datagram.c 20.1 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>
L
Linus Torvalds 已提交
33 34 35 36

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

37
static bool ipv6_mapped_addr_any(const struct in6_addr *a)
38
{
39
	return ipv6_addr_v4mapped(a) && (a->s6_addr32[3] == 0);
40 41
}

L
Linus Torvalds 已提交
42 43 44 45 46
int ip6_datagram_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
{
	struct sockaddr_in6	*usin = (struct sockaddr_in6 *) uaddr;
	struct inet_sock      	*inet = inet_sk(sk);
	struct ipv6_pinfo      	*np = inet6_sk(sk);
47
	struct in6_addr		*daddr, *final_p, final;
L
Linus Torvalds 已提交
48
	struct dst_entry	*dst;
49
	struct flowi6		fl6;
L
Linus Torvalds 已提交
50
	struct ip6_flowlabel	*flowlabel = NULL;
51
	struct ipv6_txoptions   *opt;
L
Linus Torvalds 已提交
52 53 54 55 56 57 58 59 60 61 62
	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)
63
		return -EINVAL;
L
Linus Torvalds 已提交
64

65 66
	if (usin->sin6_family != AF_INET6)
		return -EAFNOSUPPORT;
L
Linus Torvalds 已提交
67

68
	memset(&fl6, 0, sizeof(fl6));
L
Linus Torvalds 已提交
69
	if (np->sndflow) {
70 71 72
		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
73 74
			if (flowlabel == NULL)
				return -EINVAL;
A
Alexey Dobriyan 已提交
75
			usin->sin6_addr = flowlabel->dst;
L
Linus Torvalds 已提交
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
		}
	}

	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

E
Eric Dumazet 已提交
109
		ipv6_addr_set_v4mapped(inet->inet_daddr, &np->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(&np->rcv_saddr) ||
		    ipv6_mapped_addr_any(&np->rcv_saddr)) {
E
Eric Dumazet 已提交
117 118
			ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
					       &np->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
122 123 124 125 126 127 128 129 130 131 132 133 134 135 136

		goto out;
	}

	if (addr_type&IPV6_ADDR_LINKLOCAL) {
		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;
		}
	}

A
Alexey Dobriyan 已提交
147
	np->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;
A
Alexey Dobriyan 已提交
158 159
	fl6.daddr = np->daddr;
	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, true);
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

	if (ipv6_addr_any(&np->rcv_saddr)) {
A
Alexey Dobriyan 已提交
186
		np->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
		      ipv6_addr_equal(&fl6.daddr, &np->daddr) ?
194 195
		      &np->daddr : NULL,
#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
200 201 202 203 204 205

	sk->sk_state = TCP_ESTABLISHED;
out:
	fl6_sock_release(flowlabel);
	return err;
}
206
EXPORT_SYMBOL_GPL(ip6_datagram_connect);
L
Linus Torvalds 已提交
207

208
void ipv6_icmp_error(struct sock *sk, struct sk_buff *skb, int err,
A
Al Viro 已提交
209
		     __be16 port, u32 info, u8 *payload)
L
Linus Torvalds 已提交
210 211
{
	struct ipv6_pinfo *np  = inet6_sk(sk);
212
	struct icmp6hdr *icmph = icmp6_hdr(skb);
L
Linus Torvalds 已提交
213 214 215 216 217 218 219 220 221
	struct sock_exterr_skb *serr;

	if (!np->recverr)
		return;

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

222 223
	skb->protocol = htons(ETH_P_IPV6);

L
Linus Torvalds 已提交
224 225 226
	serr = SKB_EXT_ERR(skb);
	serr->ee.ee_errno = err;
	serr->ee.ee_origin = SO_EE_ORIGIN_ICMP6;
227
	serr->ee.ee_type = icmph->icmp6_type;
L
Linus Torvalds 已提交
228 229 230 231
	serr->ee.ee_code = icmph->icmp6_code;
	serr->ee.ee_pad = 0;
	serr->ee.ee_info = info;
	serr->ee.ee_data = 0;
232 233
	serr->addr_offset = (u8 *)&(((struct ipv6hdr *)(icmph + 1))->daddr) -
				  skb_network_header(skb);
L
Linus Torvalds 已提交
234 235 236
	serr->port = port;

	__skb_pull(skb, payload - skb->data);
237
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
238 239 240 241 242

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

243
void ipv6_local_error(struct sock *sk, int err, struct flowi6 *fl6, u32 info)
L
Linus Torvalds 已提交
244 245 246 247 248 249 250 251 252 253 254 255 256
{
	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;

257 258
	skb->protocol = htons(ETH_P_IPV6);

259 260
	skb_put(skb, sizeof(struct ipv6hdr));
	skb_reset_network_header(skb);
261
	iph = ipv6_hdr(skb);
A
Alexey Dobriyan 已提交
262
	iph->daddr = fl6->daddr;
L
Linus Torvalds 已提交
263 264 265 266

	serr = SKB_EXT_ERR(skb);
	serr->ee.ee_errno = err;
	serr->ee.ee_origin = SO_EE_ORIGIN_LOCAL;
267
	serr->ee.ee_type = 0;
L
Linus Torvalds 已提交
268 269 270 271
	serr->ee.ee_code = 0;
	serr->ee.ee_pad = 0;
	serr->ee.ee_info = info;
	serr->ee.ee_data = 0;
272
	serr->addr_offset = (u8 *)&iph->daddr - skb_network_header(skb);
273
	serr->port = fl6->fl6_dport;
L
Linus Torvalds 已提交
274

275
	__skb_pull(skb, skb_tail_pointer(skb) - skb->data);
276
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
277 278 279 280 281

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

282
void ipv6_local_rxpmtu(struct sock *sk, struct flowi6 *fl6, u32 mtu)
283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298
{
	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
299
	iph->daddr = fl6->daddr;
300 301 302 303 304 305 306

	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;
307
	mtu_info->ip6m_addr.sin6_scope_id = fl6->flowi6_oif;
A
Alexey Dobriyan 已提交
308
	mtu_info->ip6m_addr.sin6_addr = ipv6_hdr(skb)->daddr;
309 310 311 312 313 314 315 316

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

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

317
/*
L
Linus Torvalds 已提交
318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352
 *	Handle MSG_ERRQUEUE
 */
int ipv6_recv_error(struct sock *sk, struct msghdr *msg, int len)
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct sock_exterr_skb *serr;
	struct sk_buff *skb, *skb2;
	struct sockaddr_in6 *sin;
	struct {
		struct sock_extended_err ee;
		struct sockaddr_in6	 offender;
	} errhdr;
	int err;
	int copied;

	err = -EAGAIN;
	skb = skb_dequeue(&sk->sk_error_queue);
	if (skb == NULL)
		goto out;

	copied = skb->len;
	if (copied > len) {
		msg->msg_flags |= MSG_TRUNC;
		copied = len;
	}
	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
	if (err)
		goto out_free_skb;

	sock_recv_timestamp(msg, sk, skb);

	serr = SKB_EXT_ERR(skb);

	sin = (struct sockaddr_in6 *)msg->msg_name;
	if (sin) {
353
		const unsigned char *nh = skb_network_header(skb);
L
Linus Torvalds 已提交
354 355
		sin->sin6_family = AF_INET6;
		sin->sin6_flowinfo = 0;
356
		sin->sin6_port = serr->port;
L
Linus Torvalds 已提交
357
		sin->sin6_scope_id = 0;
358
		if (skb->protocol == htons(ETH_P_IPV6)) {
359 360 361
			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
362
			if (np->sndflow)
363
				sin->sin6_flowinfo = *(__be32 *)ip6h & IPV6_FLOWINFO_MASK;
L
Linus Torvalds 已提交
364 365 366
			if (ipv6_addr_type(&sin->sin6_addr) & IPV6_ADDR_LINKLOCAL)
				sin->sin6_scope_id = IP6CB(skb)->iif;
		} else {
B
Brian Haley 已提交
367 368
			ipv6_addr_set_v4mapped(*(__be32 *)(nh + serr->addr_offset),
					       &sin->sin6_addr);
L
Linus Torvalds 已提交
369 370 371 372 373 374 375 376 377 378
		}
	}

	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;
		sin->sin6_scope_id = 0;
379
		if (skb->protocol == htons(ETH_P_IPV6)) {
A
Alexey Dobriyan 已提交
380
			sin->sin6_addr = ipv6_hdr(skb)->saddr;
L
Linus Torvalds 已提交
381 382 383 384 385 386 387
			if (np->rxopt.all)
				datagram_recv_ctl(sk, msg, skb);
			if (ipv6_addr_type(&sin->sin6_addr) & IPV6_ADDR_LINKLOCAL)
				sin->sin6_scope_id = IP6CB(skb)->iif;
		} else {
			struct inet_sock *inet = inet_sk(sk);

B
Brian Haley 已提交
388 389
			ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
					       &sin->sin6_addr);
L
Linus Torvalds 已提交
390 391 392 393 394 395 396 397 398 399 400 401 402
			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;

	/* Reset and regenerate socket error */
403
	spin_lock_bh(&sk->sk_error_queue.lock);
L
Linus Torvalds 已提交
404 405 406
	sk->sk_err = 0;
	if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
		sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
407
		spin_unlock_bh(&sk->sk_error_queue.lock);
L
Linus Torvalds 已提交
408 409
		sk->sk_error_report(sk);
	} else {
410
		spin_unlock_bh(&sk->sk_error_queue.lock);
L
Linus Torvalds 已提交
411 412
	}

413
out_free_skb:
L
Linus Torvalds 已提交
414 415 416 417
	kfree_skb(skb);
out:
	return err;
}
418
EXPORT_SYMBOL_GPL(ipv6_recv_error);
L
Linus Torvalds 已提交
419

420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455
/*
 *	Handle IPV6_RECVPATHMTU
 */
int ipv6_recv_rxpmtu(struct sock *sk, struct msghdr *msg, int len)
{
	struct ipv6_pinfo *np = inet6_sk(sk);
	struct sk_buff *skb;
	struct sockaddr_in6 *sin;
	struct ip6_mtuinfo mtu_info;
	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;
	}
	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
	if (err)
		goto out_free_skb;

	sock_recv_timestamp(msg, sk, skb);

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

	sin = (struct sockaddr_in6 *)msg->msg_name;
	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
456
		sin->sin6_addr = mtu_info.ip6m_addr.sin6_addr;
457 458 459 460 461 462 463 464 465 466 467
	}

	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
468 469 470 471 472 473


int datagram_recv_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);
474
	unsigned char *nh = skb_network_header(skb);
L
Linus Torvalds 已提交
475 476 477 478 479

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

		src_info.ipi6_ifindex = opt->iif;
A
Alexey Dobriyan 已提交
480
		src_info.ipi6_addr = ipv6_hdr(skb)->daddr;
L
Linus Torvalds 已提交
481 482 483 484
		put_cmsg(msg, SOL_IPV6, IPV6_PKTINFO, sizeof(src_info), &src_info);
	}

	if (np->rxopt.bits.rxhlim) {
485
		int hlim = ipv6_hdr(skb)->hop_limit;
L
Linus Torvalds 已提交
486 487 488
		put_cmsg(msg, SOL_IPV6, IPV6_HOPLIMIT, sizeof(hlim), &hlim);
	}

489
	if (np->rxopt.bits.rxtclass) {
J
Jiri Benc 已提交
490
		int tclass = ipv6_tclass(ipv6_hdr(skb));
491 492 493
		put_cmsg(msg, SOL_IPV6, IPV6_TCLASS, sizeof(tclass), &tclass);
	}

494 495
	if (np->rxopt.bits.rxflow && (*(__be32 *)nh & IPV6_FLOWINFO_MASK)) {
		__be32 flowinfo = *(__be32 *)nh & IPV6_FLOWINFO_MASK;
L
Linus Torvalds 已提交
496 497
		put_cmsg(msg, SOL_IPV6, IPV6_FLOWINFO, sizeof(flowinfo), &flowinfo);
	}
498 499

	/* HbH is allowed only once */
L
Linus Torvalds 已提交
500
	if (np->rxopt.bits.hopopts && opt->hop) {
501
		u8 *ptr = nh + opt->hop;
L
Linus Torvalds 已提交
502 503
		put_cmsg(msg, SOL_IPV6, IPV6_HOPOPTS, (ptr[1]+1)<<3, ptr);
	}
504 505 506 507 508 509 510 511

	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.
		 *
512
		 * Also note that IPV6_RECVRTHDRDSTOPTS is NOT
513 514 515 516
		 * (and WILL NOT be) defined because
		 * IPV6_RECVDSTOPTS is more generic. --yoshfuji
		 */
		unsigned int off = sizeof(struct ipv6hdr);
517
		u8 nexthdr = ipv6_hdr(skb)->nexthdr;
518 519

		while (off <= opt->lastopt) {
520
			unsigned int len;
521
			u8 *ptr = nh + off;
522

523
			switch (nexthdr) {
524 525 526 527 528 529 530 531 532 533 534 535 536 537
			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];
538
				len = (ptr[1] + 2) << 2;
539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554
				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
555
		src_info.ipi6_addr = ipv6_hdr(skb)->daddr;
556 557 558
		put_cmsg(msg, SOL_IPV6, IPV6_2292PKTINFO, sizeof(src_info), &src_info);
	}
	if (np->rxopt.bits.rxohlim) {
559
		int hlim = ipv6_hdr(skb)->hop_limit;
560 561 562
		put_cmsg(msg, SOL_IPV6, IPV6_2292HOPLIMIT, sizeof(hlim), &hlim);
	}
	if (np->rxopt.bits.ohopopts && opt->hop) {
563
		u8 *ptr = nh + opt->hop;
564 565 566
		put_cmsg(msg, SOL_IPV6, IPV6_2292HOPOPTS, (ptr[1]+1)<<3, ptr);
	}
	if (np->rxopt.bits.odstopts && opt->dst0) {
567
		u8 *ptr = nh + opt->dst0;
568
		put_cmsg(msg, SOL_IPV6, IPV6_2292DSTOPTS, (ptr[1]+1)<<3, ptr);
L
Linus Torvalds 已提交
569
	}
570
	if (np->rxopt.bits.osrcrt && opt->srcrt) {
571
		struct ipv6_rt_hdr *rthdr = (struct ipv6_rt_hdr *)(nh + opt->srcrt);
572
		put_cmsg(msg, SOL_IPV6, IPV6_2292RTHDR, (rthdr->hdrlen+1) << 3, rthdr);
L
Linus Torvalds 已提交
573
	}
574
	if (np->rxopt.bits.odstopts && opt->dst1) {
575
		u8 *ptr = nh + opt->dst1;
576
		put_cmsg(msg, SOL_IPV6, IPV6_2292DSTOPTS, (ptr[1]+1)<<3, ptr);
L
Linus Torvalds 已提交
577
	}
578 579
	if (np->rxopt.bits.rxorigdstaddr) {
		struct sockaddr_in6 sin6;
E
Eric Dumazet 已提交
580
		__be16 *ports = (__be16 *) skb_transport_header(skb);
581 582 583 584 585 586 587 588

		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
589
			sin6.sin6_addr = ipv6_hdr(skb)->daddr;
590 591 592 593 594 595 596
			sin6.sin6_port = ports[1];
			sin6.sin6_flowinfo = 0;
			sin6.sin6_scope_id = 0;

			put_cmsg(msg, SOL_IPV6, IPV6_ORIGDSTADDR, sizeof(sin6), &sin6);
		}
	}
L
Linus Torvalds 已提交
597 598 599
	return 0;
}

600
int datagram_send_ctl(struct net *net, struct sock *sk,
601
		      struct msghdr *msg, struct flowi6 *fl6,
L
Linus Torvalds 已提交
602
		      struct ipv6_txoptions *opt,
603
		      int *hlimit, int *tclass, int *dontfrag)
L
Linus Torvalds 已提交
604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623
{
	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) {
624 625
		case IPV6_PKTINFO:
		case IPV6_2292PKTINFO:
626 627 628
		    {
			struct net_device *dev = NULL;

629
			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct in6_pktinfo))) {
L
Linus Torvalds 已提交
630 631 632 633 634
				err = -EINVAL;
				goto exit_f;
			}

			src_info = (struct in6_pktinfo *)CMSG_DATA(cmsg);
635

L
Linus Torvalds 已提交
636
			if (src_info->ipi6_ifindex) {
637 638
				if (fl6->flowi6_oif &&
				    src_info->ipi6_ifindex != fl6->flowi6_oif)
L
Linus Torvalds 已提交
639
					return -EINVAL;
640
				fl6->flowi6_oif = src_info->ipi6_ifindex;
L
Linus Torvalds 已提交
641 642
			}

643
			addr_type = __ipv6_addr_type(&src_info->ipi6_addr);
L
Linus Torvalds 已提交
644

645
			rcu_read_lock();
646 647
			if (fl6->flowi6_oif) {
				dev = dev_get_by_index_rcu(net, fl6->flowi6_oif);
648 649
				if (!dev) {
					rcu_read_unlock();
650
					return -ENODEV;
651 652 653
				}
			} else if (addr_type & IPV6_ADDR_LINKLOCAL) {
				rcu_read_unlock();
654
				return -EINVAL;
655
			}
656

657 658
			if (addr_type != IPV6_ADDR_ANY) {
				int strict = __ipv6_addr_src_scope(addr_type) <= IPV6_ADDR_SCOPE_LINKLOCAL;
659
				if (!(inet_sk(sk)->freebind || inet_sk(sk)->transparent) &&
660
				    !ipv6_chk_addr(net, &src_info->ipi6_addr,
661 662 663
						   strict ? dev : NULL, 0))
					err = -EINVAL;
				else
A
Alexey Dobriyan 已提交
664
					fl6->saddr = src_info->ipi6_addr;
L
Linus Torvalds 已提交
665
			}
666

667
			rcu_read_unlock();
L
Linus Torvalds 已提交
668

669 670 671
			if (err)
				goto exit_f;

L
Linus Torvalds 已提交
672
			break;
673
		    }
L
Linus Torvalds 已提交
674 675

		case IPV6_FLOWINFO:
676
			if (cmsg->cmsg_len < CMSG_LEN(4)) {
L
Linus Torvalds 已提交
677 678 679 680
				err = -EINVAL;
				goto exit_f;
			}

681 682
			if (fl6->flowlabel&IPV6_FLOWINFO_MASK) {
				if ((fl6->flowlabel^*(__be32 *)CMSG_DATA(cmsg))&~IPV6_FLOWINFO_MASK) {
L
Linus Torvalds 已提交
683 684 685 686
					err = -EINVAL;
					goto exit_f;
				}
			}
687
			fl6->flowlabel = IPV6_FLOWINFO_MASK & *(__be32 *)CMSG_DATA(cmsg);
L
Linus Torvalds 已提交
688 689
			break;

690
		case IPV6_2292HOPOPTS:
L
Linus Torvalds 已提交
691
		case IPV6_HOPOPTS:
692
			if (opt->hopopt || cmsg->cmsg_len < CMSG_LEN(sizeof(struct ipv6_opt_hdr))) {
L
Linus Torvalds 已提交
693 694 695 696 697 698 699 700 701 702
				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;
			}
703
			if (!ns_capable(net->user_ns, CAP_NET_RAW)) {
L
Linus Torvalds 已提交
704 705 706 707 708 709 710
				err = -EPERM;
				goto exit_f;
			}
			opt->opt_nflen += len;
			opt->hopopt = hdr;
			break;

711
		case IPV6_2292DSTOPTS:
712
			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct ipv6_opt_hdr))) {
L
Linus Torvalds 已提交
713 714 715 716 717 718 719 720 721 722
				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;
			}
723
			if (!ns_capable(net->user_ns, CAP_NET_RAW)) {
L
Linus Torvalds 已提交
724 725 726 727 728 729 730 731 732 733 734
				err = -EPERM;
				goto exit_f;
			}
			if (opt->dst1opt) {
				err = -EINVAL;
				goto exit_f;
			}
			opt->opt_flen += len;
			opt->dst1opt = hdr;
			break;

735 736 737 738 739 740 741 742 743 744 745 746 747
		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;
			}
748
			if (!ns_capable(net->user_ns, CAP_NET_RAW)) {
749 750 751 752 753 754 755 756 757 758 759 760 761
				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
762
		case IPV6_RTHDR:
763
			if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct ipv6_rt_hdr))) {
L
Linus Torvalds 已提交
764 765 766 767 768 769
				err = -EINVAL;
				goto exit_f;
			}

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

770
			switch (rthdr->type) {
A
Amerigo Wang 已提交
771
#if IS_ENABLED(CONFIG_IPV6_MIP6)
772
			case IPV6_SRCRT_TYPE_2:
B
Brian Haley 已提交
773 774 775 776 777
				if (rthdr->hdrlen != 2 ||
				    rthdr->segments_left != 1) {
					err = -EINVAL;
					goto exit_f;
				}
778
				break;
779
#endif
780
			default:
L
Linus Torvalds 已提交
781 782 783 784 785 786
				err = -EINVAL;
				goto exit_f;
			}

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

787
			if (cmsg->cmsg_len < CMSG_LEN(len)) {
L
Linus Torvalds 已提交
788 789 790 791 792 793 794 795 796 797 798 799 800
				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;

801
			if (cmsg->cmsg_type == IPV6_2292RTHDR && opt->dst1opt) {
L
Linus Torvalds 已提交
802 803 804 805 806 807 808 809 810 811
				int dsthdrlen = ((opt->dst1opt->hdrlen+1)<<3);

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

			break;

812
		case IPV6_2292HOPLIMIT:
L
Linus Torvalds 已提交
813 814 815 816 817 818 819
		case IPV6_HOPLIMIT:
			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int))) {
				err = -EINVAL;
				goto exit_f;
			}

			*hlimit = *(int *)CMSG_DATA(cmsg);
820 821 822 823 824
			if (*hlimit < -1 || *hlimit > 0xff) {
				err = -EINVAL;
				goto exit_f;
			}

L
Linus Torvalds 已提交
825 826
			break;

827 828 829 830 831
		case IPV6_TCLASS:
		    {
			int tc;

			err = -EINVAL;
832
			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int)))
833 834 835
				goto exit_f;

			tc = *(int *)CMSG_DATA(cmsg);
836
			if (tc < -1 || tc > 0xff)
837 838 839 840 841
				goto exit_f;

			err = 0;
			*tclass = tc;

842 843 844 845 846 847 848 849
			break;
		    }

		case IPV6_DONTFRAG:
		    {
			int df;

			err = -EINVAL;
850
			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int)))
851 852 853 854 855 856 857 858 859
				goto exit_f;

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

			err = 0;
			*dontfrag = df;

860 861
			break;
		    }
L
Linus Torvalds 已提交
862
		default:
863
			LIMIT_NETDEBUG(KERN_DEBUG "invalid cmsg type: %d\n",
864
				       cmsg->cmsg_type);
L
Linus Torvalds 已提交
865
			err = -EINVAL;
866
			goto exit_f;
867
		}
L
Linus Torvalds 已提交
868 869 870 871 872
	}

exit_f:
	return err;
}
873
EXPORT_SYMBOL_GPL(datagram_send_ctl);