af_inet6.c 24.5 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
#include <net/tcp.h>
52
#include <net/ping.h>
L
Linus Torvalds 已提交
53 54
#include <net/protocol.h>
#include <net/inet_common.h>
55
#include <net/route.h>
L
Linus Torvalds 已提交
56 57 58
#include <net/transp_v6.h>
#include <net/ip6_route.h>
#include <net/addrconf.h>
C
Cong Wang 已提交
59
#include <net/ndisc.h>
L
Linus Torvalds 已提交
60 61 62 63 64
#ifdef CONFIG_IPV6_TUNNEL
#include <net/ip6_tunnel.h>
#endif

#include <asm/uaccess.h>
65
#include <linux/mroute6.h>
L
Linus Torvalds 已提交
66 67 68 69 70

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

71
/* The inetsw6 table contains everything that inet6_create needs to
L
Linus Torvalds 已提交
72 73 74 75 76
 * build a new socket.
 */
static struct list_head inetsw6[SOCK_MAX];
static DEFINE_SPINLOCK(inetsw6_lock);

77 78 79 80 81
struct ipv6_params ipv6_defaults = {
	.disable_ipv6 = 0,
	.autoconf = 1,
};

82
static int disable_ipv6_mod;
83 84 85 86 87 88 89 90 91

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");
92

L
Linus Torvalds 已提交
93 94 95 96 97 98 99
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);
}

100 101
static int inet6_create(struct net *net, struct socket *sock, int protocol,
			int kern)
L
Linus Torvalds 已提交
102 103 104 105 106 107 108 109
{
	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;
110 111
	int try_loading_module = 0;
	int err;
L
Linus Torvalds 已提交
112 113

	/* Look for the requested type/protocol pair. */
114 115
lookup_protocol:
	err = -ESOCKTNOSUPPORT;
L
Linus Torvalds 已提交
116
	rcu_read_lock();
117
	list_for_each_entry_rcu(answer, &inetsw6[sock->type], list) {
L
Linus Torvalds 已提交
118

119
		err = 0;
L
Linus Torvalds 已提交
120 121 122 123 124 125 126 127 128 129 130 131 132
		/* 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;
		}
133
		err = -EPROTONOSUPPORT;
L
Linus Torvalds 已提交
134 135
	}

136
	if (err) {
137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158
		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;
159 160
	if (sock->type == SOCK_RAW && !kern &&
	    !ns_capable(net->user_ns, CAP_NET_RAW))
L
Linus Torvalds 已提交
161 162 163 164 165 166 167 168
		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();

169
	WARN_ON(answer_prot->slab == NULL);
L
Linus Torvalds 已提交
170

171
	err = -ENOBUFS;
172
	sk = sk_alloc(net, PF_INET6, GFP_KERNEL, answer_prot);
L
Linus Torvalds 已提交
173 174 175 176 177
	if (sk == NULL)
		goto out;

	sock_init_data(sock, sk);

178
	err = 0;
L
Linus Torvalds 已提交
179 180
	sk->sk_no_check = answer_no_check;
	if (INET_PROTOSW_REUSE & answer_flags)
181
		sk->sk_reuse = SK_CAN_REUSE;
L
Linus Torvalds 已提交
182 183

	inet = inet_sk(sk);
184
	inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
L
Linus Torvalds 已提交
185 186

	if (SOCK_RAW == sock->type) {
E
Eric Dumazet 已提交
187
		inet->inet_num = protocol;
L
Linus Torvalds 已提交
188 189 190 191
		if (IPPROTO_RAW == protocol)
			inet->hdrincl = 1;
	}

192
	sk->sk_destruct		= inet_sock_destruct;
L
Linus Torvalds 已提交
193 194 195 196 197 198 199
	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;
200
	np->mcast_hops	= IPV6_DEFAULT_MCASTHOPS;
L
Linus Torvalds 已提交
201 202
	np->mc_loop	= 1;
	np->pmtudisc	= IPV6_PMTUDISC_WANT;
203
	np->ipv6only	= net->ipv6.sysctl.bindv6only;
204

L
Linus Torvalds 已提交
205 206 207 208 209 210 211 212 213
	/* 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;
214
	inet->rcv_tos	= 0;
L
Linus Torvalds 已提交
215

216
	if (net->ipv4.sysctl_ip_no_pmtu_disc)
L
Linus Torvalds 已提交
217 218 219
		inet->pmtudisc = IP_PMTUDISC_DONT;
	else
		inet->pmtudisc = IP_PMTUDISC_WANT;
220
	/*
221 222 223 224 225 226 227 228 229
	 * 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
230

E
Eric Dumazet 已提交
231
	if (inet->inet_num) {
L
Linus Torvalds 已提交
232 233 234 235
		/* 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
236
		inet->inet_sport = htons(inet->inet_num);
L
Linus Torvalds 已提交
237 238 239
		sk->sk_prot->hash(sk);
	}
	if (sk->sk_prot->init) {
240 241
		err = sk->sk_prot->init(sk);
		if (err) {
L
Linus Torvalds 已提交
242 243 244 245 246
			sk_common_release(sk);
			goto out;
		}
	}
out:
247
	return err;
L
Linus Torvalds 已提交
248 249 250 251 252 253 254 255 256
out_rcu_unlock:
	rcu_read_unlock();
	goto out;
}


/* bind for INET6 API */
int inet6_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
{
257
	struct sockaddr_in6 *addr = (struct sockaddr_in6 *)uaddr;
L
Linus Torvalds 已提交
258 259 260
	struct sock *sk = sock->sk;
	struct inet_sock *inet = inet_sk(sk);
	struct ipv6_pinfo *np = inet6_sk(sk);
261
	struct net *net = sock_net(sk);
A
Al Viro 已提交
262
	__be32 v4addr = 0;
L
Linus Torvalds 已提交
263 264 265 266 267 268 269 270 271 272
	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;
273 274

	if (addr->sin6_family != AF_INET6)
275
		return -EAFNOSUPPORT;
276

L
Linus Torvalds 已提交
277 278 279 280 281
	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);
282
	if (snum && snum < PROT_SOCK && !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
L
Linus Torvalds 已提交
283 284 285 286 287
		return -EACCES;

	lock_sock(sk);

	/* Check these errors (active socket, double bind). */
E
Eric Dumazet 已提交
288
	if (sk->sk_state != TCP_CLOSE || inet->inet_num) {
L
Linus Torvalds 已提交
289 290 291 292 293 294
		err = -EINVAL;
		goto out;
	}

	/* Check if the address belongs to the host. */
	if (addr_type == IPV6_ADDR_MAPPED) {
295 296
		int chk_addr_ret;

297 298 299 300 301 302 303
		/* Binding to v4-mapped address on a v6-only socket
		 * makes no sense
		 */
		if (np->ipv6only) {
			err = -EINVAL;
			goto out;
		}
304

305
		/* Reproduce AF_INET checks to make the bindings consistent */
L
Linus Torvalds 已提交
306
		v4addr = addr->sin6_addr.s6_addr32[3];
307 308 309 310 311 312
		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 &&
313 314
		    chk_addr_ret != RTN_BROADCAST) {
			err = -EADDRNOTAVAIL;
L
Linus Torvalds 已提交
315
			goto out;
316
		}
L
Linus Torvalds 已提交
317 318 319 320
	} else {
		if (addr_type != IPV6_ADDR_ANY) {
			struct net_device *dev = NULL;

321
			rcu_read_lock();
322
			if (__ipv6_addr_needs_scope_id(addr_type)) {
L
Linus Torvalds 已提交
323 324 325 326 327 328 329
				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;
				}
330

L
Linus Torvalds 已提交
331 332 333
				/* Binding to link-local address requires an interface */
				if (!sk->sk_bound_dev_if) {
					err = -EINVAL;
334
					goto out_unlock;
L
Linus Torvalds 已提交
335
				}
336
				dev = dev_get_by_index_rcu(net, sk->sk_bound_dev_if);
L
Linus Torvalds 已提交
337 338
				if (!dev) {
					err = -ENODEV;
339
					goto out_unlock;
L
Linus Torvalds 已提交
340 341 342 343 344 345 346 347
				}
			}

			/* 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))	{
348
				if (!(inet->freebind || inet->transparent) &&
349
				    !ipv6_chk_addr(net, &addr->sin6_addr,
350
						   dev, 0)) {
L
Linus Torvalds 已提交
351
					err = -EADDRNOTAVAIL;
352
					goto out_unlock;
L
Linus Torvalds 已提交
353 354
				}
			}
355
			rcu_read_unlock();
L
Linus Torvalds 已提交
356 357 358
		}
	}

E
Eric Dumazet 已提交
359 360
	inet->inet_rcv_saddr = v4addr;
	inet->inet_saddr = v4addr;
L
Linus Torvalds 已提交
361

362
	sk->sk_v6_rcv_saddr = addr->sin6_addr;
363

L
Linus Torvalds 已提交
364
	if (!(addr_type & IPV6_ADDR_MULTICAST))
A
Alexey Dobriyan 已提交
365
		np->saddr = addr->sin6_addr;
L
Linus Torvalds 已提交
366 367 368 369 370 371 372 373

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

374
	if (addr_type != IPV6_ADDR_ANY) {
L
Linus Torvalds 已提交
375
		sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
376 377 378
		if (addr_type != IPV6_ADDR_MAPPED)
			np->ipv6only = 1;
	}
L
Linus Torvalds 已提交
379 380
	if (snum)
		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
E
Eric Dumazet 已提交
381 382 383
	inet->inet_sport = htons(inet->inet_num);
	inet->inet_dport = 0;
	inet->inet_daddr = 0;
L
Linus Torvalds 已提交
384 385 386
out:
	release_sock(sk);
	return err;
387 388 389
out_unlock:
	rcu_read_unlock();
	goto out;
L
Linus Torvalds 已提交
390
}
391 392
EXPORT_SYMBOL(inet6_bind);

L
Linus Torvalds 已提交
393 394 395 396 397 398 399 400 401 402 403 404 405 406 407
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);
}
408 409
EXPORT_SYMBOL(inet6_release);

410
void inet6_destroy_sock(struct sock *sk)
L
Linus Torvalds 已提交
411 412 413 414 415 416 417
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct sk_buff *skb;
	struct ipv6_txoptions *opt;

	/* Release rx options */

418 419
	skb = xchg(&np->pktoptions, NULL);
	if (skb != NULL)
L
Linus Torvalds 已提交
420
		kfree_skb(skb);
421

422 423
	skb = xchg(&np->rxpmtu, NULL);
	if (skb != NULL)
424
		kfree_skb(skb);
L
Linus Torvalds 已提交
425 426 427 428 429 430

	/* Free flowlabels */
	fl6_free_socklist(sk);

	/* Free tx options */

431 432
	opt = xchg(&np->opt, NULL);
	if (opt != NULL)
L
Linus Torvalds 已提交
433 434
		sock_kfree_s(sk, opt, opt->tot_len);
}
435 436
EXPORT_SYMBOL_GPL(inet6_destroy_sock);

L
Linus Torvalds 已提交
437 438 439
/*
 *	This does both peername and sockname.
 */
440

L
Linus Torvalds 已提交
441 442 443
int inet6_getname(struct socket *sock, struct sockaddr *uaddr,
		 int *uaddr_len, int peer)
{
444
	struct sockaddr_in6 *sin = (struct sockaddr_in6 *)uaddr;
L
Linus Torvalds 已提交
445 446 447
	struct sock *sk = sock->sk;
	struct inet_sock *inet = inet_sk(sk);
	struct ipv6_pinfo *np = inet6_sk(sk);
448

L
Linus Torvalds 已提交
449 450 451 452
	sin->sin6_family = AF_INET6;
	sin->sin6_flowinfo = 0;
	sin->sin6_scope_id = 0;
	if (peer) {
E
Eric Dumazet 已提交
453
		if (!inet->inet_dport)
L
Linus Torvalds 已提交
454 455 456 457
			return -ENOTCONN;
		if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
		    peer == 1)
			return -ENOTCONN;
E
Eric Dumazet 已提交
458
		sin->sin6_port = inet->inet_dport;
459
		sin->sin6_addr = sk->sk_v6_daddr;
L
Linus Torvalds 已提交
460 461 462
		if (np->sndflow)
			sin->sin6_flowinfo = np->flow_label;
	} else {
463
		if (ipv6_addr_any(&sk->sk_v6_rcv_saddr))
A
Alexey Dobriyan 已提交
464
			sin->sin6_addr = np->saddr;
L
Linus Torvalds 已提交
465
		else
466
			sin->sin6_addr = sk->sk_v6_rcv_saddr;
L
Linus Torvalds 已提交
467

E
Eric Dumazet 已提交
468
		sin->sin6_port = inet->inet_sport;
L
Linus Torvalds 已提交
469
	}
470 471
	sin->sin6_scope_id = ipv6_iface_scope_id(&sin->sin6_addr,
						 sk->sk_bound_dev_if);
L
Linus Torvalds 已提交
472
	*uaddr_len = sizeof(*sin);
E
Eric Dumazet 已提交
473
	return 0;
L
Linus Torvalds 已提交
474
}
475 476
EXPORT_SYMBOL(inet6_getname);

L
Linus Torvalds 已提交
477 478 479
int inet6_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
{
	struct sock *sk = sock->sk;
480
	struct net *net = sock_net(sk);
L
Linus Torvalds 已提交
481

482
	switch (cmd) {
L
Linus Torvalds 已提交
483 484 485
	case SIOCGSTAMP:
		return sock_get_timestamp(sk, (struct timeval __user *)arg);

486 487 488
	case SIOCGSTAMPNS:
		return sock_get_timestampns(sk, (struct timespec __user *)arg);

L
Linus Torvalds 已提交
489 490
	case SIOCADDRT:
	case SIOCDELRT:
491

E
Eric Dumazet 已提交
492
		return ipv6_route_ioctl(net, cmd, (void __user *)arg);
L
Linus Torvalds 已提交
493 494

	case SIOCSIFADDR:
495
		return addrconf_add_ifaddr(net, (void __user *) arg);
L
Linus Torvalds 已提交
496
	case SIOCDIFADDR:
497
		return addrconf_del_ifaddr(net, (void __user *) arg);
L
Linus Torvalds 已提交
498
	case SIOCSIFDSTADDR:
499
		return addrconf_set_dstaddr(net, (void __user *) arg);
L
Linus Torvalds 已提交
500
	default:
501 502 503
		if (!sk->sk_prot->ioctl)
			return -ENOIOCTLCMD;
		return sk->sk_prot->ioctl(sk, cmd, arg);
L
Linus Torvalds 已提交
504 505
	}
	/*NOTREACHED*/
E
Eric Dumazet 已提交
506
	return 0;
L
Linus Torvalds 已提交
507
}
508 509
EXPORT_SYMBOL(inet6_ioctl);

510
const struct proto_ops inet6_stream_ops = {
511 512 513 514 515 516 517 518 519 520 521 522 523 524
	.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		*/
525 526
	.sendmsg	   = inet_sendmsg,		/* ok		*/
	.recvmsg	   = inet_recvmsg,		/* ok		*/
527
	.mmap		   = sock_no_mmap,
528
	.sendpage	   = inet_sendpage,
529
	.splice_read	   = tcp_splice_read,
530
#ifdef CONFIG_COMPAT
531 532
	.compat_setsockopt = compat_sock_common_setsockopt,
	.compat_getsockopt = compat_sock_common_getsockopt,
533
#endif
L
Linus Torvalds 已提交
534 535
};

536
const struct proto_ops inet6_dgram_ops = {
537 538 539 540 541 542 543 544 545 546 547 548 549 550 551
	.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		*/
552
	.recvmsg	   = inet_recvmsg,		/* ok		*/
553 554
	.mmap		   = sock_no_mmap,
	.sendpage	   = sock_no_sendpage,
555
#ifdef CONFIG_COMPAT
556 557
	.compat_setsockopt = compat_sock_common_setsockopt,
	.compat_getsockopt = compat_sock_common_getsockopt,
558
#endif
L
Linus Torvalds 已提交
559 560
};

561
static const struct net_proto_family inet6_family_ops = {
L
Linus Torvalds 已提交
562 563 564 565 566
	.family = PF_INET6,
	.create = inet6_create,
	.owner	= THIS_MODULE,
};

567
int inet6_register_protosw(struct inet_protosw *p)
L
Linus Torvalds 已提交
568 569 570 571
{
	struct list_head *lh;
	struct inet_protosw *answer;
	struct list_head *last_perm;
572 573
	int protocol = p->protocol;
	int ret;
L
Linus Torvalds 已提交
574 575 576

	spin_lock_bh(&inetsw6_lock);

577
	ret = -EINVAL;
L
Linus Torvalds 已提交
578 579 580 581 582
	if (p->type >= SOCK_MAX)
		goto out_illegal;

	/* If we are trying to override a permanent protocol, bail. */
	answer = NULL;
583
	ret = -EPERM;
L
Linus Torvalds 已提交
584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602
	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
603
	 * non-permanent entry.  This means that when we remove this entry, the
L
Linus Torvalds 已提交
604 605 606
	 * system automatically returns to the old behavior.
	 */
	list_add_rcu(&p->list, last_perm);
607
	ret = 0;
L
Linus Torvalds 已提交
608 609
out:
	spin_unlock_bh(&inetsw6_lock);
610
	return ret;
L
Linus Torvalds 已提交
611 612

out_permanent:
613
	pr_err("Attempt to override permanent protocol %d\n", protocol);
L
Linus Torvalds 已提交
614 615 616
	goto out;

out_illegal:
617
	pr_err("Ignoring attempt to register invalid socket type %d\n",
L
Linus Torvalds 已提交
618 619 620
	       p->type);
	goto out;
}
621 622
EXPORT_SYMBOL(inet6_register_protosw);

L
Linus Torvalds 已提交
623 624 625 626
void
inet6_unregister_protosw(struct inet_protosw *p)
{
	if (INET_PROTOSW_PERMANENT & p->flags) {
627
		pr_err("Attempt to unregister permanent protocol %d\n",
L
Linus Torvalds 已提交
628 629 630 631 632 633 634 635 636
		       p->protocol);
	} else {
		spin_lock_bh(&inetsw6_lock);
		list_del_rcu(&p->list);
		spin_unlock_bh(&inetsw6_lock);

		synchronize_net();
	}
}
637 638
EXPORT_SYMBOL(inet6_unregister_protosw);

639 640 641
int inet6_sk_rebuild_header(struct sock *sk)
{
	struct ipv6_pinfo *np = inet6_sk(sk);
642
	struct dst_entry *dst;
643 644 645 646 647

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

	if (dst == NULL) {
		struct inet_sock *inet = inet_sk(sk);
648
		struct in6_addr *final_p, final;
649 650 651 652
		struct flowi6 fl6;

		memset(&fl6, 0, sizeof(fl6));
		fl6.flowi6_proto = sk->sk_protocol;
653
		fl6.daddr = sk->sk_v6_daddr;
A
Alexey Dobriyan 已提交
654
		fl6.saddr = np->saddr;
655 656 657
		fl6.flowlabel = np->flow_label;
		fl6.flowi6_oif = sk->sk_bound_dev_if;
		fl6.flowi6_mark = sk->sk_mark;
658 659
		fl6.fl6_dport = inet->inet_dport;
		fl6.fl6_sport = inet->inet_sport;
660 661 662 663
		security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));

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

664
		dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
665
		if (IS_ERR(dst)) {
666
			sk->sk_route_caps = 0;
667 668
			sk->sk_err_soft = -PTR_ERR(dst);
			return PTR_ERR(dst);
669 670
		}

671
		__ip6_dst_store(sk, dst, NULL, NULL);
672 673 674 675 676 677
	}

	return 0;
}
EXPORT_SYMBOL_GPL(inet6_sk_rebuild_header);

E
Eric Dumazet 已提交
678
bool ipv6_opt_accepted(const struct sock *sk, const struct sk_buff *skb)
679
{
E
Eric Dumazet 已提交
680 681
	const struct ipv6_pinfo *np = inet6_sk(sk);
	const struct inet6_skb_parm *opt = IP6CB(skb);
682 683 684 685

	if (np->rxopt.all) {
		if ((opt->hop && (np->rxopt.bits.hopopts ||
				  np->rxopt.bits.ohopopts)) ||
F
Florent Fourcot 已提交
686
		    (ip6_flowinfo((struct ipv6hdr *) skb_network_header(skb)) &&
687 688 689 690 691
		     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
692
			return true;
693
	}
E
Eric Dumazet 已提交
694
	return false;
695 696 697
}
EXPORT_SYMBOL_GPL(ipv6_opt_accepted);

698
static struct packet_type ipv6_packet_type __read_mostly = {
699
	.type = cpu_to_be16(ETH_P_IPV6),
700
	.func = ipv6_rcv,
701 702
};

703 704 705 706 707 708 709 710 711 712 713
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);
}

714 715
static int __net_init ipv6_init_mibs(struct net *net)
{
716 717
	int i;

718
	if (snmp_mib_init((void __percpu **)net->mib.udp_stats_in6,
719 720
			  sizeof(struct udp_mib),
			  __alignof__(struct udp_mib)) < 0)
721
		return -ENOMEM;
722
	if (snmp_mib_init((void __percpu **)net->mib.udplite_stats_in6,
723 724
			  sizeof(struct udp_mib),
			  __alignof__(struct udp_mib)) < 0)
725
		goto err_udplite_mib;
726
	if (snmp_mib_init((void __percpu **)net->mib.ipv6_statistics,
727 728
			  sizeof(struct ipstats_mib),
			  __alignof__(struct ipstats_mib)) < 0)
729
		goto err_ip_mib;
730 731 732 733 734 735 736 737 738 739 740 741

	for_each_possible_cpu(i) {
		struct ipstats_mib *af_inet6_stats;
		af_inet6_stats = per_cpu_ptr(net->mib.ipv6_statistics[0], i);
		u64_stats_init(&af_inet6_stats->syncp);
#if SNMP_ARRAY_SZ == 2
		af_inet6_stats = per_cpu_ptr(net->mib.ipv6_statistics[1], i);
		u64_stats_init(&af_inet6_stats->syncp);
#endif
	}


742
	if (snmp_mib_init((void __percpu **)net->mib.icmpv6_statistics,
743 744
			  sizeof(struct icmpv6_mib),
			  __alignof__(struct icmpv6_mib)) < 0)
745
		goto err_icmp_mib;
746 747 748
	net->mib.icmpv6msg_statistics = kzalloc(sizeof(struct icmpv6msg_mib),
						GFP_KERNEL);
	if (!net->mib.icmpv6msg_statistics)
749
		goto err_icmpmsg_mib;
750
	return 0;
751

752
err_icmpmsg_mib:
753
	snmp_mib_free((void __percpu **)net->mib.icmpv6_statistics);
754
err_icmp_mib:
755
	snmp_mib_free((void __percpu **)net->mib.ipv6_statistics);
756
err_ip_mib:
757
	snmp_mib_free((void __percpu **)net->mib.udplite_stats_in6);
758
err_udplite_mib:
759
	snmp_mib_free((void __percpu **)net->mib.udp_stats_in6);
760
	return -ENOMEM;
761 762
}

763
static void ipv6_cleanup_mibs(struct net *net)
764
{
765 766 767 768
	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);
769
	kfree(net->mib.icmpv6msg_statistics);
770 771
}

A
Alexey Dobriyan 已提交
772
static int __net_init inet6_net_init(struct net *net)
773
{
774 775
	int err = 0;

776
	net->ipv6.sysctl.bindv6only = 0;
777
	net->ipv6.sysctl.icmpv6_time = 1*HZ;
F
fan.du 已提交
778
	atomic_set(&net->ipv6.rt_genid, 0);
779

780 781 782
	err = ipv6_init_mibs(net);
	if (err)
		return err;
783 784 785 786
#ifdef CONFIG_PROC_FS
	err = udp6_proc_init(net);
	if (err)
		goto out;
787 788 789
	err = tcp6_proc_init(net);
	if (err)
		goto proc_tcp6_fail;
790 791 792
	err = ac6_proc_init(net);
	if (err)
		goto proc_ac6_fail;
793 794
#endif
	return err;
795 796

#ifdef CONFIG_PROC_FS
797 798
proc_ac6_fail:
	tcp6_proc_exit(net);
799 800
proc_tcp6_fail:
	udp6_proc_exit(net);
801 802 803
out:
	ipv6_cleanup_mibs(net);
	return err;
804
#endif
805 806
}

807
static void __net_exit inet6_net_exit(struct net *net)
808
{
809 810
#ifdef CONFIG_PROC_FS
	udp6_proc_exit(net);
811
	tcp6_proc_exit(net);
812
	ac6_proc_exit(net);
813
#endif
814
	ipv6_cleanup_mibs(net);
815 816 817 818 819 820 821
}

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

822 823 824 825 826
static const struct ipv6_stub ipv6_stub_impl = {
	.ipv6_sock_mc_join = ipv6_sock_mc_join,
	.ipv6_sock_mc_drop = ipv6_sock_mc_drop,
	.ipv6_dst_lookup = ip6_dst_lookup,
	.udpv6_encap_enable = udpv6_encap_enable,
C
Cong Wang 已提交
827
	.ndisc_send_na = ndisc_send_na,
828
	.nd_tbl	= &nd_tbl,
829 830
};

L
Linus Torvalds 已提交
831 832
static int __init inet6_init(void)
{
833
	struct list_head *r;
834
	int err = 0;
L
Linus Torvalds 已提交
835

836
	BUILD_BUG_ON(sizeof(struct inet6_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
837

838
	/* Register the socket-side information for inet6_create.  */
839
	for (r = &inetsw6[0]; r < &inetsw6[SOCK_MAX]; ++r)
840 841
		INIT_LIST_HEAD(r);

842
	if (disable_ipv6_mod) {
843
		pr_info("Loaded, but administratively disabled, reboot required to enable\n");
844 845 846
		goto out;
	}

L
Linus Torvalds 已提交
847 848 849 850 851 852 853 854
	err = proto_register(&tcpv6_prot, 1);
	if (err)
		goto out;

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

855
	err = proto_register(&udplitev6_prot, 1);
L
Linus Torvalds 已提交
856 857 858
	if (err)
		goto out_unregister_udp_proto;

859 860 861 862
	err = proto_register(&rawv6_prot, 1);
	if (err)
		goto out_unregister_udplite_proto;

863 864 865
	err = proto_register(&pingv6_prot, 1);
	if (err)
		goto out_unregister_ping_proto;
L
Linus Torvalds 已提交
866 867 868 869

	/* We MUST register RAW sockets before we create the ICMP6,
	 * IGMP6, or NDISC control sockets.
	 */
870 871 872
	err = rawv6_init();
	if (err)
		goto out_unregister_raw_proto;
L
Linus Torvalds 已提交
873 874 875 876

	/* Register the family here so that the init calls below will
	 * be able to create sockets. (?? is this dangerous ??)
	 */
877 878
	err = sock_register(&inet6_family_ops);
	if (err)
879
		goto out_sock_register_fail;
L
Linus Torvalds 已提交
880 881 882 883

	/*
	 *	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
884
	 *	in a host available by both INET and INET6 APIs and
L
Linus Torvalds 已提交
885 886 887
	 *	able to communicate via both network protocols.
	 */

888 889 890
	err = register_pernet_subsys(&inet6_net_ops);
	if (err)
		goto register_pernet_fail;
891
	err = icmpv6_init();
L
Linus Torvalds 已提交
892 893
	if (err)
		goto icmp_fail;
W
Wang Chen 已提交
894 895 896
	err = ip6_mr_init();
	if (err)
		goto ipmr_fail;
897
	err = ndisc_init();
L
Linus Torvalds 已提交
898 899
	if (err)
		goto ndisc_fail;
900
	err = igmp6_init();
L
Linus Torvalds 已提交
901 902
	if (err)
		goto igmp_fail;
903 904 905

	ipv6_stub = &ipv6_stub_impl;

906 907 908
	err = ipv6_netfilter_init();
	if (err)
		goto netfilter_fail;
L
Linus Torvalds 已提交
909 910 911 912 913
	/* Create /proc/foo6 entries. */
#ifdef CONFIG_PROC_FS
	err = -ENOMEM;
	if (raw6_proc_init())
		goto proc_raw6_fail;
914 915
	if (udplite6_proc_init())
		goto proc_udplite6_fail;
L
Linus Torvalds 已提交
916 917 918 919 920
	if (ipv6_misc_proc_init())
		goto proc_misc6_fail;
	if (if6_proc_init())
		goto proc_if6_fail;
#endif
921 922 923
	err = ip6_route_init();
	if (err)
		goto ip6_route_fail;
924 925 926
	err = ndisc_late_init();
	if (err)
		goto ndisc_late_fail;
927 928 929
	err = ip6_flowlabel_init();
	if (err)
		goto ip6_flowlabel_fail;
L
Linus Torvalds 已提交
930 931 932 933 934
	err = addrconf_init();
	if (err)
		goto addrconf_fail;

	/* Init v6 extension headers. */
935 936 937 938
	err = ipv6_exthdrs_init();
	if (err)
		goto ipv6_exthdrs_fail;

939 940 941
	err = ipv6_frag_init();
	if (err)
		goto ipv6_frag_fail;
L
Linus Torvalds 已提交
942 943

	/* Init v6 transport protocols. */
944 945 946
	err = udpv6_init();
	if (err)
		goto udpv6_fail;
947

948 949 950 951 952 953 954 955 956 957 958
	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;
959

960 961 962 963
	err = pingv6_init();
	if (err)
		goto pingv6_fail;

964 965 966 967 968
#ifdef CONFIG_SYSCTL
	err = ipv6_sysctl_register();
	if (err)
		goto sysctl_fail;
#endif
L
Linus Torvalds 已提交
969 970 971
out:
	return err;

972 973
#ifdef CONFIG_SYSCTL
sysctl_fail:
974
	pingv6_exit();
975
#endif
976
pingv6_fail:
977
	ipv6_packet_cleanup();
978 979 980 981 982 983 984 985
ipv6_packet_fail:
	tcpv6_exit();
tcpv6_fail:
	udplitev6_exit();
udplitev6_fail:
	udpv6_exit();
udpv6_fail:
	ipv6_frag_exit();
986 987
ipv6_frag_fail:
	ipv6_exthdrs_exit();
988 989
ipv6_exthdrs_fail:
	addrconf_cleanup();
L
Linus Torvalds 已提交
990 991
addrconf_fail:
	ip6_flowlabel_cleanup();
992
ip6_flowlabel_fail:
993 994
	ndisc_late_cleanup();
ndisc_late_fail:
L
Linus Torvalds 已提交
995
	ip6_route_cleanup();
996
ip6_route_fail:
L
Linus Torvalds 已提交
997 998 999 1000 1001
#ifdef CONFIG_PROC_FS
	if6_proc_exit();
proc_if6_fail:
	ipv6_misc_proc_exit();
proc_misc6_fail:
1002 1003
	udplite6_proc_exit();
proc_udplite6_fail:
L
Linus Torvalds 已提交
1004 1005 1006
	raw6_proc_exit();
proc_raw6_fail:
#endif
1007 1008
	ipv6_netfilter_fini();
netfilter_fail:
L
Linus Torvalds 已提交
1009 1010 1011 1012
	igmp6_cleanup();
igmp_fail:
	ndisc_cleanup();
ndisc_fail:
W
Wang Chen 已提交
1013 1014
	ip6_mr_cleanup();
ipmr_fail:
L
Linus Torvalds 已提交
1015 1016
	icmpv6_cleanup();
icmp_fail:
1017 1018
	unregister_pernet_subsys(&inet6_net_ops);
register_pernet_fail:
1019
	sock_unregister(PF_INET6);
1020
	rtnl_unregister_all(PF_INET6);
1021 1022
out_sock_register_fail:
	rawv6_exit();
1023 1024
out_unregister_ping_proto:
	proto_unregister(&pingv6_prot);
L
Linus Torvalds 已提交
1025 1026
out_unregister_raw_proto:
	proto_unregister(&rawv6_prot);
1027 1028
out_unregister_udplite_proto:
	proto_unregister(&udplitev6_prot);
L
Linus Torvalds 已提交
1029 1030 1031 1032 1033 1034 1035 1036 1037
out_unregister_udp_proto:
	proto_unregister(&udpv6_prot);
out_unregister_tcp_proto:
	proto_unregister(&tcpv6_prot);
	goto out;
}
module_init(inet6_init);

MODULE_ALIAS_NETPROTO(PF_INET6);