datagram.c 23.2 KB
Newer Older
L
Linus Torvalds 已提交
1 2
/*
 *	common UDP/RAW code
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
 *
 *	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.
 */

14
#include <linux/capability.h>
L
Linus Torvalds 已提交
15 16 17 18 19 20 21 22 23
#include <linux/errno.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/interrupt.h>
#include <linux/socket.h>
#include <linux/sockios.h>
#include <linux/in6.h>
#include <linux/ipv6.h>
#include <linux/route.h>
24
#include <linux/slab.h>
25
#include <linux/export.h>
L
Linus Torvalds 已提交
26 27 28 29 30 31

#include <net/ipv6.h>
#include <net/ndisc.h>
#include <net/addrconf.h>
#include <net/transp_v6.h>
#include <net/ip6_route.h>
32
#include <net/tcp_states.h>
33
#include <net/dsfield.h>
L
Linus Torvalds 已提交
34 35 36 37

#include <linux/errqueue.h>
#include <asm/uaccess.h>

38
static bool ipv6_mapped_addr_any(const struct in6_addr *a)
39
{
40
	return ipv6_addr_v4mapped(a) && (a->s6_addr32[3] == 0);
41 42
}

43
static int __ip6_datagram_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
L
Linus Torvalds 已提交
44 45
{
	struct sockaddr_in6	*usin = (struct sockaddr_in6 *) uaddr;
46 47 48
	struct inet_sock	*inet = inet_sk(sk);
	struct ipv6_pinfo	*np = inet6_sk(sk);
	struct in6_addr	*daddr, *final_p, final;
L
Linus Torvalds 已提交
49
	struct dst_entry	*dst;
50
	struct flowi6		fl6;
L
Linus Torvalds 已提交
51
	struct ip6_flowlabel	*flowlabel = NULL;
52
	struct ipv6_txoptions	*opt;
L
Linus Torvalds 已提交
53 54 55 56 57 58
	int			addr_type;
	int			err;

	if (usin->sin6_family == AF_INET) {
		if (__ipv6_only_sock(sk))
			return -EAFNOSUPPORT;
59
		err = __ip4_datagram_connect(sk, uaddr, addr_len);
L
Linus Torvalds 已提交
60 61 62 63
		goto ipv4_connected;
	}

	if (addr_len < SIN6_LEN_RFC2133)
64
		return -EINVAL;
L
Linus Torvalds 已提交
65

66 67
	if (usin->sin6_family != AF_INET6)
		return -EAFNOSUPPORT;
L
Linus Torvalds 已提交
68

69
	memset(&fl6, 0, sizeof(fl6));
L
Linus Torvalds 已提交
70
	if (np->sndflow) {
71 72 73
		fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
		if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
			flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
74
			if (!flowlabel)
L
Linus Torvalds 已提交
75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
				return -EINVAL;
		}
	}

	addr_type = ipv6_addr_type(&usin->sin6_addr);

	if (addr_type == IPV6_ADDR_ANY) {
		/*
		 *	connect to self
		 */
		usin->sin6_addr.s6_addr[15] = 0x01;
	}

	daddr = &usin->sin6_addr;

	if (addr_type == IPV6_ADDR_MAPPED) {
		struct sockaddr_in sin;

		if (__ipv6_only_sock(sk)) {
			err = -ENETUNREACH;
			goto out;
		}
		sin.sin_family = AF_INET;
		sin.sin_addr.s_addr = daddr->s6_addr32[3];
		sin.sin_port = usin->sin6_port;

101 102 103
		err = __ip4_datagram_connect(sk,
					     (struct sockaddr *) &sin,
					     sizeof(sin));
L
Linus Torvalds 已提交
104 105 106 107

ipv4_connected:
		if (err)
			goto out;
108

109
		ipv6_addr_set_v4mapped(inet->inet_daddr, &sk->sk_v6_daddr);
L
Linus Torvalds 已提交
110

111 112
		if (ipv6_addr_any(&np->saddr) ||
		    ipv6_mapped_addr_any(&np->saddr))
E
Eric Dumazet 已提交
113
			ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
B
Brian Haley 已提交
114

115 116
		if (ipv6_addr_any(&sk->sk_v6_rcv_saddr) ||
		    ipv6_mapped_addr_any(&sk->sk_v6_rcv_saddr)) {
E
Eric Dumazet 已提交
117
			ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
118
					       &sk->sk_v6_rcv_saddr);
E
Eric Dumazet 已提交
119 120 121
			if (sk->sk_prot->rehash)
				sk->sk_prot->rehash(sk);
		}
L
Linus Torvalds 已提交
122 123 124 125

		goto out;
	}

126
	if (__ipv6_addr_needs_scope_id(addr_type)) {
L
Linus Torvalds 已提交
127 128 129 130 131 132 133 134 135 136
		if (addr_len >= sizeof(struct sockaddr_in6) &&
		    usin->sin6_scope_id) {
			if (sk->sk_bound_dev_if &&
			    sk->sk_bound_dev_if != usin->sin6_scope_id) {
				err = -EINVAL;
				goto out;
			}
			sk->sk_bound_dev_if = usin->sin6_scope_id;
		}

137 138 139
		if (!sk->sk_bound_dev_if && (addr_type & IPV6_ADDR_MULTICAST))
			sk->sk_bound_dev_if = np->mcast_oif;

L
Linus Torvalds 已提交
140 141 142 143 144 145 146
		/* Connect to link-local address requires an interface */
		if (!sk->sk_bound_dev_if) {
			err = -EINVAL;
			goto out;
		}
	}

147
	sk->sk_v6_daddr = *daddr;
148
	np->flow_label = fl6.flowlabel;
L
Linus Torvalds 已提交
149

E
Eric Dumazet 已提交
150
	inet->inet_dport = usin->sin6_port;
L
Linus Torvalds 已提交
151 152 153 154 155 156

	/*
	 *	Check for a route to destination an obtain the
	 *	destination cache for it.
	 */

157
	fl6.flowi6_proto = sk->sk_protocol;
158
	fl6.daddr = sk->sk_v6_daddr;
A
Alexey Dobriyan 已提交
159
	fl6.saddr = np->saddr;
160 161
	fl6.flowi6_oif = sk->sk_bound_dev_if;
	fl6.flowi6_mark = sk->sk_mark;
162 163
	fl6.fl6_dport = inet->inet_dport;
	fl6.fl6_sport = inet->inet_sport;
L
Linus Torvalds 已提交
164

165 166
	if (!fl6.flowi6_oif && (addr_type&IPV6_ADDR_MULTICAST))
		fl6.flowi6_oif = np->mcast_oif;
L
Linus Torvalds 已提交
167

168
	security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
V
Venkat Yekkirala 已提交
169

170
	opt = flowlabel ? flowlabel->opt : np->opt;
171
	final_p = fl6_update_dst(&fl6, opt, &final);
L
Linus Torvalds 已提交
172

173
	dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
174 175 176
	err = 0;
	if (IS_ERR(dst)) {
		err = PTR_ERR(dst);
L
Linus Torvalds 已提交
177
		goto out;
178
	}
L
Linus Torvalds 已提交
179 180 181 182

	/* source address lookup done in ip6_dst_lookup */

	if (ipv6_addr_any(&np->saddr))
A
Alexey Dobriyan 已提交
183
		np->saddr = fl6.saddr;
L
Linus Torvalds 已提交
184

185 186
	if (ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
		sk->sk_v6_rcv_saddr = fl6.saddr;
E
Eric Dumazet 已提交
187
		inet->inet_rcv_saddr = LOOPBACK4_IPV6;
E
Eric Dumazet 已提交
188 189
		if (sk->sk_prot->rehash)
			sk->sk_prot->rehash(sk);
L
Linus Torvalds 已提交
190 191 192
	}

	ip6_dst_store(sk, dst,
193 194
		      ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ?
		      &sk->sk_v6_daddr : NULL,
195
#ifdef CONFIG_IPV6_SUBTREES
196
		      ipv6_addr_equal(&fl6.saddr, &np->saddr) ?
197 198 199
		      &np->saddr :
#endif
		      NULL);
L
Linus Torvalds 已提交
200 201

	sk->sk_state = TCP_ESTABLISHED;
202
	sk_set_txhash(sk);
L
Linus Torvalds 已提交
203 204 205 206
out:
	fl6_sock_release(flowlabel);
	return err;
}
207 208 209 210 211 212 213 214 215 216

int ip6_datagram_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
{
	int res;

	lock_sock(sk);
	res = __ip6_datagram_connect(sk, uaddr, addr_len);
	release_sock(sk);
	return res;
}
217
EXPORT_SYMBOL_GPL(ip6_datagram_connect);
L
Linus Torvalds 已提交
218

219 220 221 222 223 224 225 226 227 228
int ip6_datagram_connect_v6_only(struct sock *sk, struct sockaddr *uaddr,
				 int addr_len)
{
	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, uaddr);
	if (sin6->sin6_family != AF_INET6)
		return -EAFNOSUPPORT;
	return ip6_datagram_connect(sk, uaddr, addr_len);
}
EXPORT_SYMBOL_GPL(ip6_datagram_connect_v6_only);

229
void ipv6_icmp_error(struct sock *sk, struct sk_buff *skb, int err,
A
Al Viro 已提交
230
		     __be16 port, u32 info, u8 *payload)
L
Linus Torvalds 已提交
231 232
{
	struct ipv6_pinfo *np  = inet6_sk(sk);
233
	struct icmp6hdr *icmph = icmp6_hdr(skb);
L
Linus Torvalds 已提交
234 235 236 237 238 239 240 241 242
	struct sock_exterr_skb *serr;

	if (!np->recverr)
		return;

	skb = skb_clone(skb, GFP_ATOMIC);
	if (!skb)
		return;

243 244
	skb->protocol = htons(ETH_P_IPV6);

L
Linus Torvalds 已提交
245 246 247
	serr = SKB_EXT_ERR(skb);
	serr->ee.ee_errno = err;
	serr->ee.ee_origin = SO_EE_ORIGIN_ICMP6;
248
	serr->ee.ee_type = icmph->icmp6_type;
L
Linus Torvalds 已提交
249 250 251 252
	serr->ee.ee_code = icmph->icmp6_code;
	serr->ee.ee_pad = 0;
	serr->ee.ee_info = info;
	serr->ee.ee_data = 0;
253 254
	serr->addr_offset = (u8 *)&(((struct ipv6hdr *)(icmph + 1))->daddr) -
				  skb_network_header(skb);
L
Linus Torvalds 已提交
255 256 257
	serr->port = port;

	__skb_pull(skb, payload - skb->data);
258
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
259 260 261 262 263

	if (sock_queue_err_skb(sk, skb))
		kfree_skb(skb);
}

264
void ipv6_local_error(struct sock *sk, int err, struct flowi6 *fl6, u32 info)
L
Linus Torvalds 已提交
265 266 267 268 269 270 271 272 273 274 275 276 277
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct sock_exterr_skb *serr;
	struct ipv6hdr *iph;
	struct sk_buff *skb;

	if (!np->recverr)
		return;

	skb = alloc_skb(sizeof(struct ipv6hdr), GFP_ATOMIC);
	if (!skb)
		return;

278 279
	skb->protocol = htons(ETH_P_IPV6);

280 281
	skb_put(skb, sizeof(struct ipv6hdr));
	skb_reset_network_header(skb);
282
	iph = ipv6_hdr(skb);
A
Alexey Dobriyan 已提交
283
	iph->daddr = fl6->daddr;
L
Linus Torvalds 已提交
284 285 286 287

	serr = SKB_EXT_ERR(skb);
	serr->ee.ee_errno = err;
	serr->ee.ee_origin = SO_EE_ORIGIN_LOCAL;
288
	serr->ee.ee_type = 0;
L
Linus Torvalds 已提交
289 290 291 292
	serr->ee.ee_code = 0;
	serr->ee.ee_pad = 0;
	serr->ee.ee_info = info;
	serr->ee.ee_data = 0;
293
	serr->addr_offset = (u8 *)&iph->daddr - skb_network_header(skb);
294
	serr->port = fl6->fl6_dport;
L
Linus Torvalds 已提交
295

296
	__skb_pull(skb, skb_tail_pointer(skb) - skb->data);
297
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
298 299 300 301 302

	if (sock_queue_err_skb(sk, skb))
		kfree_skb(skb);
}

303
void ipv6_local_rxpmtu(struct sock *sk, struct flowi6 *fl6, u32 mtu)
304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct ipv6hdr *iph;
	struct sk_buff *skb;
	struct ip6_mtuinfo *mtu_info;

	if (!np->rxopt.bits.rxpmtu)
		return;

	skb = alloc_skb(sizeof(struct ipv6hdr), GFP_ATOMIC);
	if (!skb)
		return;

	skb_put(skb, sizeof(struct ipv6hdr));
	skb_reset_network_header(skb);
	iph = ipv6_hdr(skb);
A
Alexey Dobriyan 已提交
320
	iph->daddr = fl6->daddr;
321 322 323 324 325 326 327

	mtu_info = IP6CBMTU(skb);

	mtu_info->ip6m_mtu = mtu;
	mtu_info->ip6m_addr.sin6_family = AF_INET6;
	mtu_info->ip6m_addr.sin6_port = 0;
	mtu_info->ip6m_addr.sin6_flowinfo = 0;
328
	mtu_info->ip6m_addr.sin6_scope_id = fl6->flowi6_oif;
A
Alexey Dobriyan 已提交
329
	mtu_info->ip6m_addr.sin6_addr = ipv6_hdr(skb)->daddr;
330 331 332 333 334 335 336 337

	__skb_pull(skb, skb_tail_pointer(skb) - skb->data);
	skb_reset_transport_header(skb);

	skb = xchg(&np->rxpmtu, skb);
	kfree_skb(skb);
}

338 339 340 341 342 343 344 345 346 347
/* For some errors we have valid addr_offset even with zero payload and
 * zero port. Also, addr_offset should be supported if port is set.
 */
static inline bool ipv6_datagram_support_addr(struct sock_exterr_skb *serr)
{
	return serr->ee.ee_origin == SO_EE_ORIGIN_ICMP6 ||
	       serr->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
	       serr->ee.ee_origin == SO_EE_ORIGIN_LOCAL || serr->port;
}

348 349 350 351 352 353 354 355 356 357 358
/* IPv6 supports cmsg on all origins aside from SO_EE_ORIGIN_LOCAL.
 *
 * At one point, excluding local errors was a quick test to identify icmp/icmp6
 * errors. This is no longer true, but the test remained, so the v6 stack,
 * unlike v4, also honors cmsg requests on all wifi and timestamp errors.
 *
 * Timestamp code paths do not initialize the fields expected by cmsg:
 * the PKTINFO fields in skb->cb[]. Fill those in here.
 */
static bool ip6_datagram_support_cmsg(struct sk_buff *skb,
				      struct sock_exterr_skb *serr)
359
{
360 361 362 363 364 365 366 367 368
	if (serr->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
	    serr->ee.ee_origin == SO_EE_ORIGIN_ICMP6)
		return true;

	if (serr->ee.ee_origin == SO_EE_ORIGIN_LOCAL)
		return false;

	if (!skb->dev)
		return false;
369 370

	if (skb->protocol == htons(ETH_P_IPV6))
371
		IP6CB(skb)->iif = skb->dev->ifindex;
372
	else
373 374 375
		PKTINFO_SKB_CB(skb)->ipi_ifindex = skb->dev->ifindex;

	return true;
376 377
}

378
/*
L
Linus Torvalds 已提交
379 380
 *	Handle MSG_ERRQUEUE
 */
381
int ipv6_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
L
Linus Torvalds 已提交
382 383 384
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct sock_exterr_skb *serr;
385
	struct sk_buff *skb;
386
	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin, msg->msg_name);
L
Linus Torvalds 已提交
387 388 389 390 391 392 393 394
	struct {
		struct sock_extended_err ee;
		struct sockaddr_in6	 offender;
	} errhdr;
	int err;
	int copied;

	err = -EAGAIN;
395
	skb = sock_dequeue_err_skb(sk);
396
	if (!skb)
L
Linus Torvalds 已提交
397 398 399 400 401 402 403
		goto out;

	copied = skb->len;
	if (copied > len) {
		msg->msg_flags |= MSG_TRUNC;
		copied = len;
	}
404
	err = skb_copy_datagram_msg(skb, 0, msg, copied);
L
Linus Torvalds 已提交
405 406 407 408 409 410 411
	if (err)
		goto out_free_skb;

	sock_recv_timestamp(msg, sk, skb);

	serr = SKB_EXT_ERR(skb);

412
	if (sin && ipv6_datagram_support_addr(serr)) {
413
		const unsigned char *nh = skb_network_header(skb);
L
Linus Torvalds 已提交
414 415
		sin->sin6_family = AF_INET6;
		sin->sin6_flowinfo = 0;
416
		sin->sin6_port = serr->port;
417
		if (skb->protocol == htons(ETH_P_IPV6)) {
418 419 420
			const struct ipv6hdr *ip6h = container_of((struct in6_addr *)(nh + serr->addr_offset),
								  struct ipv6hdr, daddr);
			sin->sin6_addr = ip6h->daddr;
L
Linus Torvalds 已提交
421
			if (np->sndflow)
422
				sin->sin6_flowinfo = ip6_flowinfo(ip6h);
423 424 425
			sin->sin6_scope_id =
				ipv6_iface_scope_id(&sin->sin6_addr,
						    IP6CB(skb)->iif);
L
Linus Torvalds 已提交
426
		} else {
B
Brian Haley 已提交
427 428
			ipv6_addr_set_v4mapped(*(__be32 *)(nh + serr->addr_offset),
					       &sin->sin6_addr);
429
			sin->sin6_scope_id = 0;
L
Linus Torvalds 已提交
430
		}
431
		*addr_len = sizeof(*sin);
L
Linus Torvalds 已提交
432 433 434 435
	}

	memcpy(&errhdr.ee, &serr->ee, sizeof(struct sock_extended_err));
	sin = &errhdr.offender;
436
	memset(sin, 0, sizeof(*sin));
437 438

	if (ip6_datagram_support_cmsg(skb, serr)) {
L
Linus Torvalds 已提交
439
		sin->sin6_family = AF_INET6;
440
		if (np->rxopt.all)
441
			ip6_datagram_recv_common_ctl(sk, msg, skb);
442
		if (skb->protocol == htons(ETH_P_IPV6)) {
A
Alexey Dobriyan 已提交
443
			sin->sin6_addr = ipv6_hdr(skb)->saddr;
L
Linus Torvalds 已提交
444
			if (np->rxopt.all)
445
				ip6_datagram_recv_specific_ctl(sk, msg, skb);
446 447 448
			sin->sin6_scope_id =
				ipv6_iface_scope_id(&sin->sin6_addr,
						    IP6CB(skb)->iif);
L
Linus Torvalds 已提交
449
		} else {
B
Brian Haley 已提交
450 451
			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
					       &sin->sin6_addr);
452
			if (inet_sk(sk)->cmsg_flags)
L
Linus Torvalds 已提交
453 454 455 456 457 458 459 460 461 462 463
				ip_cmsg_recv(msg, skb);
		}
	}

	put_cmsg(msg, SOL_IPV6, IPV6_RECVERR, sizeof(errhdr), &errhdr);

	/* Now we could try to dump offended packet options */

	msg->msg_flags |= MSG_ERRQUEUE;
	err = copied;

464
out_free_skb:
L
Linus Torvalds 已提交
465 466 467 468
	kfree_skb(skb);
out:
	return err;
}
469
EXPORT_SYMBOL_GPL(ipv6_recv_error);
L
Linus Torvalds 已提交
470

471 472 473
/*
 *	Handle IPV6_RECVPATHMTU
 */
474 475
int ipv6_recv_rxpmtu(struct sock *sk, struct msghdr *msg, int len,
		     int *addr_len)
476 477 478 479
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct sk_buff *skb;
	struct ip6_mtuinfo mtu_info;
480
	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin, msg->msg_name);
481 482 483 484 485
	int err;
	int copied;

	err = -EAGAIN;
	skb = xchg(&np->rxpmtu, NULL);
486
	if (!skb)
487 488 489 490 491 492 493
		goto out;

	copied = skb->len;
	if (copied > len) {
		msg->msg_flags |= MSG_TRUNC;
		copied = len;
	}
494
	err = skb_copy_datagram_msg(skb, 0, msg, copied);
495 496 497 498 499 500 501 502 503 504 505 506
	if (err)
		goto out_free_skb;

	sock_recv_timestamp(msg, sk, skb);

	memcpy(&mtu_info, IP6CBMTU(skb), sizeof(mtu_info));

	if (sin) {
		sin->sin6_family = AF_INET6;
		sin->sin6_flowinfo = 0;
		sin->sin6_port = 0;
		sin->sin6_scope_id = mtu_info.ip6m_addr.sin6_scope_id;
A
Alexey Dobriyan 已提交
507
		sin->sin6_addr = mtu_info.ip6m_addr.sin6_addr;
508
		*addr_len = sizeof(*sin);
509 510 511 512 513 514 515 516 517 518 519
	}

	put_cmsg(msg, SOL_IPV6, IPV6_PATHMTU, sizeof(mtu_info), &mtu_info);

	err = copied;

out_free_skb:
	kfree_skb(skb);
out:
	return err;
}
L
Linus Torvalds 已提交
520 521


522 523
void ip6_datagram_recv_common_ctl(struct sock *sk, struct msghdr *msg,
				 struct sk_buff *skb)
L
Linus Torvalds 已提交
524 525
{
	struct ipv6_pinfo *np = inet6_sk(sk);
526
	bool is_ipv6 = skb->protocol == htons(ETH_P_IPV6);
L
Linus Torvalds 已提交
527 528 529 530

	if (np->rxopt.bits.rxinfo) {
		struct in6_pktinfo src_info;

531 532 533 534 535 536 537 538 539
		if (is_ipv6) {
			src_info.ipi6_ifindex = IP6CB(skb)->iif;
			src_info.ipi6_addr = ipv6_hdr(skb)->daddr;
		} else {
			src_info.ipi6_ifindex =
				PKTINFO_SKB_CB(skb)->ipi_ifindex;
			ipv6_addr_set_v4mapped(ip_hdr(skb)->daddr,
					       &src_info.ipi6_addr);
		}
540 541 542 543

		if (src_info.ipi6_ifindex >= 0)
			put_cmsg(msg, SOL_IPV6, IPV6_PKTINFO,
				 sizeof(src_info), &src_info);
L
Linus Torvalds 已提交
544
	}
545 546 547 548 549 550 551 552
}

void ip6_datagram_recv_specific_ctl(struct sock *sk, struct msghdr *msg,
				    struct sk_buff *skb)
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct inet6_skb_parm *opt = IP6CB(skb);
	unsigned char *nh = skb_network_header(skb);
L
Linus Torvalds 已提交
553 554

	if (np->rxopt.bits.rxhlim) {
555
		int hlim = ipv6_hdr(skb)->hop_limit;
L
Linus Torvalds 已提交
556 557 558
		put_cmsg(msg, SOL_IPV6, IPV6_HOPLIMIT, sizeof(hlim), &hlim);
	}

559
	if (np->rxopt.bits.rxtclass) {
560
		int tclass = ipv6_get_dsfield(ipv6_hdr(skb));
561 562 563
		put_cmsg(msg, SOL_IPV6, IPV6_TCLASS, sizeof(tclass), &tclass);
	}

564 565 566 567
	if (np->rxopt.bits.rxflow) {
		__be32 flowinfo = ip6_flowinfo((struct ipv6hdr *)nh);
		if (flowinfo)
			put_cmsg(msg, SOL_IPV6, IPV6_FLOWINFO, sizeof(flowinfo), &flowinfo);
L
Linus Torvalds 已提交
568
	}
569 570

	/* HbH is allowed only once */
571 572
	if (np->rxopt.bits.hopopts && (opt->flags & IP6SKB_HOPBYHOP)) {
		u8 *ptr = nh + sizeof(struct ipv6hdr);
L
Linus Torvalds 已提交
573 574
		put_cmsg(msg, SOL_IPV6, IPV6_HOPOPTS, (ptr[1]+1)<<3, ptr);
	}
575 576 577 578 579 580 581 582

	if (opt->lastopt &&
	    (np->rxopt.bits.dstopts || np->rxopt.bits.srcrt)) {
		/*
		 * Silly enough, but we need to reparse in order to
		 * report extension headers (except for HbH)
		 * in order.
		 *
583
		 * Also note that IPV6_RECVRTHDRDSTOPTS is NOT
584 585 586 587
		 * (and WILL NOT be) defined because
		 * IPV6_RECVDSTOPTS is more generic. --yoshfuji
		 */
		unsigned int off = sizeof(struct ipv6hdr);
588
		u8 nexthdr = ipv6_hdr(skb)->nexthdr;
589 590

		while (off <= opt->lastopt) {
591
			unsigned int len;
592
			u8 *ptr = nh + off;
593

594
			switch (nexthdr) {
595 596 597 598 599 600 601 602 603 604 605 606 607 608
			case IPPROTO_DSTOPTS:
				nexthdr = ptr[0];
				len = (ptr[1] + 1) << 3;
				if (np->rxopt.bits.dstopts)
					put_cmsg(msg, SOL_IPV6, IPV6_DSTOPTS, len, ptr);
				break;
			case IPPROTO_ROUTING:
				nexthdr = ptr[0];
				len = (ptr[1] + 1) << 3;
				if (np->rxopt.bits.srcrt)
					put_cmsg(msg, SOL_IPV6, IPV6_RTHDR, len, ptr);
				break;
			case IPPROTO_AH:
				nexthdr = ptr[0];
609
				len = (ptr[1] + 2) << 2;
610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625
				break;
			default:
				nexthdr = ptr[0];
				len = (ptr[1] + 1) << 3;
				break;
			}

			off += len;
		}
	}

	/* socket options in old style */
	if (np->rxopt.bits.rxoinfo) {
		struct in6_pktinfo src_info;

		src_info.ipi6_ifindex = opt->iif;
A
Alexey Dobriyan 已提交
626
		src_info.ipi6_addr = ipv6_hdr(skb)->daddr;
627 628 629
		put_cmsg(msg, SOL_IPV6, IPV6_2292PKTINFO, sizeof(src_info), &src_info);
	}
	if (np->rxopt.bits.rxohlim) {
630
		int hlim = ipv6_hdr(skb)->hop_limit;
631 632
		put_cmsg(msg, SOL_IPV6, IPV6_2292HOPLIMIT, sizeof(hlim), &hlim);
	}
633 634
	if (np->rxopt.bits.ohopopts && (opt->flags & IP6SKB_HOPBYHOP)) {
		u8 *ptr = nh + sizeof(struct ipv6hdr);
635 636 637
		put_cmsg(msg, SOL_IPV6, IPV6_2292HOPOPTS, (ptr[1]+1)<<3, ptr);
	}
	if (np->rxopt.bits.odstopts && opt->dst0) {
638
		u8 *ptr = nh + opt->dst0;
639
		put_cmsg(msg, SOL_IPV6, IPV6_2292DSTOPTS, (ptr[1]+1)<<3, ptr);
L
Linus Torvalds 已提交
640
	}
641
	if (np->rxopt.bits.osrcrt && opt->srcrt) {
642
		struct ipv6_rt_hdr *rthdr = (struct ipv6_rt_hdr *)(nh + opt->srcrt);
643
		put_cmsg(msg, SOL_IPV6, IPV6_2292RTHDR, (rthdr->hdrlen+1) << 3, rthdr);
L
Linus Torvalds 已提交
644
	}
645
	if (np->rxopt.bits.odstopts && opt->dst1) {
646
		u8 *ptr = nh + opt->dst1;
647
		put_cmsg(msg, SOL_IPV6, IPV6_2292DSTOPTS, (ptr[1]+1)<<3, ptr);
L
Linus Torvalds 已提交
648
	}
649 650
	if (np->rxopt.bits.rxorigdstaddr) {
		struct sockaddr_in6 sin6;
E
Eric Dumazet 已提交
651
		__be16 *ports = (__be16 *) skb_transport_header(skb);
652 653 654 655 656 657 658 659

		if (skb_transport_offset(skb) + 4 <= skb->len) {
			/* All current transport protocols have the port numbers in the
			 * first four bytes of the transport header and this function is
			 * written with this assumption in mind.
			 */

			sin6.sin6_family = AF_INET6;
A
Alexey Dobriyan 已提交
660
			sin6.sin6_addr = ipv6_hdr(skb)->daddr;
661 662
			sin6.sin6_port = ports[1];
			sin6.sin6_flowinfo = 0;
663 664 665
			sin6.sin6_scope_id =
				ipv6_iface_scope_id(&ipv6_hdr(skb)->daddr,
						    opt->iif);
666 667 668 669

			put_cmsg(msg, SOL_IPV6, IPV6_ORIGDSTADDR, sizeof(sin6), &sin6);
		}
	}
670 671 672 673 674 675 676
}

void ip6_datagram_recv_ctl(struct sock *sk, struct msghdr *msg,
			  struct sk_buff *skb)
{
	ip6_datagram_recv_common_ctl(sk, msg, skb);
	ip6_datagram_recv_specific_ctl(sk, msg, skb);
L
Linus Torvalds 已提交
677
}
T
Tom Parkin 已提交
678
EXPORT_SYMBOL_GPL(ip6_datagram_recv_ctl);
L
Linus Torvalds 已提交
679

680 681 682 683
int ip6_datagram_send_ctl(struct net *net, struct sock *sk,
			  struct msghdr *msg, struct flowi6 *fl6,
			  struct ipv6_txoptions *opt,
			  int *hlimit, int *tclass, int *dontfrag)
L
Linus Torvalds 已提交
684 685 686 687 688 689 690 691
{
	struct in6_pktinfo *src_info;
	struct cmsghdr *cmsg;
	struct ipv6_rt_hdr *rthdr;
	struct ipv6_opt_hdr *hdr;
	int len;
	int err = 0;

692
	for_each_cmsghdr(cmsg, msg) {
L
Linus Torvalds 已提交
693 694 695 696 697 698 699 700 701 702 703
		int addr_type;

		if (!CMSG_OK(msg, cmsg)) {
			err = -EINVAL;
			goto exit_f;
		}

		if (cmsg->cmsg_level != SOL_IPV6)
			continue;

		switch (cmsg->cmsg_type) {
704 705
		case IPV6_PKTINFO:
		case IPV6_2292PKTINFO:
706 707 708
		    {
			struct net_device *dev = NULL;

709
			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct in6_pktinfo))) {
L
Linus Torvalds 已提交
710 711 712 713 714
				err = -EINVAL;
				goto exit_f;
			}

			src_info = (struct in6_pktinfo *)CMSG_DATA(cmsg);
715

L
Linus Torvalds 已提交
716
			if (src_info->ipi6_ifindex) {
717 718
				if (fl6->flowi6_oif &&
				    src_info->ipi6_ifindex != fl6->flowi6_oif)
L
Linus Torvalds 已提交
719
					return -EINVAL;
720
				fl6->flowi6_oif = src_info->ipi6_ifindex;
L
Linus Torvalds 已提交
721 722
			}

723
			addr_type = __ipv6_addr_type(&src_info->ipi6_addr);
L
Linus Torvalds 已提交
724

725
			rcu_read_lock();
726 727
			if (fl6->flowi6_oif) {
				dev = dev_get_by_index_rcu(net, fl6->flowi6_oif);
728 729
				if (!dev) {
					rcu_read_unlock();
730
					return -ENODEV;
731 732 733
				}
			} else if (addr_type & IPV6_ADDR_LINKLOCAL) {
				rcu_read_unlock();
734
				return -EINVAL;
735
			}
736

737 738
			if (addr_type != IPV6_ADDR_ANY) {
				int strict = __ipv6_addr_src_scope(addr_type) <= IPV6_ADDR_SCOPE_LINKLOCAL;
739
				if (!(inet_sk(sk)->freebind || inet_sk(sk)->transparent) &&
740
				    !ipv6_chk_addr(net, &src_info->ipi6_addr,
741 742 743
						   strict ? dev : NULL, 0) &&
				    !ipv6_chk_acast_addr_src(net, dev,
							     &src_info->ipi6_addr))
744 745
					err = -EINVAL;
				else
A
Alexey Dobriyan 已提交
746
					fl6->saddr = src_info->ipi6_addr;
L
Linus Torvalds 已提交
747
			}
748

749
			rcu_read_unlock();
L
Linus Torvalds 已提交
750

751 752 753
			if (err)
				goto exit_f;

L
Linus Torvalds 已提交
754
			break;
755
		    }
L
Linus Torvalds 已提交
756 757

		case IPV6_FLOWINFO:
758
			if (cmsg->cmsg_len < CMSG_LEN(4)) {
L
Linus Torvalds 已提交
759 760 761 762
				err = -EINVAL;
				goto exit_f;
			}

763 764
			if (fl6->flowlabel&IPV6_FLOWINFO_MASK) {
				if ((fl6->flowlabel^*(__be32 *)CMSG_DATA(cmsg))&~IPV6_FLOWINFO_MASK) {
L
Linus Torvalds 已提交
765 766 767 768
					err = -EINVAL;
					goto exit_f;
				}
			}
769
			fl6->flowlabel = IPV6_FLOWINFO_MASK & *(__be32 *)CMSG_DATA(cmsg);
L
Linus Torvalds 已提交
770 771
			break;

772
		case IPV6_2292HOPOPTS:
L
Linus Torvalds 已提交
773
		case IPV6_HOPOPTS:
774
			if (opt->hopopt || cmsg->cmsg_len < CMSG_LEN(sizeof(struct ipv6_opt_hdr))) {
L
Linus Torvalds 已提交
775 776 777 778 779 780 781 782 783 784
				err = -EINVAL;
				goto exit_f;
			}

			hdr = (struct ipv6_opt_hdr *)CMSG_DATA(cmsg);
			len = ((hdr->hdrlen + 1) << 3);
			if (cmsg->cmsg_len < CMSG_LEN(len)) {
				err = -EINVAL;
				goto exit_f;
			}
785
			if (!ns_capable(net->user_ns, CAP_NET_RAW)) {
L
Linus Torvalds 已提交
786 787 788 789 790 791 792
				err = -EPERM;
				goto exit_f;
			}
			opt->opt_nflen += len;
			opt->hopopt = hdr;
			break;

793
		case IPV6_2292DSTOPTS:
794
			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct ipv6_opt_hdr))) {
L
Linus Torvalds 已提交
795 796 797 798 799 800 801 802 803 804
				err = -EINVAL;
				goto exit_f;
			}

			hdr = (struct ipv6_opt_hdr *)CMSG_DATA(cmsg);
			len = ((hdr->hdrlen + 1) << 3);
			if (cmsg->cmsg_len < CMSG_LEN(len)) {
				err = -EINVAL;
				goto exit_f;
			}
805
			if (!ns_capable(net->user_ns, CAP_NET_RAW)) {
L
Linus Torvalds 已提交
806 807 808 809 810 811 812 813 814 815 816
				err = -EPERM;
				goto exit_f;
			}
			if (opt->dst1opt) {
				err = -EINVAL;
				goto exit_f;
			}
			opt->opt_flen += len;
			opt->dst1opt = hdr;
			break;

817 818 819 820 821 822 823 824 825 826 827 828 829
		case IPV6_DSTOPTS:
		case IPV6_RTHDRDSTOPTS:
			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct ipv6_opt_hdr))) {
				err = -EINVAL;
				goto exit_f;
			}

			hdr = (struct ipv6_opt_hdr *)CMSG_DATA(cmsg);
			len = ((hdr->hdrlen + 1) << 3);
			if (cmsg->cmsg_len < CMSG_LEN(len)) {
				err = -EINVAL;
				goto exit_f;
			}
830
			if (!ns_capable(net->user_ns, CAP_NET_RAW)) {
831 832 833 834 835 836 837 838 839 840 841 842 843
				err = -EPERM;
				goto exit_f;
			}
			if (cmsg->cmsg_type == IPV6_DSTOPTS) {
				opt->opt_flen += len;
				opt->dst1opt = hdr;
			} else {
				opt->opt_nflen += len;
				opt->dst0opt = hdr;
			}
			break;

		case IPV6_2292RTHDR:
L
Linus Torvalds 已提交
844
		case IPV6_RTHDR:
845
			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct ipv6_rt_hdr))) {
L
Linus Torvalds 已提交
846 847 848 849 850 851
				err = -EINVAL;
				goto exit_f;
			}

			rthdr = (struct ipv6_rt_hdr *)CMSG_DATA(cmsg);

852
			switch (rthdr->type) {
A
Amerigo Wang 已提交
853
#if IS_ENABLED(CONFIG_IPV6_MIP6)
854
			case IPV6_SRCRT_TYPE_2:
B
Brian Haley 已提交
855 856 857 858 859
				if (rthdr->hdrlen != 2 ||
				    rthdr->segments_left != 1) {
					err = -EINVAL;
					goto exit_f;
				}
860
				break;
861
#endif
862
			default:
L
Linus Torvalds 已提交
863 864 865 866 867 868
				err = -EINVAL;
				goto exit_f;
			}

			len = ((rthdr->hdrlen + 1) << 3);

869
			if (cmsg->cmsg_len < CMSG_LEN(len)) {
L
Linus Torvalds 已提交
870 871 872 873 874 875 876 877 878 879 880 881 882
				err = -EINVAL;
				goto exit_f;
			}

			/* segments left must also match */
			if ((rthdr->hdrlen >> 1) != rthdr->segments_left) {
				err = -EINVAL;
				goto exit_f;
			}

			opt->opt_nflen += len;
			opt->srcrt = rthdr;

883
			if (cmsg->cmsg_type == IPV6_2292RTHDR && opt->dst1opt) {
L
Linus Torvalds 已提交
884 885 886 887 888 889 890 891 892 893
				int dsthdrlen = ((opt->dst1opt->hdrlen+1)<<3);

				opt->opt_nflen += dsthdrlen;
				opt->dst0opt = opt->dst1opt;
				opt->dst1opt = NULL;
				opt->opt_flen -= dsthdrlen;
			}

			break;

894
		case IPV6_2292HOPLIMIT:
L
Linus Torvalds 已提交
895 896 897 898 899 900 901
		case IPV6_HOPLIMIT:
			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int))) {
				err = -EINVAL;
				goto exit_f;
			}

			*hlimit = *(int *)CMSG_DATA(cmsg);
902 903 904 905 906
			if (*hlimit < -1 || *hlimit > 0xff) {
				err = -EINVAL;
				goto exit_f;
			}

L
Linus Torvalds 已提交
907 908
			break;

909 910 911 912 913
		case IPV6_TCLASS:
		    {
			int tc;

			err = -EINVAL;
914
			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int)))
915 916 917
				goto exit_f;

			tc = *(int *)CMSG_DATA(cmsg);
918
			if (tc < -1 || tc > 0xff)
919 920 921 922 923
				goto exit_f;

			err = 0;
			*tclass = tc;

924 925 926 927 928 929 930 931
			break;
		    }

		case IPV6_DONTFRAG:
		    {
			int df;

			err = -EINVAL;
932
			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int)))
933 934 935 936 937 938 939 940 941
				goto exit_f;

			df = *(int *)CMSG_DATA(cmsg);
			if (df < 0 || df > 1)
				goto exit_f;

			err = 0;
			*dontfrag = df;

942 943
			break;
		    }
L
Linus Torvalds 已提交
944
		default:
945 946
			net_dbg_ratelimited("invalid cmsg type: %d\n",
					    cmsg->cmsg_type);
L
Linus Torvalds 已提交
947
			err = -EINVAL;
948
			goto exit_f;
949
		}
L
Linus Torvalds 已提交
950 951 952 953 954
	}

exit_f:
	return err;
}
955
EXPORT_SYMBOL_GPL(ip6_datagram_send_ctl);
956 957 958 959 960 961

void ip6_dgram_sock_seq_show(struct seq_file *seq, struct sock *sp,
			     __u16 srcp, __u16 destp, int bucket)
{
	const struct in6_addr *dest, *src;

962 963
	dest  = &sp->sk_v6_daddr;
	src   = &sp->sk_v6_rcv_saddr;
964 965
	seq_printf(seq,
		   "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
966
		   "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d\n",
967 968 969 970 971 972 973 974 975 976 977 978 979 980 981
		   bucket,
		   src->s6_addr32[0], src->s6_addr32[1],
		   src->s6_addr32[2], src->s6_addr32[3], srcp,
		   dest->s6_addr32[0], dest->s6_addr32[1],
		   dest->s6_addr32[2], dest->s6_addr32[3], destp,
		   sp->sk_state,
		   sk_wmem_alloc_get(sp),
		   sk_rmem_alloc_get(sp),
		   0, 0L, 0,
		   from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
		   0,
		   sock_i_ino(sp),
		   atomic_read(&sp->sk_refcnt), sp,
		   atomic_read(&sp->sk_drops));
}