datagram.c 21.8 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
}

L
Linus Torvalds 已提交
43 44 45
int ip6_datagram_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
{
	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 59 60 61 62 63
	int			addr_type;
	int			err;

	if (usin->sin6_family == AF_INET) {
		if (__ipv6_only_sock(sk))
			return -EAFNOSUPPORT;
		err = ip4_datagram_connect(sk, uaddr, addr_len);
		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);
L
Linus Torvalds 已提交
74 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
			if (flowlabel == NULL)
				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
		err = ip4_datagram_connect(sk,
102
					   (struct sockaddr *) &sin,
L
Linus Torvalds 已提交
103 104 105 106 107
					   sizeof(sin));

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
	ip6_set_txhash(sk);
L
Linus Torvalds 已提交
203 204 205 206
out:
	fl6_sock_release(flowlabel);
	return err;
}
207
EXPORT_SYMBOL_GPL(ip6_datagram_connect);
L
Linus Torvalds 已提交
208

209 210 211 212 213 214 215 216 217 218
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);

219
void ipv6_icmp_error(struct sock *sk, struct sk_buff *skb, int err,
A
Al Viro 已提交
220
		     __be16 port, u32 info, u8 *payload)
L
Linus Torvalds 已提交
221 222
{
	struct ipv6_pinfo *np  = inet6_sk(sk);
223
	struct icmp6hdr *icmph = icmp6_hdr(skb);
L
Linus Torvalds 已提交
224 225 226 227 228 229 230 231 232
	struct sock_exterr_skb *serr;

	if (!np->recverr)
		return;

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

233 234
	skb->protocol = htons(ETH_P_IPV6);

L
Linus Torvalds 已提交
235 236 237
	serr = SKB_EXT_ERR(skb);
	serr->ee.ee_errno = err;
	serr->ee.ee_origin = SO_EE_ORIGIN_ICMP6;
238
	serr->ee.ee_type = icmph->icmp6_type;
L
Linus Torvalds 已提交
239 240 241 242
	serr->ee.ee_code = icmph->icmp6_code;
	serr->ee.ee_pad = 0;
	serr->ee.ee_info = info;
	serr->ee.ee_data = 0;
243 244
	serr->addr_offset = (u8 *)&(((struct ipv6hdr *)(icmph + 1))->daddr) -
				  skb_network_header(skb);
L
Linus Torvalds 已提交
245 246 247
	serr->port = port;

	__skb_pull(skb, payload - skb->data);
248
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
249 250 251 252 253

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

254
void ipv6_local_error(struct sock *sk, int err, struct flowi6 *fl6, u32 info)
L
Linus Torvalds 已提交
255 256 257 258 259 260 261 262 263 264 265 266 267
{
	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;

268 269
	skb->protocol = htons(ETH_P_IPV6);

270 271
	skb_put(skb, sizeof(struct ipv6hdr));
	skb_reset_network_header(skb);
272
	iph = ipv6_hdr(skb);
A
Alexey Dobriyan 已提交
273
	iph->daddr = fl6->daddr;
L
Linus Torvalds 已提交
274 275 276 277

	serr = SKB_EXT_ERR(skb);
	serr->ee.ee_errno = err;
	serr->ee.ee_origin = SO_EE_ORIGIN_LOCAL;
278
	serr->ee.ee_type = 0;
L
Linus Torvalds 已提交
279 280 281 282
	serr->ee.ee_code = 0;
	serr->ee.ee_pad = 0;
	serr->ee.ee_info = info;
	serr->ee.ee_data = 0;
283
	serr->addr_offset = (u8 *)&iph->daddr - skb_network_header(skb);
284
	serr->port = fl6->fl6_dport;
L
Linus Torvalds 已提交
285

286
	__skb_pull(skb, skb_tail_pointer(skb) - skb->data);
287
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
288 289 290 291 292

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

293
void ipv6_local_rxpmtu(struct sock *sk, struct flowi6 *fl6, u32 mtu)
294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309
{
	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
310
	iph->daddr = fl6->daddr;
311 312 313 314 315 316 317

	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;
318
	mtu_info->ip6m_addr.sin6_scope_id = fl6->flowi6_oif;
A
Alexey Dobriyan 已提交
319
	mtu_info->ip6m_addr.sin6_addr = ipv6_hdr(skb)->daddr;
320 321 322 323 324 325 326 327

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

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

328
/*
L
Linus Torvalds 已提交
329 330
 *	Handle MSG_ERRQUEUE
 */
331
int ipv6_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
L
Linus Torvalds 已提交
332 333 334
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct sock_exterr_skb *serr;
335
	struct sk_buff *skb;
336
	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin, msg->msg_name);
L
Linus Torvalds 已提交
337 338 339 340 341 342 343 344
	struct {
		struct sock_extended_err ee;
		struct sockaddr_in6	 offender;
	} errhdr;
	int err;
	int copied;

	err = -EAGAIN;
345
	skb = sock_dequeue_err_skb(sk);
L
Linus Torvalds 已提交
346 347 348 349 350 351 352 353
	if (skb == NULL)
		goto out;

	copied = skb->len;
	if (copied > len) {
		msg->msg_flags |= MSG_TRUNC;
		copied = len;
	}
354
	err = skb_copy_datagram_msg(skb, 0, msg, copied);
L
Linus Torvalds 已提交
355 356 357 358 359 360 361 362
	if (err)
		goto out_free_skb;

	sock_recv_timestamp(msg, sk, skb);

	serr = SKB_EXT_ERR(skb);

	if (sin) {
363
		const unsigned char *nh = skb_network_header(skb);
L
Linus Torvalds 已提交
364 365
		sin->sin6_family = AF_INET6;
		sin->sin6_flowinfo = 0;
366
		sin->sin6_port = serr->port;
367
		if (skb->protocol == htons(ETH_P_IPV6)) {
368 369 370
			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
371
			if (np->sndflow)
372
				sin->sin6_flowinfo = ip6_flowinfo(ip6h);
373 374 375
			sin->sin6_scope_id =
				ipv6_iface_scope_id(&sin->sin6_addr,
						    IP6CB(skb)->iif);
L
Linus Torvalds 已提交
376
		} else {
B
Brian Haley 已提交
377 378
			ipv6_addr_set_v4mapped(*(__be32 *)(nh + serr->addr_offset),
					       &sin->sin6_addr);
379
			sin->sin6_scope_id = 0;
L
Linus Torvalds 已提交
380
		}
381
		*addr_len = sizeof(*sin);
L
Linus Torvalds 已提交
382 383 384 385 386 387 388 389
	}

	memcpy(&errhdr.ee, &serr->ee, sizeof(struct sock_extended_err));
	sin = &errhdr.offender;
	sin->sin6_family = AF_UNSPEC;
	if (serr->ee.ee_origin != SO_EE_ORIGIN_LOCAL) {
		sin->sin6_family = AF_INET6;
		sin->sin6_flowinfo = 0;
390
		sin->sin6_port = 0;
391 392
		if (np->rxopt.all)
			ip6_datagram_recv_common_ctl(sk, msg, skb);
393
		if (skb->protocol == htons(ETH_P_IPV6)) {
A
Alexey Dobriyan 已提交
394
			sin->sin6_addr = ipv6_hdr(skb)->saddr;
L
Linus Torvalds 已提交
395
			if (np->rxopt.all)
396
				ip6_datagram_recv_specific_ctl(sk, msg, skb);
397 398 399
			sin->sin6_scope_id =
				ipv6_iface_scope_id(&sin->sin6_addr,
						    IP6CB(skb)->iif);
L
Linus Torvalds 已提交
400 401 402
		} else {
			struct inet_sock *inet = inet_sk(sk);

B
Brian Haley 已提交
403 404
			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
					       &sin->sin6_addr);
405
			sin->sin6_scope_id = 0;
L
Linus Torvalds 已提交
406 407 408 409 410 411 412 413 414 415 416 417
			if (inet->cmsg_flags)
				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;

418
out_free_skb:
L
Linus Torvalds 已提交
419 420 421 422
	kfree_skb(skb);
out:
	return err;
}
423
EXPORT_SYMBOL_GPL(ipv6_recv_error);
L
Linus Torvalds 已提交
424

425 426 427
/*
 *	Handle IPV6_RECVPATHMTU
 */
428 429
int ipv6_recv_rxpmtu(struct sock *sk, struct msghdr *msg, int len,
		     int *addr_len)
430 431 432 433
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct sk_buff *skb;
	struct ip6_mtuinfo mtu_info;
434
	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin, msg->msg_name);
435 436 437 438 439 440 441 442 443 444 445 446 447
	int err;
	int copied;

	err = -EAGAIN;
	skb = xchg(&np->rxpmtu, NULL);
	if (skb == NULL)
		goto out;

	copied = skb->len;
	if (copied > len) {
		msg->msg_flags |= MSG_TRUNC;
		copied = len;
	}
448
	err = skb_copy_datagram_msg(skb, 0, msg, copied);
449 450 451 452 453 454 455 456 457 458 459 460
	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
461
		sin->sin6_addr = mtu_info.ip6m_addr.sin6_addr;
462
		*addr_len = sizeof(*sin);
463 464 465 466 467 468 469 470 471 472 473
	}

	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
474 475


476 477
void ip6_datagram_recv_common_ctl(struct sock *sk, struct msghdr *msg,
				 struct sk_buff *skb)
L
Linus Torvalds 已提交
478 479
{
	struct ipv6_pinfo *np = inet6_sk(sk);
480
	bool is_ipv6 = skb->protocol == htons(ETH_P_IPV6);
L
Linus Torvalds 已提交
481 482 483 484

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

485 486 487 488 489 490 491 492 493
		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);
		}
L
Linus Torvalds 已提交
494 495
		put_cmsg(msg, SOL_IPV6, IPV6_PKTINFO, sizeof(src_info), &src_info);
	}
496 497 498 499 500 501 502 503
}

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
504 505

	if (np->rxopt.bits.rxhlim) {
506
		int hlim = ipv6_hdr(skb)->hop_limit;
L
Linus Torvalds 已提交
507 508 509
		put_cmsg(msg, SOL_IPV6, IPV6_HOPLIMIT, sizeof(hlim), &hlim);
	}

510
	if (np->rxopt.bits.rxtclass) {
511
		int tclass = ipv6_get_dsfield(ipv6_hdr(skb));
512 513 514
		put_cmsg(msg, SOL_IPV6, IPV6_TCLASS, sizeof(tclass), &tclass);
	}

515 516 517 518
	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
519
	}
520 521

	/* HbH is allowed only once */
L
Linus Torvalds 已提交
522
	if (np->rxopt.bits.hopopts && opt->hop) {
523
		u8 *ptr = nh + opt->hop;
L
Linus Torvalds 已提交
524 525
		put_cmsg(msg, SOL_IPV6, IPV6_HOPOPTS, (ptr[1]+1)<<3, ptr);
	}
526 527 528 529 530 531 532 533

	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.
		 *
534
		 * Also note that IPV6_RECVRTHDRDSTOPTS is NOT
535 536 537 538
		 * (and WILL NOT be) defined because
		 * IPV6_RECVDSTOPTS is more generic. --yoshfuji
		 */
		unsigned int off = sizeof(struct ipv6hdr);
539
		u8 nexthdr = ipv6_hdr(skb)->nexthdr;
540 541

		while (off <= opt->lastopt) {
542
			unsigned int len;
543
			u8 *ptr = nh + off;
544

545
			switch (nexthdr) {
546 547 548 549 550 551 552 553 554 555 556 557 558 559
			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];
560
				len = (ptr[1] + 2) << 2;
561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576
				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
577
		src_info.ipi6_addr = ipv6_hdr(skb)->daddr;
578 579 580
		put_cmsg(msg, SOL_IPV6, IPV6_2292PKTINFO, sizeof(src_info), &src_info);
	}
	if (np->rxopt.bits.rxohlim) {
581
		int hlim = ipv6_hdr(skb)->hop_limit;
582 583 584
		put_cmsg(msg, SOL_IPV6, IPV6_2292HOPLIMIT, sizeof(hlim), &hlim);
	}
	if (np->rxopt.bits.ohopopts && opt->hop) {
585
		u8 *ptr = nh + opt->hop;
586 587 588
		put_cmsg(msg, SOL_IPV6, IPV6_2292HOPOPTS, (ptr[1]+1)<<3, ptr);
	}
	if (np->rxopt.bits.odstopts && opt->dst0) {
589
		u8 *ptr = nh + opt->dst0;
590
		put_cmsg(msg, SOL_IPV6, IPV6_2292DSTOPTS, (ptr[1]+1)<<3, ptr);
L
Linus Torvalds 已提交
591
	}
592
	if (np->rxopt.bits.osrcrt && opt->srcrt) {
593
		struct ipv6_rt_hdr *rthdr = (struct ipv6_rt_hdr *)(nh + opt->srcrt);
594
		put_cmsg(msg, SOL_IPV6, IPV6_2292RTHDR, (rthdr->hdrlen+1) << 3, rthdr);
L
Linus Torvalds 已提交
595
	}
596
	if (np->rxopt.bits.odstopts && opt->dst1) {
597
		u8 *ptr = nh + opt->dst1;
598
		put_cmsg(msg, SOL_IPV6, IPV6_2292DSTOPTS, (ptr[1]+1)<<3, ptr);
L
Linus Torvalds 已提交
599
	}
600 601
	if (np->rxopt.bits.rxorigdstaddr) {
		struct sockaddr_in6 sin6;
E
Eric Dumazet 已提交
602
		__be16 *ports = (__be16 *) skb_transport_header(skb);
603 604 605 606 607 608 609 610

		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
611
			sin6.sin6_addr = ipv6_hdr(skb)->daddr;
612 613
			sin6.sin6_port = ports[1];
			sin6.sin6_flowinfo = 0;
614 615 616
			sin6.sin6_scope_id =
				ipv6_iface_scope_id(&ipv6_hdr(skb)->daddr,
						    opt->iif);
617 618 619 620

			put_cmsg(msg, SOL_IPV6, IPV6_ORIGDSTADDR, sizeof(sin6), &sin6);
		}
	}
621 622 623 624 625 626 627
}

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
628
}
T
Tom Parkin 已提交
629
EXPORT_SYMBOL_GPL(ip6_datagram_recv_ctl);
L
Linus Torvalds 已提交
630

631 632 633 634
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
635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654
{
	struct in6_pktinfo *src_info;
	struct cmsghdr *cmsg;
	struct ipv6_rt_hdr *rthdr;
	struct ipv6_opt_hdr *hdr;
	int len;
	int err = 0;

	for (cmsg = CMSG_FIRSTHDR(msg); cmsg; cmsg = CMSG_NXTHDR(msg, cmsg)) {
		int addr_type;

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

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

		switch (cmsg->cmsg_type) {
655 656
		case IPV6_PKTINFO:
		case IPV6_2292PKTINFO:
657 658 659
		    {
			struct net_device *dev = NULL;

660
			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct in6_pktinfo))) {
L
Linus Torvalds 已提交
661 662 663 664 665
				err = -EINVAL;
				goto exit_f;
			}

			src_info = (struct in6_pktinfo *)CMSG_DATA(cmsg);
666

L
Linus Torvalds 已提交
667
			if (src_info->ipi6_ifindex) {
668 669
				if (fl6->flowi6_oif &&
				    src_info->ipi6_ifindex != fl6->flowi6_oif)
L
Linus Torvalds 已提交
670
					return -EINVAL;
671
				fl6->flowi6_oif = src_info->ipi6_ifindex;
L
Linus Torvalds 已提交
672 673
			}

674
			addr_type = __ipv6_addr_type(&src_info->ipi6_addr);
L
Linus Torvalds 已提交
675

676
			rcu_read_lock();
677 678
			if (fl6->flowi6_oif) {
				dev = dev_get_by_index_rcu(net, fl6->flowi6_oif);
679 680
				if (!dev) {
					rcu_read_unlock();
681
					return -ENODEV;
682 683 684
				}
			} else if (addr_type & IPV6_ADDR_LINKLOCAL) {
				rcu_read_unlock();
685
				return -EINVAL;
686
			}
687

688 689
			if (addr_type != IPV6_ADDR_ANY) {
				int strict = __ipv6_addr_src_scope(addr_type) <= IPV6_ADDR_SCOPE_LINKLOCAL;
690
				if (!(inet_sk(sk)->freebind || inet_sk(sk)->transparent) &&
691
				    !ipv6_chk_addr(net, &src_info->ipi6_addr,
692 693 694
						   strict ? dev : NULL, 0) &&
				    !ipv6_chk_acast_addr_src(net, dev,
							     &src_info->ipi6_addr))
695 696
					err = -EINVAL;
				else
A
Alexey Dobriyan 已提交
697
					fl6->saddr = src_info->ipi6_addr;
L
Linus Torvalds 已提交
698
			}
699

700
			rcu_read_unlock();
L
Linus Torvalds 已提交
701

702 703 704
			if (err)
				goto exit_f;

L
Linus Torvalds 已提交
705
			break;
706
		    }
L
Linus Torvalds 已提交
707 708

		case IPV6_FLOWINFO:
709
			if (cmsg->cmsg_len < CMSG_LEN(4)) {
L
Linus Torvalds 已提交
710 711 712 713
				err = -EINVAL;
				goto exit_f;
			}

714 715
			if (fl6->flowlabel&IPV6_FLOWINFO_MASK) {
				if ((fl6->flowlabel^*(__be32 *)CMSG_DATA(cmsg))&~IPV6_FLOWINFO_MASK) {
L
Linus Torvalds 已提交
716 717 718 719
					err = -EINVAL;
					goto exit_f;
				}
			}
720
			fl6->flowlabel = IPV6_FLOWINFO_MASK & *(__be32 *)CMSG_DATA(cmsg);
L
Linus Torvalds 已提交
721 722
			break;

723
		case IPV6_2292HOPOPTS:
L
Linus Torvalds 已提交
724
		case IPV6_HOPOPTS:
725
			if (opt->hopopt || cmsg->cmsg_len < CMSG_LEN(sizeof(struct ipv6_opt_hdr))) {
L
Linus Torvalds 已提交
726 727 728 729 730 731 732 733 734 735
				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;
			}
736
			if (!ns_capable(net->user_ns, CAP_NET_RAW)) {
L
Linus Torvalds 已提交
737 738 739 740 741 742 743
				err = -EPERM;
				goto exit_f;
			}
			opt->opt_nflen += len;
			opt->hopopt = hdr;
			break;

744
		case IPV6_2292DSTOPTS:
745
			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct ipv6_opt_hdr))) {
L
Linus Torvalds 已提交
746 747 748 749 750 751 752 753 754 755
				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;
			}
756
			if (!ns_capable(net->user_ns, CAP_NET_RAW)) {
L
Linus Torvalds 已提交
757 758 759 760 761 762 763 764 765 766 767
				err = -EPERM;
				goto exit_f;
			}
			if (opt->dst1opt) {
				err = -EINVAL;
				goto exit_f;
			}
			opt->opt_flen += len;
			opt->dst1opt = hdr;
			break;

768 769 770 771 772 773 774 775 776 777 778 779 780
		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;
			}
781
			if (!ns_capable(net->user_ns, CAP_NET_RAW)) {
782 783 784 785 786 787 788 789 790 791 792 793 794
				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
795
		case IPV6_RTHDR:
796
			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct ipv6_rt_hdr))) {
L
Linus Torvalds 已提交
797 798 799 800 801 802
				err = -EINVAL;
				goto exit_f;
			}

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

803
			switch (rthdr->type) {
A
Amerigo Wang 已提交
804
#if IS_ENABLED(CONFIG_IPV6_MIP6)
805
			case IPV6_SRCRT_TYPE_2:
B
Brian Haley 已提交
806 807 808 809 810
				if (rthdr->hdrlen != 2 ||
				    rthdr->segments_left != 1) {
					err = -EINVAL;
					goto exit_f;
				}
811
				break;
812
#endif
813
			default:
L
Linus Torvalds 已提交
814 815 816 817 818 819
				err = -EINVAL;
				goto exit_f;
			}

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

820
			if (cmsg->cmsg_len < CMSG_LEN(len)) {
L
Linus Torvalds 已提交
821 822 823 824 825 826 827 828 829 830 831 832 833
				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;

834
			if (cmsg->cmsg_type == IPV6_2292RTHDR && opt->dst1opt) {
L
Linus Torvalds 已提交
835 836 837 838 839 840 841 842 843 844
				int dsthdrlen = ((opt->dst1opt->hdrlen+1)<<3);

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

			break;

845
		case IPV6_2292HOPLIMIT:
L
Linus Torvalds 已提交
846 847 848 849 850 851 852
		case IPV6_HOPLIMIT:
			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int))) {
				err = -EINVAL;
				goto exit_f;
			}

			*hlimit = *(int *)CMSG_DATA(cmsg);
853 854 855 856 857
			if (*hlimit < -1 || *hlimit > 0xff) {
				err = -EINVAL;
				goto exit_f;
			}

L
Linus Torvalds 已提交
858 859
			break;

860 861 862 863 864
		case IPV6_TCLASS:
		    {
			int tc;

			err = -EINVAL;
865
			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int)))
866 867 868
				goto exit_f;

			tc = *(int *)CMSG_DATA(cmsg);
869
			if (tc < -1 || tc > 0xff)
870 871 872 873 874
				goto exit_f;

			err = 0;
			*tclass = tc;

875 876 877 878 879 880 881 882
			break;
		    }

		case IPV6_DONTFRAG:
		    {
			int df;

			err = -EINVAL;
883
			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int)))
884 885 886 887 888 889 890 891 892
				goto exit_f;

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

			err = 0;
			*dontfrag = df;

893 894
			break;
		    }
L
Linus Torvalds 已提交
895
		default:
896 897
			net_dbg_ratelimited("invalid cmsg type: %d\n",
					    cmsg->cmsg_type);
L
Linus Torvalds 已提交
898
			err = -EINVAL;
899
			goto exit_f;
900
		}
L
Linus Torvalds 已提交
901 902 903 904 905
	}

exit_f:
	return err;
}
906
EXPORT_SYMBOL_GPL(ip6_datagram_send_ctl);
907 908 909 910 911 912

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;

913 914
	dest  = &sp->sk_v6_daddr;
	src   = &sp->sk_v6_rcv_saddr;
915 916
	seq_printf(seq,
		   "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
917
		   "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d\n",
918 919 920 921 922 923 924 925 926 927 928 929 930 931 932
		   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));
}