alignof__(struct ipstats_mib)) < 0)
732
		goto err_ip_mib;
733
	if (snmp_mib_init((void __percpu **)net->mib.icmpv6_statistics,
734 735
			  sizeof(struct icmpv6_mib),
			  __alignof__(struct icmpv6_mib)) < 0)
736
		goto err_icmp_mib;
737 738 739
	net->mib.icmpv6msg_statistics = kzalloc(sizeof(struct icmpv6msg_mib),
						GFP_KERNEL);
	if (!net->mib.icmpv6msg_statistics)
740
		goto err_icmpmsg_mib;
741
	return 0;
742

743
err_icmpmsg_mib:
744
	snmp_mib_free((void __percpu **)net->mib.icmpv6_statistics);
745
err_icmp_mib:
746
	snmp_mib_free((void __percpu **)net->mib.ipv6_statistics);
747
err_ip_mib:
748
	snmp_mib_free((void __percpu **)net->mib.udplite_stats_in6);
749
err_udplite_mib:
750
	snmp_mib_free((void __percpu **)net->mib.udp_stats_in6);
751
	return -ENOMEM;
752 753
}

754
static void ipv6_cleanup_mibs(struct net *net)
755
{
756 757 758 759
	snmp_mib_free((void __percpu **)net->mib.udp_stats_in6);
	snmp_mib_free((void __percpu **)net->mib.udplite_stats_in6);
	snmp_mib_free((void __percpu **)net->mib.ipv6_statistics);
	snmp_mib_free((void __percpu **)net->mib.icmpv6_statistics);
760
	kfree(net->mib.icmpv6msg_statistics);
761 762
}

A
Alexey Dobriyan 已提交
763
static int __net_init inet6_net_init(struct net *net)
764
{
765 766
	int err = 0;

767
	net->ipv6.sysctl.bindv6only = 0;
768
	net->ipv6.sysctl.icmpv6_time = 1*HZ;
769

770 771 772
	err = ipv6_init_mibs(net);
	if (err)
		return err;
773 774 775 776
#ifdef CONFIG_PROC_FS
	err = udp6_proc_init(net);
	if (err)
		goto out;
777 778 779
	err = tcp6_proc_init(net);
	if (err)
		goto proc_tcp6_fail;
780 781 782
	err = ac6_proc_init(net);
	if (err)
		goto proc_ac6_fail;
783 784
#endif
	return err;
785 786

#ifdef CONFIG_PROC_FS
787 788
proc_ac6_fail:
	tcp6_proc_exit(net);
789 790
proc_tcp6_fail:
	udp6_proc_exit(net);
791 792 793
out:
	ipv6_cleanup_mibs(net);
	return err;
794
#endif
795 796
}

797
static void __net_exit inet6_net_exit(struct net *net)
798
{
799 800
#ifdef CONFIG_PROC_FS
	udp6_proc_exit(net);
801
	tcp6_proc_exit(net);
802
	ac6_proc_exit(net);
803
#endif
804
	ipv6_cleanup_mibs(net);
805 806 807 808 809 810 811
}

static struct pernet_operations inet6_net_ops = {
	.init = inet6_net_init,
	.exit = inet6_net_exit,
};

L
Linus Torvalds 已提交
812 813 814
static int __init inet6_init(void)
{
	struct sk_buff *dummy_skb;
815
	struct list_head *r;
816
	int err = 0;
L
Linus Torvalds 已提交
817

818 819
	BUILD_BUG_ON(sizeof(struct inet6_skb_parm) > sizeof(dummy_skb->cb));

820
	/* Register the socket-side information for inet6_create.  */
821
	for (r = &inetsw6[0]; r < &inetsw6[SOCK_MAX]; ++r)
822 823
		INIT_LIST_HEAD(r);

824
	if (disable_ipv6_mod) {
825
		pr_info("Loaded, but administratively disabled, reboot required to enable\n");
826 827 828
		goto out;
	}

L
Linus Torvalds 已提交
829 830 831 832 833 834 835 836
	err = proto_register(&tcpv6_prot, 1);
	if (err)
		goto out;

	err = proto_register(&udpv6_prot, 1);
	if (err)
		goto out_unregister_tcp_proto;

837
	err = proto_register(&udplitev6_prot, 1);
L
Linus Torvalds 已提交
838 839 840
	if (err)
		goto out_unregister_udp_proto;

841 842 843 844
	err = proto_register(&rawv6_prot, 1);
	if (err)
		goto out_unregister_udplite_proto;

L
Linus Torvalds 已提交
845 846 847 848

	/* We MUST register RAW sockets before we create the ICMP6,
	 * IGMP6, or NDISC control sockets.
	 */
849 850 851
	err = rawv6_init();
	if (err)
		goto out_unregister_raw_proto;
L
Linus Torvalds 已提交
852 853 854 855

	/* Register the family here so that the init calls below will
	 * be able to create sockets. (?? is this dangerous ??)
	 */
856 857
	err = sock_register(&inet6_family_ops);
	if (err)
858
		goto out_sock_register_fail;
L
Linus Torvalds 已提交
859

G
Glauber Costa 已提交
860 861
	tcpv6_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;

L
Linus Torvalds 已提交
862 863 864
	/*
	 *	ipngwg API draft makes clear that the correct semantics
	 *	for TCP and UDP is to consider one TCP and UDP instance
L
Lucas De Marchi 已提交
865
	 *	in a host available by both INET and INET6 APIs and
L
Linus Torvalds 已提交
866 867 868
	 *	able to communicate via both network protocols.
	 */

869 870 871
	err = register_pernet_subsys(&inet6_net_ops);
	if (err)
		goto register_pernet_fail;
872
	err = icmpv6_init();
L
Linus Torvalds 已提交
873 874
	if (err)
		goto icmp_fail;
W
Wang Chen 已提交
875 876 877
	err = ip6_mr_init();
	if (err)
		goto ipmr_fail;
878
	err = ndisc_init();
L
Linus Torvalds 已提交
879 880
	if (err)
		goto ndisc_fail;
881
	err = igmp6_init();
L
Linus Torvalds 已提交
882 883
	if (err)
		goto igmp_fail;
884 885 886
	err = ipv6_netfilter_init();
	if (err)
		goto netfilter_fail;
L
Linus Torvalds 已提交
887 888 889 890 891
	/* Create /proc/foo6 entries. */
#ifdef CONFIG_PROC_FS
	err = -ENOMEM;
	if (raw6_proc_init())
		goto proc_raw6_fail;
892 893
	if (udplite6_proc_init())
		goto proc_udplite6_fail;
L
Linus Torvalds 已提交
894 895 896 897 898
	if (ipv6_misc_proc_init())
		goto proc_misc6_fail;
	if (if6_proc_init())
		goto proc_if6_fail;
#endif
899 900 901
	err = ip6_route_init();
	if (err)
		goto ip6_route_fail;
902 903 904
	err = ip6_flowlabel_init();
	if (err)
		goto ip6_flowlabel_fail;
L
Linus Torvalds 已提交
905 906 907 908 909
	err = addrconf_init();
	if (err)
		goto addrconf_fail;

	/* Init v6 extension headers. */
910 911 912 913
	err = ipv6_exthdrs_init();
	if (err)
		goto ipv6_exthdrs_fail;

914 915 916
	err = ipv6_frag_init();
	if (err)
		goto ipv6_frag_fail;
L
Linus Torvalds 已提交
917 918

	/* Init v6 transport protocols. */
919 920 921
	err = udpv6_init();
	if (err)
		goto udpv6_fail;
922

923 924 925 926 927 928 929 930 931 932 933
	err = udplitev6_init();
	if (err)
		goto udplitev6_fail;

	err = tcpv6_init();
	if (err)
		goto tcpv6_fail;

	err = ipv6_packet_init();
	if (err)
		goto ipv6_packet_fail;
934 935 936 937 938 939

#ifdef CONFIG_SYSCTL
	err = ipv6_sysctl_register();
	if (err)
		goto sysctl_fail;
#endif
L
Linus Torvalds 已提交
940 941 942
out:
	return err;

943 944 945 946
#ifdef CONFIG_SYSCTL
sysctl_fail:
	ipv6_packet_cleanup();
#endif
947 948 949 950 951 952 953 954
ipv6_packet_fail:
	tcpv6_exit();
tcpv6_fail:
	udplitev6_exit();
udplitev6_fail:
	udpv6_exit();
udpv6_fail:
	ipv6_frag_exit();
955 956
ipv6_frag_fail:
	ipv6_exthdrs_exit();
957 958
ipv6_exthdrs_fail:
	addrconf_cleanup();
L
Linus Torvalds 已提交
959 960
addrconf_fail:
	ip6_flowlabel_cleanup();
961
ip6_flowlabel_fail:
L
Linus Torvalds 已提交
962
	ip6_route_cleanup();
963
ip6_route_fail:
L
Linus Torvalds 已提交
964 965 966 967 968
#ifdef CONFIG_PROC_FS
	if6_proc_exit();
proc_if6_fail:
	ipv6_misc_proc_exit();
proc_misc6_fail:
969 970
	udplite6_proc_exit();
proc_udplite6_fail:
L
Linus Torvalds 已提交
971 972 973
	raw6_proc_exit();
proc_raw6_fail:
#endif
974 975
	ipv6_netfilter_fini();
netfilter_fail:
L
Linus Torvalds 已提交
976 977 978 979
	igmp6_cleanup();
igmp_fail:
	ndisc_cleanup();
ndisc_fail:
W
Wang Chen 已提交
980 981
	ip6_mr_cleanup();
ipmr_fail:
L
Linus Torvalds 已提交
982 983
	icmpv6_cleanup();
icmp_fail:
984 985
	unregister_pernet_subsys(&inet6_net_ops);
register_pernet_fail:
986
	sock_unregister(PF_INET6);
987
	rtnl_unregister_all(PF_INET6);
988 989
out_sock_register_fail:
	rawv6_exit();
L
Linus Torvalds 已提交
990 991
out_unregister_raw_proto:
	proto_unregister(&rawv6_prot);
992 993
out_unregister_udplite_proto:
	proto_unregister(&udplitev6_prot);
L
Linus Torvalds 已提交
994 995 996 997 998 999 1000 1001 1002 1003
out_unregister_udp_proto:
	proto_unregister(&udpv6_prot);
out_unregister_tcp_proto:
	proto_unregister(&tcpv6_prot);
	goto out;
}
module_init(inet6_init);

static void __exit inet6_exit(void)
{
1004
	if (disable_ipv6_mod)
1005 1006
		return;

L
Linus Torvalds 已提交
1007 1008
	/* First of all disallow new sockets creation. */
	sock_unregister(PF_INET6);
1009 1010
	/* Disallow any further netlink messages */
	rtnl_unregister_all(PF_INET6);
1011

1012 1013 1014 1015
	udpv6_exit();
	udplitev6_exit();
	tcpv6_exit();

1016 1017
	/* Cleanup code parts. */
	ipv6_packet_cleanup();
1018
	ipv6_frag_exit();
1019
	ipv6_exthdrs_exit();
1020 1021 1022
	addrconf_cleanup();
	ip6_flowlabel_cleanup();
	ip6_route_cleanup();
L
Linus Torvalds 已提交
1023
#ifdef CONFIG_PROC_FS
1024 1025

	/* Cleanup code parts. */
L
Linus Torvalds 已提交
1026
	if6_proc_exit();
1027 1028 1029
	ipv6_misc_proc_exit();
	udplite6_proc_exit();
	raw6_proc_exit();
1030
#endif
1031
	ipv6_netfilter_fini();
1032
	igmp6_cleanup();
L
Linus Torvalds 已提交
1033
	ndisc_cleanup();
W
Wang Chen 已提交
1034
	ip6_mr_cleanup();
L
Linus Torvalds 已提交
1035
	icmpv6_cleanup();
1036
	rawv6_exit();
1037

1038
	unregister_pernet_subsys(&inet6_net_ops);
L
Linus Torvalds 已提交
1039
	proto_unregister(&rawv6_prot);
1040
	proto_unregister(&udplitev6_prot);
L
Linus Torvalds 已提交
1041 1042
	proto_unregister(&udpv6_prot);
	proto_unregister(&tcpv6_prot);
1043 1044

	rcu_barrier(); /* Wait for completion of call_rcu()'s */
L
Linus Torvalds 已提交
1045 1046 1047 1048
}
module_exit(inet6_exit);

MODULE_ALIAS_NETPROTO(PF_INET6);