ip_sockglue.c 35.4 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6
/*
 * INET		An implementation of the TCP/IP protocol suite for the LINUX
 *		operating system.  INET is implemented using the  BSD Socket
 *		interface as the means of communication with the user level.
 *
 *		The IP to API glue.
7
 *
L
Linus Torvalds 已提交
8 9 10 11 12
 * Authors:	see ip.c
 *
 * Fixes:
 *		Many		:	Split from ip.c , see ip.c for history.
 *		Martin Mares	:	TOS setting fixed.
13
 *		Alan Cox	:	Fixed a couple of oopses in Martin's
L
Linus Torvalds 已提交
14 15 16 17 18 19 20 21 22 23
 *					TOS tweaks.
 *		Mike McLagan	:	Routing by source
 */

#include <linux/module.h>
#include <linux/types.h>
#include <linux/mm.h>
#include <linux/skbuff.h>
#include <linux/ip.h>
#include <linux/icmp.h>
24
#include <linux/inetdevice.h>
L
Linus Torvalds 已提交
25
#include <linux/netdevice.h>
26
#include <linux/slab.h>
L
Linus Torvalds 已提交
27 28 29
#include <net/sock.h>
#include <net/ip.h>
#include <net/icmp.h>
30
#include <net/tcp_states.h>
L
Linus Torvalds 已提交
31 32 33 34 35
#include <linux/udp.h>
#include <linux/igmp.h>
#include <linux/netfilter.h>
#include <linux/route.h>
#include <linux/mroute.h>
36
#include <net/inet_ecn.h>
L
Linus Torvalds 已提交
37 38
#include <net/route.h>
#include <net/xfrm.h>
39
#include <net/compat.h>
40
#include <net/checksum.h>
E
Eric Dumazet 已提交
41
#if IS_ENABLED(CONFIG_IPV6)
L
Linus Torvalds 已提交
42 43
#include <net/transp_v6.h>
#endif
44
#include <net/ip_fib.h>
L
Linus Torvalds 已提交
45 46 47 48 49 50 51 52 53 54

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

/*
 *	SOL_IP control messages.
 */

static void ip_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
{
55
	struct in_pktinfo info = *PKTINFO_SKB_CB(skb);
L
Linus Torvalds 已提交
56

57
	info.ipi_addr.s_addr = ip_hdr(skb)->daddr;
L
Linus Torvalds 已提交
58 59 60 61 62 63

	put_cmsg(msg, SOL_IP, IP_PKTINFO, sizeof(info), &info);
}

static void ip_cmsg_recv_ttl(struct msghdr *msg, struct sk_buff *skb)
{
64
	int ttl = ip_hdr(skb)->ttl;
L
Linus Torvalds 已提交
65 66 67 68 69
	put_cmsg(msg, SOL_IP, IP_TTL, sizeof(int), &ttl);
}

static void ip_cmsg_recv_tos(struct msghdr *msg, struct sk_buff *skb)
{
70
	put_cmsg(msg, SOL_IP, IP_TOS, 1, &ip_hdr(skb)->tos);
L
Linus Torvalds 已提交
71 72 73 74 75 76 77
}

static void ip_cmsg_recv_opts(struct msghdr *msg, struct sk_buff *skb)
{
	if (IPCB(skb)->opt.optlen == 0)
		return;

78 79
	put_cmsg(msg, SOL_IP, IP_RECVOPTS, IPCB(skb)->opt.optlen,
		 ip_hdr(skb) + 1);
L
Linus Torvalds 已提交
80 81 82 83 84 85
}


static void ip_cmsg_recv_retopts(struct msghdr *msg, struct sk_buff *skb)
{
	unsigned char optbuf[sizeof(struct ip_options) + 40];
D
Daniel Baluta 已提交
86
	struct ip_options *opt = (struct ip_options *)optbuf;
L
Linus Torvalds 已提交
87 88 89 90 91 92 93 94 95 96 97 98 99

	if (IPCB(skb)->opt.optlen == 0)
		return;

	if (ip_options_echo(opt, skb)) {
		msg->msg_flags |= MSG_CTRUNC;
		return;
	}
	ip_options_undo(opt);

	put_cmsg(msg, SOL_IP, IP_RETOPTS, opt->optlen, opt->__data);
}

100 101 102 103 104 105 106 107 108
static void ip_cmsg_recv_checksum(struct msghdr *msg, struct sk_buff *skb,
				  int offset)
{
	__wsum csum = skb->csum;

	if (skb->ip_summed != CHECKSUM_COMPLETE)
		return;

	if (offset != 0)
109 110
		csum = csum_sub(csum, csum_partial(skb_transport_header(skb),
						   offset, 0));
111 112 113 114

	put_cmsg(msg, SOL_IP, IP_CHECKSUM, sizeof(__wsum), &csum);
}

C
Catherine Zhang 已提交
115 116 117
static void ip_cmsg_recv_security(struct msghdr *msg, struct sk_buff *skb)
{
	char *secdata;
118
	u32 seclen, secid;
C
Catherine Zhang 已提交
119 120
	int err;

121 122 123 124 125
	err = security_socket_getpeersec_dgram(NULL, skb, &secid);
	if (err)
		return;

	err = security_secid_to_secctx(secid, &secdata, &seclen);
C
Catherine Zhang 已提交
126 127 128 129
	if (err)
		return;

	put_cmsg(msg, SOL_IP, SCM_SECURITY, seclen, secdata);
130
	security_release_secctx(secdata, seclen);
C
Catherine Zhang 已提交
131 132
}

133
static void ip_cmsg_recv_dstaddr(struct msghdr *msg, struct sk_buff *skb)
134 135
{
	struct sockaddr_in sin;
136
	const struct iphdr *iph = ip_hdr(skb);
137
	__be16 *ports = (__be16 *)skb_transport_header(skb);
138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153

	if (skb_transport_offset(skb) + 4 > skb->len)
		return;

	/* 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.
	 */

	sin.sin_family = AF_INET;
	sin.sin_addr.s_addr = iph->daddr;
	sin.sin_port = ports[1];
	memset(sin.sin_zero, 0, sizeof(sin.sin_zero));

	put_cmsg(msg, SOL_IP, IP_ORIGDSTADDR, sizeof(sin), &sin);
}
L
Linus Torvalds 已提交
154

155 156
void ip_cmsg_recv_offset(struct msghdr *msg, struct sk_buff *skb,
			 int offset)
L
Linus Torvalds 已提交
157 158
{
	struct inet_sock *inet = inet_sk(skb->sk);
159
	unsigned int flags = inet->cmsg_flags;
L
Linus Torvalds 已提交
160 161

	/* Ordered by supposed usage frequency */
T
Tom Herbert 已提交
162
	if (flags & IP_CMSG_PKTINFO) {
L
Linus Torvalds 已提交
163 164
		ip_cmsg_recv_pktinfo(msg, skb);

T
Tom Herbert 已提交
165 166 167 168 169 170
		flags &= ~IP_CMSG_PKTINFO;
		if (!flags)
			return;
	}

	if (flags & IP_CMSG_TTL) {
L
Linus Torvalds 已提交
171 172
		ip_cmsg_recv_ttl(msg, skb);

T
Tom Herbert 已提交
173 174 175 176 177 178
		flags &= ~IP_CMSG_TTL;
		if (!flags)
			return;
	}

	if (flags & IP_CMSG_TOS) {
L
Linus Torvalds 已提交
179 180
		ip_cmsg_recv_tos(msg, skb);

T
Tom Herbert 已提交
181 182 183 184 185 186
		flags &= ~IP_CMSG_TOS;
		if (!flags)
			return;
	}

	if (flags & IP_CMSG_RECVOPTS) {
L
Linus Torvalds 已提交
187 188
		ip_cmsg_recv_opts(msg, skb);

T
Tom Herbert 已提交
189 190 191 192 193 194
		flags &= ~IP_CMSG_RECVOPTS;
		if (!flags)
			return;
	}

	if (flags & IP_CMSG_RETOPTS) {
L
Linus Torvalds 已提交
195
		ip_cmsg_recv_retopts(msg, skb);
C
Catherine Zhang 已提交
196

T
Tom Herbert 已提交
197 198 199 200 201 202
		flags &= ~IP_CMSG_RETOPTS;
		if (!flags)
			return;
	}

	if (flags & IP_CMSG_PASSSEC) {
C
Catherine Zhang 已提交
203
		ip_cmsg_recv_security(msg, skb);
204

T
Tom Herbert 已提交
205 206 207 208 209
		flags &= ~IP_CMSG_PASSSEC;
		if (!flags)
			return;
	}

210
	if (flags & IP_CMSG_ORIGDSTADDR) {
211 212
		ip_cmsg_recv_dstaddr(msg, skb);

213 214 215 216 217 218 219
		flags &= ~IP_CMSG_ORIGDSTADDR;
		if (!flags)
			return;
	}

	if (flags & IP_CMSG_CHECKSUM)
		ip_cmsg_recv_checksum(msg, skb, offset);
L
Linus Torvalds 已提交
220
}
221
EXPORT_SYMBOL(ip_cmsg_recv_offset);
L
Linus Torvalds 已提交
222

223
int ip_cmsg_send(struct sock *sk, struct msghdr *msg, struct ipcm_cookie *ipc,
224
		 bool allow_ipv6)
L
Linus Torvalds 已提交
225
{
226
	int err, val;
L
Linus Torvalds 已提交
227
	struct cmsghdr *cmsg;
228
	struct net *net = sock_net(sk);
L
Linus Torvalds 已提交
229

230
	for_each_cmsghdr(cmsg, msg) {
L
Linus Torvalds 已提交
231 232
		if (!CMSG_OK(msg, cmsg))
			return -EINVAL;
233
#if IS_ENABLED(CONFIG_IPV6)
234 235 236 237 238 239 240 241 242 243 244 245 246 247 248
		if (allow_ipv6 &&
		    cmsg->cmsg_level == SOL_IPV6 &&
		    cmsg->cmsg_type == IPV6_PKTINFO) {
			struct in6_pktinfo *src_info;

			if (cmsg->cmsg_len < CMSG_LEN(sizeof(*src_info)))
				return -EINVAL;
			src_info = (struct in6_pktinfo *)CMSG_DATA(cmsg);
			if (!ipv6_addr_v4mapped(&src_info->ipi6_addr))
				return -EINVAL;
			ipc->oif = src_info->ipi6_ifindex;
			ipc->addr = src_info->ipi6_addr.s6_addr32[3];
			continue;
		}
#endif
249 250 251 252 253 254
		if (cmsg->cmsg_level == SOL_SOCKET) {
			if (__sock_cmsg_send(sk, msg, cmsg, &ipc->sockc))
				return -EINVAL;
			continue;
		}

L
Linus Torvalds 已提交
255 256 257 258 259
		if (cmsg->cmsg_level != SOL_IP)
			continue;
		switch (cmsg->cmsg_type) {
		case IP_RETOPTS:
			err = cmsg->cmsg_len - CMSG_ALIGN(sizeof(struct cmsghdr));
260 261

			/* Our caller is responsible for freeing ipc->opt */
E
Eric Dumazet 已提交
262 263
			err = ip_options_get(net, &ipc->opt, CMSG_DATA(cmsg),
					     err < 40 ? err : 40);
L
Linus Torvalds 已提交
264 265 266 267 268 269 270 271 272 273 274 275 276
			if (err)
				return err;
			break;
		case IP_PKTINFO:
		{
			struct in_pktinfo *info;
			if (cmsg->cmsg_len != CMSG_LEN(sizeof(struct in_pktinfo)))
				return -EINVAL;
			info = (struct in_pktinfo *)CMSG_DATA(cmsg);
			ipc->oif = info->ipi_ifindex;
			ipc->addr = info->ipi_spec_dst.s_addr;
			break;
		}
277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294
		case IP_TTL:
			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int)))
				return -EINVAL;
			val = *(int *)CMSG_DATA(cmsg);
			if (val < 1 || val > 255)
				return -EINVAL;
			ipc->ttl = val;
			break;
		case IP_TOS:
			if (cmsg->cmsg_len != CMSG_LEN(sizeof(int)))
				return -EINVAL;
			val = *(int *)CMSG_DATA(cmsg);
			if (val < 0 || val > 255)
				return -EINVAL;
			ipc->tos = val;
			ipc->priority = rt_tos2priority(ipc->tos);
			break;

L
Linus Torvalds 已提交
295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312
		default:
			return -EINVAL;
		}
	}
	return 0;
}


/* Special input handler for packets caught by router alert option.
   They are selected only by protocol field, and then processed likely
   local ones; but only if someone wants them! Otherwise, router
   not running rsvpd will kill RSVP.

   It is user level problem, what it will make with them.
   I have no idea, how it will masquearde or NAT them (it is joke, joke :-)),
   but receiver should be enough clever f.e. to forward mtrace requests,
   sent to multicast group to reach destination designated router.
 */
313
struct ip_ra_chain __rcu *ip_ra_chain;
E
Eric Dumazet 已提交
314 315
static DEFINE_SPINLOCK(ip_ra_lock);

E
Eric Dumazet 已提交
316 317

static void ip_ra_destroy_rcu(struct rcu_head *head)
E
Eric Dumazet 已提交
318
{
E
Eric Dumazet 已提交
319 320 321 322
	struct ip_ra_chain *ra = container_of(head, struct ip_ra_chain, rcu);

	sock_put(ra->saved_sk);
	kfree(ra);
E
Eric Dumazet 已提交
323
}
L
Linus Torvalds 已提交
324

E
Eric Dumazet 已提交
325 326
int ip_ra_control(struct sock *sk, unsigned char on,
		  void (*destructor)(struct sock *))
L
Linus Torvalds 已提交
327
{
328 329
	struct ip_ra_chain *ra, *new_ra;
	struct ip_ra_chain __rcu **rap;
L
Linus Torvalds 已提交
330

E
Eric Dumazet 已提交
331
	if (sk->sk_type != SOCK_RAW || inet_sk(sk)->inet_num == IPPROTO_RAW)
L
Linus Torvalds 已提交
332 333 334 335
		return -EINVAL;

	new_ra = on ? kmalloc(sizeof(*new_ra), GFP_KERNEL) : NULL;

E
Eric Dumazet 已提交
336
	spin_lock_bh(&ip_ra_lock);
337 338 339 340
	for (rap = &ip_ra_chain;
	     (ra = rcu_dereference_protected(*rap,
			lockdep_is_held(&ip_ra_lock))) != NULL;
	     rap = &ra->next) {
L
Linus Torvalds 已提交
341 342
		if (ra->sk == sk) {
			if (on) {
E
Eric Dumazet 已提交
343
				spin_unlock_bh(&ip_ra_lock);
J
Jesper Juhl 已提交
344
				kfree(new_ra);
L
Linus Torvalds 已提交
345 346
				return -EADDRINUSE;
			}
E
Eric Dumazet 已提交
347 348
			/* dont let ip_call_ra_chain() use sk again */
			ra->sk = NULL;
349
			RCU_INIT_POINTER(*rap, ra->next);
E
Eric Dumazet 已提交
350
			spin_unlock_bh(&ip_ra_lock);
L
Linus Torvalds 已提交
351 352 353

			if (ra->destructor)
				ra->destructor(sk);
E
Eric Dumazet 已提交
354 355 356 357 358 359 360
			/*
			 * Delay sock_put(sk) and kfree(ra) after one rcu grace
			 * period. This guarantee ip_call_ra_chain() dont need
			 * to mess with socket refcounts.
			 */
			ra->saved_sk = sk;
			call_rcu(&ra->rcu, ip_ra_destroy_rcu);
L
Linus Torvalds 已提交
361 362 363
			return 0;
		}
	}
364
	if (!new_ra) {
E
Eric Dumazet 已提交
365
		spin_unlock_bh(&ip_ra_lock);
L
Linus Torvalds 已提交
366 367 368 369 370
		return -ENOBUFS;
	}
	new_ra->sk = sk;
	new_ra->destructor = destructor;

371
	RCU_INIT_POINTER(new_ra->next, ra);
E
Eric Dumazet 已提交
372
	rcu_assign_pointer(*rap, new_ra);
L
Linus Torvalds 已提交
373
	sock_hold(sk);
E
Eric Dumazet 已提交
374
	spin_unlock_bh(&ip_ra_lock);
L
Linus Torvalds 已提交
375 376 377 378

	return 0;
}

379
void ip_icmp_error(struct sock *sk, struct sk_buff *skb, int err,
A
Al Viro 已提交
380
		   __be16 port, u32 info, u8 *payload)
L
Linus Torvalds 已提交
381 382 383 384 385 386 387
{
	struct sock_exterr_skb *serr;

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

388
	serr = SKB_EXT_ERR(skb);
L
Linus Torvalds 已提交
389 390
	serr->ee.ee_errno = err;
	serr->ee.ee_origin = SO_EE_ORIGIN_ICMP;
391 392
	serr->ee.ee_type = icmp_hdr(skb)->type;
	serr->ee.ee_code = icmp_hdr(skb)->code;
L
Linus Torvalds 已提交
393 394 395
	serr->ee.ee_pad = 0;
	serr->ee.ee_info = info;
	serr->ee.ee_data = 0;
396
	serr->addr_offset = (u8 *)&(((struct iphdr *)(icmp_hdr(skb) + 1))->daddr) -
397
				   skb_network_header(skb);
L
Linus Torvalds 已提交
398 399
	serr->port = port;

400
	if (skb_pull(skb, payload - skb->data)) {
401 402 403 404 405
		skb_reset_transport_header(skb);
		if (sock_queue_err_skb(sk, skb) == 0)
			return;
	}
	kfree_skb(skb);
L
Linus Torvalds 已提交
406 407
}

A
Al Viro 已提交
408
void ip_local_error(struct sock *sk, int err, __be32 daddr, __be16 port, u32 info)
L
Linus Torvalds 已提交
409 410 411 412 413 414 415 416 417 418 419 420 421
{
	struct inet_sock *inet = inet_sk(sk);
	struct sock_exterr_skb *serr;
	struct iphdr *iph;
	struct sk_buff *skb;

	if (!inet->recverr)
		return;

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

422 423
	skb_put(skb, sizeof(struct iphdr));
	skb_reset_network_header(skb);
424
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
425 426
	iph->daddr = daddr;

427
	serr = SKB_EXT_ERR(skb);
L
Linus Torvalds 已提交
428 429
	serr->ee.ee_errno = err;
	serr->ee.ee_origin = SO_EE_ORIGIN_LOCAL;
430
	serr->ee.ee_type = 0;
L
Linus Torvalds 已提交
431 432 433 434
	serr->ee.ee_code = 0;
	serr->ee.ee_pad = 0;
	serr->ee.ee_info = info;
	serr->ee.ee_data = 0;
435
	serr->addr_offset = (u8 *)&iph->daddr - skb_network_header(skb);
L
Linus Torvalds 已提交
436 437
	serr->port = port;

438
	__skb_pull(skb, skb_tail_pointer(skb) - skb->data);
439
	skb_reset_transport_header(skb);
L
Linus Torvalds 已提交
440 441 442 443 444

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

445 446 447 448 449 450 451 452 453
/* 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 ipv4_datagram_support_addr(struct sock_exterr_skb *serr)
{
	return serr->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
	       serr->ee.ee_origin == SO_EE_ORIGIN_LOCAL || serr->port;
}

454 455 456 457 458 459 460 461
/* IPv4 supports cmsg on all imcp errors and some timestamps
 *
 * Timestamp code paths do not initialize the fields expected by cmsg:
 * the PKTINFO fields in skb->cb[]. Fill those in here.
 */
static bool ipv4_datagram_support_cmsg(const struct sock *sk,
				       struct sk_buff *skb,
				       int ee_origin)
462
{
463 464 465 466
	struct in_pktinfo *info;

	if (ee_origin == SO_EE_ORIGIN_ICMP)
		return true;
467

468 469 470 471 472 473 474 475
	if (ee_origin == SO_EE_ORIGIN_LOCAL)
		return false;

	/* Support IP_PKTINFO on tstamp packets if requested, to correlate
	 * timestamp with egress dev. Not possible for packets without dev
	 * or without payload (SOF_TIMESTAMPING_OPT_TSONLY).
	 */
	if ((!(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_CMSG)) ||
476 477 478
	    (!skb->dev))
		return false;

479
	info = PKTINFO_SKB_CB(skb);
480 481 482 483 484
	info->ipi_spec_dst.s_addr = ip_hdr(skb)->saddr;
	info->ipi_ifindex = skb->dev->ifindex;
	return true;
}

485
/*
L
Linus Torvalds 已提交
486 487
 *	Handle MSG_ERRQUEUE
 */
488
int ip_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
L
Linus Torvalds 已提交
489 490
{
	struct sock_exterr_skb *serr;
491
	struct sk_buff *skb;
492
	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
L
Linus Torvalds 已提交
493 494 495 496 497 498 499
	struct {
		struct sock_extended_err ee;
		struct sockaddr_in	 offender;
	} errhdr;
	int err;
	int copied;

500 501
	WARN_ON_ONCE(sk->sk_family == AF_INET6);

L
Linus Torvalds 已提交
502
	err = -EAGAIN;
503
	skb = sock_dequeue_err_skb(sk);
504
	if (!skb)
L
Linus Torvalds 已提交
505 506 507 508 509 510 511
		goto out;

	copied = skb->len;
	if (copied > len) {
		msg->msg_flags |= MSG_TRUNC;
		copied = len;
	}
512
	err = skb_copy_datagram_msg(skb, 0, msg, copied);
L
Linus Torvalds 已提交
513 514 515 516 517 518 519
	if (err)
		goto out_free_skb;

	sock_recv_timestamp(msg, sk, skb);

	serr = SKB_EXT_ERR(skb);

520
	if (sin && ipv4_datagram_support_addr(serr)) {
L
Linus Torvalds 已提交
521
		sin->sin_family = AF_INET;
522 523
		sin->sin_addr.s_addr = *(__be32 *)(skb_network_header(skb) +
						   serr->addr_offset);
L
Linus Torvalds 已提交
524 525
		sin->sin_port = serr->port;
		memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
526
		*addr_len = sizeof(*sin);
L
Linus Torvalds 已提交
527 528 529 530
	}

	memcpy(&errhdr.ee, &serr->ee, sizeof(struct sock_extended_err));
	sin = &errhdr.offender;
531
	memset(sin, 0, sizeof(*sin));
532

533
	if (ipv4_datagram_support_cmsg(sk, skb, serr->ee.ee_origin)) {
L
Linus Torvalds 已提交
534
		sin->sin_family = AF_INET;
535
		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
536
		if (inet_sk(sk)->cmsg_flags)
L
Linus Torvalds 已提交
537 538 539 540 541 542 543 544 545 546
			ip_cmsg_recv(msg, skb);
	}

	put_cmsg(msg, SOL_IP, IP_RECVERR, sizeof(errhdr), &errhdr);

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

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

547
out_free_skb:
L
Linus Torvalds 已提交
548 549 550 551 552 553 554
	kfree_skb(skb);
out:
	return err;
}


/*
E
Eric Dumazet 已提交
555 556
 *	Socket option code for IP. This is the end of the line after any
 *	TCP,UDP etc options on an IP socket.
L
Linus Torvalds 已提交
557
 */
558 559 560 561 562
static bool setsockopt_needs_rtnl(int optname)
{
	switch (optname) {
	case IP_ADD_MEMBERSHIP:
	case IP_ADD_SOURCE_MEMBERSHIP:
563
	case IP_BLOCK_SOURCE:
564
	case IP_DROP_MEMBERSHIP:
565 566 567 568 569
	case IP_DROP_SOURCE_MEMBERSHIP:
	case IP_MSFILTER:
	case IP_UNBLOCK_SOURCE:
	case MCAST_BLOCK_SOURCE:
	case MCAST_MSFILTER:
570
	case MCAST_JOIN_GROUP:
571
	case MCAST_JOIN_SOURCE_GROUP:
572
	case MCAST_LEAVE_GROUP:
573 574
	case MCAST_LEAVE_SOURCE_GROUP:
	case MCAST_UNBLOCK_SOURCE:
575 576 577 578
		return true;
	}
	return false;
}
L
Linus Torvalds 已提交
579

580
static int do_ip_setsockopt(struct sock *sk, int level,
581
			    int optname, char __user *optval, unsigned int optlen)
L
Linus Torvalds 已提交
582 583
{
	struct inet_sock *inet = inet_sk(sk);
584
	struct net *net = sock_net(sk);
585
	int val = 0, err;
586
	bool needs_rtnl = setsockopt_needs_rtnl(optname);
L
Linus Torvalds 已提交
587

588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604
	switch (optname) {
	case IP_PKTINFO:
	case IP_RECVTTL:
	case IP_RECVOPTS:
	case IP_RECVTOS:
	case IP_RETOPTS:
	case IP_TOS:
	case IP_TTL:
	case IP_HDRINCL:
	case IP_MTU_DISCOVER:
	case IP_RECVERR:
	case IP_ROUTER_ALERT:
	case IP_FREEBIND:
	case IP_PASSSEC:
	case IP_TRANSPARENT:
	case IP_MINTTL:
	case IP_NODEFRAG:
605
	case IP_BIND_ADDRESS_NO_PORT:
606 607 608 609 610
	case IP_UNICAST_IF:
	case IP_MULTICAST_TTL:
	case IP_MULTICAST_ALL:
	case IP_MULTICAST_LOOP:
	case IP_RECVORIGDSTADDR:
611
	case IP_CHECKSUM:
L
Linus Torvalds 已提交
612 613 614 615 616 617 618 619 620 621 622 623 624 625
		if (optlen >= sizeof(int)) {
			if (get_user(val, (int __user *) optval))
				return -EFAULT;
		} else if (optlen >= sizeof(char)) {
			unsigned char ucval;

			if (get_user(ucval, (unsigned char __user *) optval))
				return -EFAULT;
			val = (int) ucval;
		}
	}

	/* If optlen==0, it is equivalent to val == 0 */

626
	if (ip_mroute_opt(optname))
627
		return ip_mroute_setsockopt(sk, optname, optval, optlen);
L
Linus Torvalds 已提交
628 629

	err = 0;
630 631
	if (needs_rtnl)
		rtnl_lock();
L
Linus Torvalds 已提交
632 633 634
	lock_sock(sk);

	switch (optname) {
S
Stephen Hemminger 已提交
635 636
	case IP_OPTIONS:
	{
637 638
		struct ip_options_rcu *old, *opt = NULL;

639
		if (optlen > 40)
S
Stephen Hemminger 已提交
640
			goto e_inval;
641
		err = ip_options_get_from_user(sock_net(sk), &opt,
642
					       optval, optlen);
S
Stephen Hemminger 已提交
643 644
		if (err)
			break;
645
		old = rcu_dereference_protected(inet->inet_opt,
646
						lockdep_sock_is_held(sk));
S
Stephen Hemminger 已提交
647 648
		if (inet->is_icsk) {
			struct inet_connection_sock *icsk = inet_csk(sk);
E
Eric Dumazet 已提交
649
#if IS_ENABLED(CONFIG_IPV6)
S
Stephen Hemminger 已提交
650 651 652
			if (sk->sk_family == PF_INET ||
			    (!((1 << sk->sk_state) &
			       (TCPF_LISTEN | TCPF_CLOSE)) &&
E
Eric Dumazet 已提交
653
			     inet->inet_daddr != LOOPBACK4_IPV6)) {
L
Linus Torvalds 已提交
654
#endif
655 656
				if (old)
					icsk->icsk_ext_hdr_len -= old->opt.optlen;
S
Stephen Hemminger 已提交
657
				if (opt)
658
					icsk->icsk_ext_hdr_len += opt->opt.optlen;
S
Stephen Hemminger 已提交
659
				icsk->icsk_sync_mss(sk, icsk->icsk_pmtu_cookie);
E
Eric Dumazet 已提交
660
#if IS_ENABLED(CONFIG_IPV6)
L
Linus Torvalds 已提交
661
			}
S
Stephen Hemminger 已提交
662
#endif
L
Linus Torvalds 已提交
663
		}
664 665
		rcu_assign_pointer(inet->inet_opt, opt);
		if (old)
666
			kfree_rcu(old, rcu);
S
Stephen Hemminger 已提交
667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704
		break;
	}
	case IP_PKTINFO:
		if (val)
			inet->cmsg_flags |= IP_CMSG_PKTINFO;
		else
			inet->cmsg_flags &= ~IP_CMSG_PKTINFO;
		break;
	case IP_RECVTTL:
		if (val)
			inet->cmsg_flags |=  IP_CMSG_TTL;
		else
			inet->cmsg_flags &= ~IP_CMSG_TTL;
		break;
	case IP_RECVTOS:
		if (val)
			inet->cmsg_flags |=  IP_CMSG_TOS;
		else
			inet->cmsg_flags &= ~IP_CMSG_TOS;
		break;
	case IP_RECVOPTS:
		if (val)
			inet->cmsg_flags |=  IP_CMSG_RECVOPTS;
		else
			inet->cmsg_flags &= ~IP_CMSG_RECVOPTS;
		break;
	case IP_RETOPTS:
		if (val)
			inet->cmsg_flags |= IP_CMSG_RETOPTS;
		else
			inet->cmsg_flags &= ~IP_CMSG_RETOPTS;
		break;
	case IP_PASSSEC:
		if (val)
			inet->cmsg_flags |= IP_CMSG_PASSSEC;
		else
			inet->cmsg_flags &= ~IP_CMSG_PASSSEC;
		break;
705 706 707 708 709 710
	case IP_RECVORIGDSTADDR:
		if (val)
			inet->cmsg_flags |= IP_CMSG_ORIGDSTADDR;
		else
			inet->cmsg_flags &= ~IP_CMSG_ORIGDSTADDR;
		break;
711 712 713 714 715 716 717 718 719 720 721 722 723
	case IP_CHECKSUM:
		if (val) {
			if (!(inet->cmsg_flags & IP_CMSG_CHECKSUM)) {
				inet_inc_convert_csum(sk);
				inet->cmsg_flags |= IP_CMSG_CHECKSUM;
			}
		} else {
			if (inet->cmsg_flags & IP_CMSG_CHECKSUM) {
				inet_dec_convert_csum(sk);
				inet->cmsg_flags &= ~IP_CMSG_CHECKSUM;
			}
		}
		break;
S
Stephen Hemminger 已提交
724 725
	case IP_TOS:	/* This sets both TOS and Precedence */
		if (sk->sk_type == SOCK_STREAM) {
726 727
			val &= ~INET_ECN_MASK;
			val |= inet->tos & INET_ECN_MASK;
S
Stephen Hemminger 已提交
728 729 730 731 732 733 734 735
		}
		if (inet->tos != val) {
			inet->tos = val;
			sk->sk_priority = rt_tos2priority(val);
			sk_dst_reset(sk);
		}
		break;
	case IP_TTL:
E
Eric Dumazet 已提交
736
		if (optlen < 1)
S
Stephen Hemminger 已提交
737
			goto e_inval;
738
		if (val != -1 && (val < 1 || val > 255))
S
Stephen Hemminger 已提交
739 740 741 742 743 744
			goto e_inval;
		inet->uc_ttl = val;
		break;
	case IP_HDRINCL:
		if (sk->sk_type != SOCK_RAW) {
			err = -ENOPROTOOPT;
C
Catherine Zhang 已提交
745
			break;
S
Stephen Hemminger 已提交
746 747 748
		}
		inet->hdrincl = val ? 1 : 0;
		break;
749 750 751 752 753 754 755
	case IP_NODEFRAG:
		if (sk->sk_type != SOCK_RAW) {
			err = -ENOPROTOOPT;
			break;
		}
		inet->nodefrag = val ? 1 : 0;
		break;
756 757 758
	case IP_BIND_ADDRESS_NO_PORT:
		inet->bind_address_no_port = val ? 1 : 0;
		break;
S
Stephen Hemminger 已提交
759
	case IP_MTU_DISCOVER:
760
		if (val < IP_PMTUDISC_DONT || val > IP_PMTUDISC_OMIT)
S
Stephen Hemminger 已提交
761 762 763 764 765 766 767 768 769 770 771
			goto e_inval;
		inet->pmtudisc = val;
		break;
	case IP_RECVERR:
		inet->recverr = !!val;
		if (!val)
			skb_queue_purge(&sk->sk_error_queue);
		break;
	case IP_MULTICAST_TTL:
		if (sk->sk_type == SOCK_STREAM)
			goto e_inval;
E
Eric Dumazet 已提交
772
		if (optlen < 1)
S
Stephen Hemminger 已提交
773
			goto e_inval;
774
		if (val == -1)
S
Stephen Hemminger 已提交
775 776 777 778 779 780
			val = 1;
		if (val < 0 || val > 255)
			goto e_inval;
		inet->mc_ttl = val;
		break;
	case IP_MULTICAST_LOOP:
E
Eric Dumazet 已提交
781
		if (optlen < 1)
S
Stephen Hemminger 已提交
782 783 784
			goto e_inval;
		inet->mc_loop = !!val;
		break;
785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813
	case IP_UNICAST_IF:
	{
		struct net_device *dev = NULL;
		int ifindex;

		if (optlen != sizeof(int))
			goto e_inval;

		ifindex = (__force int)ntohl((__force __be32)val);
		if (ifindex == 0) {
			inet->uc_index = 0;
			err = 0;
			break;
		}

		dev = dev_get_by_index(sock_net(sk), ifindex);
		err = -EADDRNOTAVAIL;
		if (!dev)
			break;
		dev_put(dev);

		err = -EINVAL;
		if (sk->sk_bound_dev_if)
			break;

		inet->uc_index = ifindex;
		err = 0;
		break;
	}
S
Stephen Hemminger 已提交
814 815 816 817 818 819 820 821 822 823 824
	case IP_MULTICAST_IF:
	{
		struct ip_mreqn mreq;
		struct net_device *dev = NULL;

		if (sk->sk_type == SOCK_STREAM)
			goto e_inval;
		/*
		 *	Check the arguments are allowable
		 */

825 826 827
		if (optlen < sizeof(struct in_addr))
			goto e_inval;

S
Stephen Hemminger 已提交
828 829
		err = -EFAULT;
		if (optlen >= sizeof(struct ip_mreqn)) {
830
			if (copy_from_user(&mreq, optval, sizeof(mreq)))
L
Linus Torvalds 已提交
831
				break;
S
Stephen Hemminger 已提交
832 833
		} else {
			memset(&mreq, 0, sizeof(mreq));
834 835 836 837 838 839 840 841 842
			if (optlen >= sizeof(struct ip_mreq)) {
				if (copy_from_user(&mreq, optval,
						   sizeof(struct ip_mreq)))
					break;
			} else if (optlen >= sizeof(struct in_addr)) {
				if (copy_from_user(&mreq.imr_address, optval,
						   sizeof(struct in_addr)))
					break;
			}
S
Stephen Hemminger 已提交
843 844 845
		}

		if (!mreq.imr_ifindex) {
A
Al Viro 已提交
846
			if (mreq.imr_address.s_addr == htonl(INADDR_ANY)) {
S
Stephen Hemminger 已提交
847 848 849
				inet->mc_index = 0;
				inet->mc_addr  = 0;
				err = 0;
L
Linus Torvalds 已提交
850 851
				break;
			}
852
			dev = ip_dev_find(sock_net(sk), mreq.imr_address.s_addr);
E
Eric Dumazet 已提交
853
			if (dev)
S
Stephen Hemminger 已提交
854 855
				mreq.imr_ifindex = dev->ifindex;
		} else
E
Eric Dumazet 已提交
856
			dev = dev_get_by_index(sock_net(sk), mreq.imr_ifindex);
L
Linus Torvalds 已提交
857 858


S
Stephen Hemminger 已提交
859 860 861
		err = -EADDRNOTAVAIL;
		if (!dev)
			break;
E
Eric Dumazet 已提交
862
		dev_put(dev);
S
Stephen Hemminger 已提交
863 864 865 866 867

		err = -EINVAL;
		if (sk->sk_bound_dev_if &&
		    mreq.imr_ifindex != sk->sk_bound_dev_if)
			break;
L
Linus Torvalds 已提交
868

S
Stephen Hemminger 已提交
869 870 871 872 873
		inet->mc_index = mreq.imr_ifindex;
		inet->mc_addr  = mreq.imr_address.s_addr;
		err = 0;
		break;
	}
L
Linus Torvalds 已提交
874

S
Stephen Hemminger 已提交
875 876 877 878
	case IP_ADD_MEMBERSHIP:
	case IP_DROP_MEMBERSHIP:
	{
		struct ip_mreqn mreq;
L
Linus Torvalds 已提交
879

880 881 882 883
		err = -EPROTO;
		if (inet_sk(sk)->is_icsk)
			break;

S
Stephen Hemminger 已提交
884 885 886 887
		if (optlen < sizeof(struct ip_mreq))
			goto e_inval;
		err = -EFAULT;
		if (optlen >= sizeof(struct ip_mreqn)) {
888
			if (copy_from_user(&mreq, optval, sizeof(mreq)))
L
Linus Torvalds 已提交
889
				break;
S
Stephen Hemminger 已提交
890 891
		} else {
			memset(&mreq, 0, sizeof(mreq));
892
			if (copy_from_user(&mreq, optval, sizeof(struct ip_mreq)))
L
Linus Torvalds 已提交
893
				break;
S
Stephen Hemminger 已提交
894
		}
L
Linus Torvalds 已提交
895

S
Stephen Hemminger 已提交
896
		if (optname == IP_ADD_MEMBERSHIP)
897
			err = ip_mc_join_group(sk, &mreq);
S
Stephen Hemminger 已提交
898
		else
899
			err = ip_mc_leave_group(sk, &mreq);
S
Stephen Hemminger 已提交
900 901 902 903 904 905 906 907 908 909
		break;
	}
	case IP_MSFILTER:
	{
		struct ip_msfilter *msf;

		if (optlen < IP_MSFILTER_SIZE(0))
			goto e_inval;
		if (optlen > sysctl_optmem_max) {
			err = -ENOBUFS;
L
Linus Torvalds 已提交
910 911
			break;
		}
S
Stephen Hemminger 已提交
912
		msf = kmalloc(optlen, GFP_KERNEL);
S
Stephen Hemminger 已提交
913
		if (!msf) {
S
Stephen Hemminger 已提交
914 915 916 917 918 919 920 921 922 923
			err = -ENOBUFS;
			break;
		}
		err = -EFAULT;
		if (copy_from_user(msf, optval, optlen)) {
			kfree(msf);
			break;
		}
		/* numsrc >= (1G-4) overflow in 32 bits */
		if (msf->imsf_numsrc >= 0x3ffffffcU ||
924
		    msf->imsf_numsrc > net->ipv4.sysctl_igmp_max_msf) {
S
Stephen Hemminger 已提交
925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944
			kfree(msf);
			err = -ENOBUFS;
			break;
		}
		if (IP_MSFILTER_SIZE(msf->imsf_numsrc) > optlen) {
			kfree(msf);
			err = -EINVAL;
			break;
		}
		err = ip_mc_msfilter(sk, msf, 0);
		kfree(msf);
		break;
	}
	case IP_BLOCK_SOURCE:
	case IP_UNBLOCK_SOURCE:
	case IP_ADD_SOURCE_MEMBERSHIP:
	case IP_DROP_SOURCE_MEMBERSHIP:
	{
		struct ip_mreq_source mreqs;
		int omode, add;
L
Linus Torvalds 已提交
945

S
Stephen Hemminger 已提交
946 947 948
		if (optlen != sizeof(struct ip_mreq_source))
			goto e_inval;
		if (copy_from_user(&mreqs, optval, sizeof(mreqs))) {
L
Linus Torvalds 已提交
949 950 951
			err = -EFAULT;
			break;
		}
S
Stephen Hemminger 已提交
952 953 954 955 956 957 958 959
		if (optname == IP_BLOCK_SOURCE) {
			omode = MCAST_EXCLUDE;
			add = 1;
		} else if (optname == IP_UNBLOCK_SOURCE) {
			omode = MCAST_EXCLUDE;
			add = 0;
		} else if (optname == IP_ADD_SOURCE_MEMBERSHIP) {
			struct ip_mreqn mreq;
L
Linus Torvalds 已提交
960

S
Stephen Hemminger 已提交
961 962 963
			mreq.imr_multiaddr.s_addr = mreqs.imr_multiaddr;
			mreq.imr_address.s_addr = mreqs.imr_interface;
			mreq.imr_ifindex = 0;
964
			err = ip_mc_join_group(sk, &mreq);
S
Stephen Hemminger 已提交
965
			if (err && err != -EADDRINUSE)
L
Linus Torvalds 已提交
966
				break;
S
Stephen Hemminger 已提交
967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995
			omode = MCAST_INCLUDE;
			add = 1;
		} else /* IP_DROP_SOURCE_MEMBERSHIP */ {
			omode = MCAST_INCLUDE;
			add = 0;
		}
		err = ip_mc_source(add, omode, sk, &mreqs, 0);
		break;
	}
	case MCAST_JOIN_GROUP:
	case MCAST_LEAVE_GROUP:
	{
		struct group_req greq;
		struct sockaddr_in *psin;
		struct ip_mreqn mreq;

		if (optlen < sizeof(struct group_req))
			goto e_inval;
		err = -EFAULT;
		if (copy_from_user(&greq, optval, sizeof(greq)))
			break;
		psin = (struct sockaddr_in *)&greq.gr_group;
		if (psin->sin_family != AF_INET)
			goto e_inval;
		memset(&mreq, 0, sizeof(mreq));
		mreq.imr_multiaddr = psin->sin_addr;
		mreq.imr_ifindex = greq.gr_interface;

		if (optname == MCAST_JOIN_GROUP)
996
			err = ip_mc_join_group(sk, &mreq);
S
Stephen Hemminger 已提交
997
		else
998
			err = ip_mc_leave_group(sk, &mreq);
S
Stephen Hemminger 已提交
999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013
		break;
	}
	case MCAST_JOIN_SOURCE_GROUP:
	case MCAST_LEAVE_SOURCE_GROUP:
	case MCAST_BLOCK_SOURCE:
	case MCAST_UNBLOCK_SOURCE:
	{
		struct group_source_req greqs;
		struct ip_mreq_source mreqs;
		struct sockaddr_in *psin;
		int omode, add;

		if (optlen != sizeof(struct group_source_req))
			goto e_inval;
		if (copy_from_user(&greqs, optval, sizeof(greqs))) {
L
Linus Torvalds 已提交
1014 1015 1016
			err = -EFAULT;
			break;
		}
S
Stephen Hemminger 已提交
1017 1018 1019
		if (greqs.gsr_group.ss_family != AF_INET ||
		    greqs.gsr_source.ss_family != AF_INET) {
			err = -EADDRNOTAVAIL;
L
Linus Torvalds 已提交
1020 1021
			break;
		}
S
Stephen Hemminger 已提交
1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034
		psin = (struct sockaddr_in *)&greqs.gsr_group;
		mreqs.imr_multiaddr = psin->sin_addr.s_addr;
		psin = (struct sockaddr_in *)&greqs.gsr_source;
		mreqs.imr_sourceaddr = psin->sin_addr.s_addr;
		mreqs.imr_interface = 0; /* use index for mc_source */

		if (optname == MCAST_BLOCK_SOURCE) {
			omode = MCAST_EXCLUDE;
			add = 1;
		} else if (optname == MCAST_UNBLOCK_SOURCE) {
			omode = MCAST_EXCLUDE;
			add = 0;
		} else if (optname == MCAST_JOIN_SOURCE_GROUP) {
L
Linus Torvalds 已提交
1035 1036
			struct ip_mreqn mreq;

S
Stephen Hemminger 已提交
1037
			psin = (struct sockaddr_in *)&greqs.gsr_group;
L
Linus Torvalds 已提交
1038
			mreq.imr_multiaddr = psin->sin_addr;
S
Stephen Hemminger 已提交
1039 1040
			mreq.imr_address.s_addr = 0;
			mreq.imr_ifindex = greqs.gsr_interface;
1041
			err = ip_mc_join_group(sk, &mreq);
S
Stephen Hemminger 已提交
1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065
			if (err && err != -EADDRINUSE)
				break;
			greqs.gsr_interface = mreq.imr_ifindex;
			omode = MCAST_INCLUDE;
			add = 1;
		} else /* MCAST_LEAVE_SOURCE_GROUP */ {
			omode = MCAST_INCLUDE;
			add = 0;
		}
		err = ip_mc_source(add, omode, sk, &mreqs,
				   greqs.gsr_interface);
		break;
	}
	case MCAST_MSFILTER:
	{
		struct sockaddr_in *psin;
		struct ip_msfilter *msf = NULL;
		struct group_filter *gsf = NULL;
		int msize, i, ifindex;

		if (optlen < GROUP_FILTER_SIZE(0))
			goto e_inval;
		if (optlen > sysctl_optmem_max) {
			err = -ENOBUFS;
L
Linus Torvalds 已提交
1066 1067
			break;
		}
1068
		gsf = kmalloc(optlen, GFP_KERNEL);
S
Stephen Hemminger 已提交
1069
		if (!gsf) {
S
Stephen Hemminger 已提交
1070
			err = -ENOBUFS;
L
Linus Torvalds 已提交
1071 1072
			break;
		}
S
Stephen Hemminger 已提交
1073
		err = -EFAULT;
E
Eric Dumazet 已提交
1074
		if (copy_from_user(gsf, optval, optlen))
S
Stephen Hemminger 已提交
1075
			goto mc_msf_out;
E
Eric Dumazet 已提交
1076

S
Stephen Hemminger 已提交
1077 1078
		/* numsrc >= (4G-140)/128 overflow in 32 bits */
		if (gsf->gf_numsrc >= 0x1ffffff ||
1079
		    gsf->gf_numsrc > net->ipv4.sysctl_igmp_max_msf) {
S
Stephen Hemminger 已提交
1080 1081 1082 1083 1084 1085 1086 1087
			err = -ENOBUFS;
			goto mc_msf_out;
		}
		if (GROUP_FILTER_SIZE(gsf->gf_numsrc) > optlen) {
			err = -EINVAL;
			goto mc_msf_out;
		}
		msize = IP_MSFILTER_SIZE(gsf->gf_numsrc);
1088
		msf = kmalloc(msize, GFP_KERNEL);
S
Stephen Hemminger 已提交
1089
		if (!msf) {
S
Stephen Hemminger 已提交
1090 1091 1092 1093 1094 1095
			err = -ENOBUFS;
			goto mc_msf_out;
		}
		ifindex = gsf->gf_interface;
		psin = (struct sockaddr_in *)&gsf->gf_group;
		if (psin->sin_family != AF_INET) {
L
Linus Torvalds 已提交
1096
			err = -EADDRNOTAVAIL;
S
Stephen Hemminger 已提交
1097
			goto mc_msf_out;
L
Linus Torvalds 已提交
1098
		}
S
Stephen Hemminger 已提交
1099 1100 1101 1102 1103
		msf->imsf_multiaddr = psin->sin_addr.s_addr;
		msf->imsf_interface = 0;
		msf->imsf_fmode = gsf->gf_fmode;
		msf->imsf_numsrc = gsf->gf_numsrc;
		err = -EADDRNOTAVAIL;
E
Eric Dumazet 已提交
1104
		for (i = 0; i < gsf->gf_numsrc; ++i) {
S
Stephen Hemminger 已提交
1105
			psin = (struct sockaddr_in *)&gsf->gf_slist[i];
1106

S
Stephen Hemminger 已提交
1107 1108 1109 1110 1111 1112 1113 1114
			if (psin->sin_family != AF_INET)
				goto mc_msf_out;
			msf->imsf_slist[i] = psin->sin_addr.s_addr;
		}
		kfree(gsf);
		gsf = NULL;

		err = ip_mc_msfilter(sk, msf, ifindex);
E
Eric Dumazet 已提交
1115
mc_msf_out:
S
Stephen Hemminger 已提交
1116 1117 1118 1119
		kfree(msf);
		kfree(gsf);
		break;
	}
1120 1121 1122 1123 1124 1125 1126
	case IP_MULTICAST_ALL:
		if (optlen < 1)
			goto e_inval;
		if (val != 0 && val != 1)
			goto e_inval;
		inet->mc_all = val;
		break;
S
Stephen Hemminger 已提交
1127 1128 1129 1130 1131
	case IP_ROUTER_ALERT:
		err = ip_ra_control(sk, val ? 1 : 0, NULL);
		break;

	case IP_FREEBIND:
E
Eric Dumazet 已提交
1132
		if (optlen < 1)
S
Stephen Hemminger 已提交
1133 1134 1135 1136 1137 1138 1139
			goto e_inval;
		inet->freebind = !!val;
		break;

	case IP_IPSEC_POLICY:
	case IP_XFRM_POLICY:
		err = -EPERM;
1140
		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
L
Linus Torvalds 已提交
1141
			break;
S
Stephen Hemminger 已提交
1142 1143
		err = xfrm_user_policy(sk, optname, optval, optlen);
		break;
L
Linus Torvalds 已提交
1144

1145
	case IP_TRANSPARENT:
1146 1147
		if (!!val && !ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1148 1149 1150 1151 1152 1153 1154 1155
			err = -EPERM;
			break;
		}
		if (optlen < 1)
			goto e_inval;
		inet->transparent = !!val;
		break;

1156 1157 1158 1159 1160 1161 1162 1163
	case IP_MINTTL:
		if (optlen < 1)
			goto e_inval;
		if (val < 0 || val > 255)
			goto e_inval;
		inet->min_ttl = val;
		break;

S
Stephen Hemminger 已提交
1164 1165 1166
	default:
		err = -ENOPROTOOPT;
		break;
L
Linus Torvalds 已提交
1167 1168
	}
	release_sock(sk);
1169 1170
	if (needs_rtnl)
		rtnl_unlock();
L
Linus Torvalds 已提交
1171 1172 1173 1174
	return err;

e_inval:
	release_sock(sk);
1175 1176
	if (needs_rtnl)
		rtnl_unlock();
L
Linus Torvalds 已提交
1177 1178 1179
	return -EINVAL;
}

E
Eric Dumazet 已提交
1180
/**
1181
 * ipv4_pktinfo_prepare - transfer some info from rtable to skb
E
Eric Dumazet 已提交
1182 1183 1184
 * @sk: socket
 * @skb: buffer
 *
1185 1186
 * To support IP_CMSG_PKTINFO option, we store rt_iif and specific
 * destination in skb->cb[] before dst drop.
S
stephen hemminger 已提交
1187
 * This way, receiver doesn't make cache line misses to read rtable.
E
Eric Dumazet 已提交
1188
 */
1189
void ipv4_pktinfo_prepare(const struct sock *sk, struct sk_buff *skb)
E
Eric Dumazet 已提交
1190
{
1191
	struct in_pktinfo *pktinfo = PKTINFO_SKB_CB(skb);
1192 1193
	bool prepare = (inet_sk(sk)->cmsg_flags & IP_CMSG_PKTINFO) ||
		       ipv6_sk_rxinfo(sk);
1194

1195
	if (prepare && skb_rtable(skb)) {
1196
		pktinfo->ipi_ifindex = inet_iif(skb);
1197
		pktinfo->ipi_spec_dst.s_addr = fib_compute_spec_dst(skb);
1198 1199 1200 1201 1202
	} else {
		pktinfo->ipi_ifindex = 0;
		pktinfo->ipi_spec_dst.s_addr = 0;
	}
	skb_dst_drop(skb);
E
Eric Dumazet 已提交
1203 1204
}

1205
int ip_setsockopt(struct sock *sk, int level,
1206
		int optname, char __user *optval, unsigned int optlen)
1207 1208 1209 1210 1211 1212 1213 1214 1215 1216
{
	int err;

	if (level != SOL_IP)
		return -ENOPROTOOPT;

	err = do_ip_setsockopt(sk, level, optname, optval, optlen);
#ifdef CONFIG_NETFILTER
	/* we need to exclude all possible ENOPROTOOPTs except default case */
	if (err == -ENOPROTOOPT && optname != IP_HDRINCL &&
1217 1218 1219
			optname != IP_IPSEC_POLICY &&
			optname != IP_XFRM_POLICY &&
			!ip_mroute_opt(optname)) {
1220 1221 1222 1223 1224 1225 1226
		lock_sock(sk);
		err = nf_setsockopt(sk, PF_INET, optname, optval, optlen);
		release_sock(sk);
	}
#endif
	return err;
}
E
Eric Dumazet 已提交
1227
EXPORT_SYMBOL(ip_setsockopt);
1228 1229

#ifdef CONFIG_COMPAT
1230
int compat_ip_setsockopt(struct sock *sk, int level, int optname,
1231
			 char __user *optval, unsigned int optlen)
1232 1233 1234 1235 1236 1237
{
	int err;

	if (level != SOL_IP)
		return -ENOPROTOOPT;

1238 1239 1240 1241
	if (optname >= MCAST_JOIN_GROUP && optname <= MCAST_MSFILTER)
		return compat_mc_setsockopt(sk, level, optname, optval, optlen,
			ip_setsockopt);

1242 1243 1244 1245
	err = do_ip_setsockopt(sk, level, optname, optval, optlen);
#ifdef CONFIG_NETFILTER
	/* we need to exclude all possible ENOPROTOOPTs except default case */
	if (err == -ENOPROTOOPT && optname != IP_HDRINCL &&
1246 1247 1248
			optname != IP_IPSEC_POLICY &&
			optname != IP_XFRM_POLICY &&
			!ip_mroute_opt(optname)) {
1249
		lock_sock(sk);
1250 1251
		err = compat_nf_setsockopt(sk, PF_INET, optname,
					   optval, optlen);
1252 1253 1254 1255 1256
		release_sock(sk);
	}
#endif
	return err;
}
1257
EXPORT_SYMBOL(compat_ip_setsockopt);
1258 1259
#endif

L
Linus Torvalds 已提交
1260
/*
E
Eric Dumazet 已提交
1261 1262
 *	Get the options. Note for future reference. The GET of IP options gets
 *	the _received_ ones. The set sets the _sent_ ones.
L
Linus Torvalds 已提交
1263 1264
 */

1265 1266 1267 1268 1269 1270 1271 1272 1273 1274
static bool getsockopt_needs_rtnl(int optname)
{
	switch (optname) {
	case IP_MSFILTER:
	case MCAST_MSFILTER:
		return true;
	}
	return false;
}

1275
static int do_ip_getsockopt(struct sock *sk, int level, int optname,
1276
			    char __user *optval, int __user *optlen, unsigned int flags)
L
Linus Torvalds 已提交
1277 1278
{
	struct inet_sock *inet = inet_sk(sk);
1279 1280
	bool needs_rtnl = getsockopt_needs_rtnl(optname);
	int val, err = 0;
L
Linus Torvalds 已提交
1281
	int len;
1282

S
Stephen Hemminger 已提交
1283
	if (level != SOL_IP)
L
Linus Torvalds 已提交
1284 1285
		return -EOPNOTSUPP;

1286
	if (ip_mroute_opt(optname))
1287
		return ip_mroute_getsockopt(sk, optname, optval, optlen);
L
Linus Torvalds 已提交
1288

1289
	if (get_user(len, optlen))
L
Linus Torvalds 已提交
1290
		return -EFAULT;
S
Stephen Hemminger 已提交
1291
	if (len < 0)
L
Linus Torvalds 已提交
1292
		return -EINVAL;
1293

1294 1295
	if (needs_rtnl)
		rtnl_lock();
L
Linus Torvalds 已提交
1296 1297
	lock_sock(sk);

S
Stephen Hemminger 已提交
1298 1299 1300 1301
	switch (optname) {
	case IP_OPTIONS:
	{
		unsigned char optbuf[sizeof(struct ip_options)+40];
1302 1303 1304 1305
		struct ip_options *opt = (struct ip_options *)optbuf;
		struct ip_options_rcu *inet_opt;

		inet_opt = rcu_dereference_protected(inet->inet_opt,
1306
						     lockdep_sock_is_held(sk));
S
Stephen Hemminger 已提交
1307
		opt->optlen = 0;
1308 1309 1310 1311
		if (inet_opt)
			memcpy(optbuf, &inet_opt->opt,
			       sizeof(struct ip_options) +
			       inet_opt->opt.optlen);
S
Stephen Hemminger 已提交
1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343
		release_sock(sk);

		if (opt->optlen == 0)
			return put_user(0, optlen);

		ip_options_undo(opt);

		len = min_t(unsigned int, len, opt->optlen);
		if (put_user(len, optlen))
			return -EFAULT;
		if (copy_to_user(optval, opt->__data, len))
			return -EFAULT;
		return 0;
	}
	case IP_PKTINFO:
		val = (inet->cmsg_flags & IP_CMSG_PKTINFO) != 0;
		break;
	case IP_RECVTTL:
		val = (inet->cmsg_flags & IP_CMSG_TTL) != 0;
		break;
	case IP_RECVTOS:
		val = (inet->cmsg_flags & IP_CMSG_TOS) != 0;
		break;
	case IP_RECVOPTS:
		val = (inet->cmsg_flags & IP_CMSG_RECVOPTS) != 0;
		break;
	case IP_RETOPTS:
		val = (inet->cmsg_flags & IP_CMSG_RETOPTS) != 0;
		break;
	case IP_PASSSEC:
		val = (inet->cmsg_flags & IP_CMSG_PASSSEC) != 0;
		break;
1344 1345 1346
	case IP_RECVORIGDSTADDR:
		val = (inet->cmsg_flags & IP_CMSG_ORIGDSTADDR) != 0;
		break;
1347 1348 1349
	case IP_CHECKSUM:
		val = (inet->cmsg_flags & IP_CMSG_CHECKSUM) != 0;
		break;
S
Stephen Hemminger 已提交
1350 1351 1352 1353
	case IP_TOS:
		val = inet->tos;
		break;
	case IP_TTL:
1354 1355
	{
		struct net *net = sock_net(sk);
S
Stephen Hemminger 已提交
1356
		val = (inet->uc_ttl == -1 ?
1357
		       net->ipv4.sysctl_ip_default_ttl :
S
Stephen Hemminger 已提交
1358 1359
		       inet->uc_ttl);
		break;
1360
	}
S
Stephen Hemminger 已提交
1361 1362 1363
	case IP_HDRINCL:
		val = inet->hdrincl;
		break;
1364 1365 1366
	case IP_NODEFRAG:
		val = inet->nodefrag;
		break;
1367 1368 1369
	case IP_BIND_ADDRESS_NO_PORT:
		val = inet->bind_address_no_port;
		break;
S
Stephen Hemminger 已提交
1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380
	case IP_MTU_DISCOVER:
		val = inet->pmtudisc;
		break;
	case IP_MTU:
	{
		struct dst_entry *dst;
		val = 0;
		dst = sk_dst_get(sk);
		if (dst) {
			val = dst_mtu(dst);
			dst_release(dst);
L
Linus Torvalds 已提交
1381
		}
S
Stephen Hemminger 已提交
1382
		if (!val) {
L
Linus Torvalds 已提交
1383
			release_sock(sk);
S
Stephen Hemminger 已提交
1384
			return -ENOTCONN;
L
Linus Torvalds 已提交
1385
		}
S
Stephen Hemminger 已提交
1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396
		break;
	}
	case IP_RECVERR:
		val = inet->recverr;
		break;
	case IP_MULTICAST_TTL:
		val = inet->mc_ttl;
		break;
	case IP_MULTICAST_LOOP:
		val = inet->mc_loop;
		break;
1397 1398 1399
	case IP_UNICAST_IF:
		val = (__force int)htonl((__u32) inet->uc_index);
		break;
S
Stephen Hemminger 已提交
1400 1401 1402 1403 1404 1405
	case IP_MULTICAST_IF:
	{
		struct in_addr addr;
		len = min_t(unsigned int, len, sizeof(struct in_addr));
		addr.s_addr = inet->mc_addr;
		release_sock(sk);
L
Linus Torvalds 已提交
1406

S
Stephen Hemminger 已提交
1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417
		if (put_user(len, optlen))
			return -EFAULT;
		if (copy_to_user(optval, &addr, len))
			return -EFAULT;
		return 0;
	}
	case IP_MSFILTER:
	{
		struct ip_msfilter msf;

		if (len < IP_MSFILTER_SIZE(0)) {
1418 1419
			err = -EINVAL;
			goto out;
L
Linus Torvalds 已提交
1420
		}
S
Stephen Hemminger 已提交
1421
		if (copy_from_user(&msf, optval, IP_MSFILTER_SIZE(0))) {
1422 1423
			err = -EFAULT;
			goto out;
L
Linus Torvalds 已提交
1424
		}
S
Stephen Hemminger 已提交
1425 1426
		err = ip_mc_msfget(sk, &msf,
				   (struct ip_msfilter __user *)optval, optlen);
1427
		goto out;
S
Stephen Hemminger 已提交
1428 1429 1430 1431
	}
	case MCAST_MSFILTER:
	{
		struct group_filter gsf;
L
Linus Torvalds 已提交
1432

S
Stephen Hemminger 已提交
1433
		if (len < GROUP_FILTER_SIZE(0)) {
1434 1435
			err = -EINVAL;
			goto out;
S
Stephen Hemminger 已提交
1436 1437
		}
		if (copy_from_user(&gsf, optval, GROUP_FILTER_SIZE(0))) {
1438 1439
			err = -EFAULT;
			goto out;
S
Stephen Hemminger 已提交
1440 1441
		}
		err = ip_mc_gsfget(sk, &gsf,
E
Eric Dumazet 已提交
1442 1443
				   (struct group_filter __user *)optval,
				   optlen);
1444
		goto out;
S
Stephen Hemminger 已提交
1445
	}
1446 1447 1448
	case IP_MULTICAST_ALL:
		val = inet->mc_all;
		break;
S
Stephen Hemminger 已提交
1449 1450 1451
	case IP_PKTOPTIONS:
	{
		struct msghdr msg;
L
Linus Torvalds 已提交
1452

S
Stephen Hemminger 已提交
1453
		release_sock(sk);
L
Linus Torvalds 已提交
1454

S
Stephen Hemminger 已提交
1455 1456
		if (sk->sk_type != SOCK_STREAM)
			return -ENOPROTOOPT;
L
Linus Torvalds 已提交
1457

1458
		msg.msg_control = (__force void *) optval;
S
Stephen Hemminger 已提交
1459
		msg.msg_controllen = len;
1460
		msg.msg_flags = flags;
L
Linus Torvalds 已提交
1461

S
Stephen Hemminger 已提交
1462 1463 1464
		if (inet->cmsg_flags & IP_CMSG_PKTINFO) {
			struct in_pktinfo info;

E
Eric Dumazet 已提交
1465 1466
			info.ipi_addr.s_addr = inet->inet_rcv_saddr;
			info.ipi_spec_dst.s_addr = inet->inet_rcv_saddr;
S
Stephen Hemminger 已提交
1467 1468
			info.ipi_ifindex = inet->mc_index;
			put_cmsg(&msg, SOL_IP, IP_PKTINFO, sizeof(info), &info);
L
Linus Torvalds 已提交
1469
		}
S
Stephen Hemminger 已提交
1470 1471 1472 1473
		if (inet->cmsg_flags & IP_CMSG_TTL) {
			int hlim = inet->mc_ttl;
			put_cmsg(&msg, SOL_IP, IP_TTL, sizeof(hlim), &hlim);
		}
1474 1475 1476 1477
		if (inet->cmsg_flags & IP_CMSG_TOS) {
			int tos = inet->rcv_tos;
			put_cmsg(&msg, SOL_IP, IP_TOS, sizeof(tos), &tos);
		}
S
Stephen Hemminger 已提交
1478 1479 1480 1481 1482 1483
		len -= msg.msg_controllen;
		return put_user(len, optlen);
	}
	case IP_FREEBIND:
		val = inet->freebind;
		break;
1484 1485 1486
	case IP_TRANSPARENT:
		val = inet->transparent;
		break;
1487 1488 1489
	case IP_MINTTL:
		val = inet->min_ttl;
		break;
S
Stephen Hemminger 已提交
1490 1491 1492
	default:
		release_sock(sk);
		return -ENOPROTOOPT;
L
Linus Torvalds 已提交
1493 1494
	}
	release_sock(sk);
1495

E
Eric Dumazet 已提交
1496
	if (len < sizeof(int) && len > 0 && val >= 0 && val <= 255) {
L
Linus Torvalds 已提交
1497 1498
		unsigned char ucval = (unsigned char)val;
		len = 1;
S
Stephen Hemminger 已提交
1499
		if (put_user(len, optlen))
L
Linus Torvalds 已提交
1500
			return -EFAULT;
1501
		if (copy_to_user(optval, &ucval, 1))
L
Linus Torvalds 已提交
1502 1503 1504
			return -EFAULT;
	} else {
		len = min_t(unsigned int, sizeof(int), len);
S
Stephen Hemminger 已提交
1505
		if (put_user(len, optlen))
L
Linus Torvalds 已提交
1506
			return -EFAULT;
1507
		if (copy_to_user(optval, &val, len))
L
Linus Torvalds 已提交
1508 1509 1510
			return -EFAULT;
	}
	return 0;
1511 1512 1513 1514 1515 1516

out:
	release_sock(sk);
	if (needs_rtnl)
		rtnl_unlock();
	return err;
L
Linus Torvalds 已提交
1517 1518
}

1519
int ip_getsockopt(struct sock *sk, int level,
S
Stephen Hemminger 已提交
1520
		  int optname, char __user *optval, int __user *optlen)
1521 1522 1523
{
	int err;

1524
	err = do_ip_getsockopt(sk, level, optname, optval, optlen, 0);
1525 1526
#ifdef CONFIG_NETFILTER
	/* we need to exclude all possible ENOPROTOOPTs except default case */
1527 1528
	if (err == -ENOPROTOOPT && optname != IP_PKTOPTIONS &&
			!ip_mroute_opt(optname)) {
1529
		int len;
1530

1531
		if (get_user(len, optlen))
1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544
			return -EFAULT;

		lock_sock(sk);
		err = nf_getsockopt(sk, PF_INET, optname, optval,
				&len);
		release_sock(sk);
		if (err >= 0)
			err = put_user(len, optlen);
		return err;
	}
#endif
	return err;
}
E
Eric Dumazet 已提交
1545
EXPORT_SYMBOL(ip_getsockopt);
1546 1547

#ifdef CONFIG_COMPAT
1548 1549
int compat_ip_getsockopt(struct sock *sk, int level, int optname,
			 char __user *optval, int __user *optlen)
1550
{
1551 1552 1553 1554 1555 1556
	int err;

	if (optname == MCAST_MSFILTER)
		return compat_mc_getsockopt(sk, level, optname, optval, optlen,
			ip_getsockopt);

1557 1558
	err = do_ip_getsockopt(sk, level, optname, optval, optlen,
		MSG_CMSG_COMPAT);
1559

1560 1561
#ifdef CONFIG_NETFILTER
	/* we need to exclude all possible ENOPROTOOPTs except default case */
1562 1563
	if (err == -ENOPROTOOPT && optname != IP_PKTOPTIONS &&
			!ip_mroute_opt(optname)) {
1564
		int len;
1565

1566
		if (get_user(len, optlen))
1567 1568 1569
			return -EFAULT;

		lock_sock(sk);
1570
		err = compat_nf_getsockopt(sk, PF_INET, optname, optval, &len);
1571 1572 1573 1574 1575 1576 1577 1578
		release_sock(sk);
		if (err >= 0)
			err = put_user(len, optlen);
		return err;
	}
#endif
	return err;
}
1579
EXPORT_SYMBOL(compat_ip_getsockopt);
1580
#endif