raw.c

   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *	RAW sockets for IPv6
   4 *	Linux INET6 implementation
   5 *
   6 *	Authors:
   7 *	Pedro Roque		<roque@di.fc.ul.pt>
   8 *
   9 *	Adapted from linux/net/ipv4/raw.c
  10 *
  11 *	Fixes:
  12 *	Hideaki YOSHIFUJI	:	sin6_scope_id support
  13 *	YOSHIFUJI,H.@USAGI	:	raw checksum (RFC2292(bis) compliance)
  14 *	Kazunori MIYAZAWA @USAGI:	change process style to use ip6_append_data
  15 */
  16
  17#include <linux/errno.h>
  18#include <linux/types.h>
  19#include <linux/socket.h>
  20#include <linux/slab.h>
  21#include <linux/sockios.h>
  22#include <linux/net.h>
  23#include <linux/in6.h>
  24#include <linux/netdevice.h>
  25#include <linux/if_arp.h>
  26#include <linux/icmpv6.h>
  27#include <linux/netfilter.h>
  28#include <linux/netfilter_ipv6.h>
  29#include <linux/skbuff.h>
  30#include <linux/compat.h>
  31#include <linux/uaccess.h>
  32#include <asm/ioctls.h>
  33
  34#include <net/net_namespace.h>
  35#include <net/ip.h>
  36#include <net/sock.h>
  37#include <net/snmp.h>
  38
  39#include <net/ipv6.h>
  40#include <net/ndisc.h>
  41#include <net/protocol.h>
  42#include <net/ip6_route.h>
  43#include <net/ip6_checksum.h>
  44#include <net/addrconf.h>
  45#include <net/transp_v6.h>
  46#include <net/udp.h>
  47#include <net/inet_common.h>
  48#include <net/tcp_states.h>
  49#if IS_ENABLED(CONFIG_IPV6_MIP6)
  50#include <net/mip6.h>
  51#endif
  52#include <linux/mroute6.h>
  53
  54#include <net/raw.h>
  55#include <net/rawv6.h>
  56#include <net/xfrm.h>
  57
  58#include <linux/proc_fs.h>
  59#include <linux/seq_file.h>
  60#include <linux/export.h>
  61
  62#define	ICMPV6_HDRLEN	4	/* ICMPv6 header, RFC 4443 Section 2.1 */
  63
  64struct raw_hashinfo raw_v6_hashinfo;
  65EXPORT_SYMBOL_GPL(raw_v6_hashinfo);
  66
  67bool raw_v6_match(struct net *net, const struct sock *sk, unsigned short num,
  68		  const struct in6_addr *loc_addr,
  69		  const struct in6_addr *rmt_addr, int dif, int sdif)
  70{
  71	if (inet_sk(sk)->inet_num != num ||
  72	    !net_eq(sock_net(sk), net) ||
  73	    (!ipv6_addr_any(&sk->sk_v6_daddr) &&
  74	     !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
  75	    !raw_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
  76				 dif, sdif))
  77		return false;
  78
  79	if (ipv6_addr_any(&sk->sk_v6_rcv_saddr) ||
  80	    ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr) ||
  81	    (ipv6_addr_is_multicast(loc_addr) &&
  82	     inet6_mc_check(sk, loc_addr, rmt_addr)))
  83		return true;
  84
  85	return false;
  86}
  87EXPORT_SYMBOL_GPL(raw_v6_match);
  88
  89/*
  90 *	0 - deliver
  91 *	1 - block
  92 */
  93static int icmpv6_filter(const struct sock *sk, const struct sk_buff *skb)
  94{
  95	struct icmp6hdr _hdr;
  96	const struct icmp6hdr *hdr;
  97
  98	/* We require only the four bytes of the ICMPv6 header, not any
  99	 * additional bytes of message body in "struct icmp6hdr".
 100	 */
 101	hdr = skb_header_pointer(skb, skb_transport_offset(skb),
 102				 ICMPV6_HDRLEN, &_hdr);
 103	if (hdr) {
 104		const __u32 *data = &raw6_sk(sk)->filter.data[0];
 105		unsigned int type = hdr->icmp6_type;
 106
 107		return (data[type >> 5] & (1U << (type & 31))) != 0;
 108	}
 109	return 1;
 110}
 111
 112#if IS_ENABLED(CONFIG_IPV6_MIP6)
 113typedef int mh_filter_t(struct sock *sock, struct sk_buff *skb);
 114
 115static mh_filter_t __rcu *mh_filter __read_mostly;
 116
 117int rawv6_mh_filter_register(mh_filter_t filter)
 118{
 119	rcu_assign_pointer(mh_filter, filter);
 120	return 0;
 121}
 122EXPORT_SYMBOL(rawv6_mh_filter_register);
 123
 124int rawv6_mh_filter_unregister(mh_filter_t filter)
 125{
 126	RCU_INIT_POINTER(mh_filter, NULL);
 127	synchronize_rcu();
 128	return 0;
 129}
 130EXPORT_SYMBOL(rawv6_mh_filter_unregister);
 131
 132#endif
 133
 134/*
 135 *	demultiplex raw sockets.
 136 *	(should consider queueing the skb in the sock receive_queue
 137 *	without calling rawv6.c)
 138 *
 139 *	Caller owns SKB so we must make clones.
 140 */
 141static bool ipv6_raw_deliver(struct sk_buff *skb, int nexthdr)
 142{
 143	struct net *net = dev_net(skb->dev);
 144	const struct in6_addr *saddr;
 145	const struct in6_addr *daddr;
 146	struct hlist_head *hlist;
 147	struct sock *sk;
 148	bool delivered = false;
 149	__u8 hash;
 150
 151	saddr = &ipv6_hdr(skb)->saddr;
 152	daddr = saddr + 1;
 153
 154	hash = raw_hashfunc(net, nexthdr);
 155	hlist = &raw_v6_hashinfo.ht[hash];
 156	rcu_read_lock();
 157	sk_for_each_rcu(sk, hlist) {
 158		int filtered;
 159
 160		if (!raw_v6_match(net, sk, nexthdr, daddr, saddr,
 161				  inet6_iif(skb), inet6_sdif(skb)))
 162			continue;
 163
 164		if (atomic_read(&sk->sk_rmem_alloc) >=
 165		    READ_ONCE(sk->sk_rcvbuf)) {
 166			sk_drops_inc(sk);
 167			continue;
 168		}
 169
 170		delivered = true;
 171		switch (nexthdr) {
 172		case IPPROTO_ICMPV6:
 173			filtered = icmpv6_filter(sk, skb);
 174			break;
 175
 176#if IS_ENABLED(CONFIG_IPV6_MIP6)
 177		case IPPROTO_MH:
 178		{
 179			/* XXX: To validate MH only once for each packet,
 180			 * this is placed here. It should be after checking
 181			 * xfrm policy, however it doesn't. The checking xfrm
 182			 * policy is placed in rawv6_rcv() because it is
 183			 * required for each socket.
 184			 */
 185			mh_filter_t *filter;
 186
 187			filter = rcu_dereference(mh_filter);
 188			filtered = filter ? (*filter)(sk, skb) : 0;
 189			break;
 190		}
 191#endif
 192		default:
 193			filtered = 0;
 194			break;
 195		}
 196
 197		if (filtered < 0)
 198			break;
 199		if (filtered == 0) {
 200			struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
 201
 202			/* Not releasing hash table! */
 203			if (clone)
 204				rawv6_rcv(sk, clone);
 205		}
 206	}
 207	rcu_read_unlock();
 208	return delivered;
 209}
 210
 211bool raw6_local_deliver(struct sk_buff *skb, int nexthdr)
 212{
 213	return ipv6_raw_deliver(skb, nexthdr);
 214}
 215
 216/* This cleans up af_inet6 a bit. -DaveM */
 217static int rawv6_bind(struct sock *sk, struct sockaddr_unsized *uaddr,
 218		      int addr_len)
 219{
 220	struct inet_sock *inet = inet_sk(sk);
 221	struct ipv6_pinfo *np = inet6_sk(sk);
 222	struct sockaddr_in6 *addr = (struct sockaddr_in6 *) uaddr;
 223	__be32 v4addr = 0;
 224	int addr_type;
 225	int err;
 226
 227	if (addr_len < SIN6_LEN_RFC2133)
 228		return -EINVAL;
 229
 230	if (addr->sin6_family != AF_INET6)
 231		return -EINVAL;
 232
 233	addr_type = ipv6_addr_type(&addr->sin6_addr);
 234
 235	/* Raw sockets are IPv6 only */
 236	if (addr_type == IPV6_ADDR_MAPPED)
 237		return -EADDRNOTAVAIL;
 238
 239	lock_sock(sk);
 240
 241	err = -EINVAL;
 242	if (sk->sk_state != TCP_CLOSE)
 243		goto out;
 244
 245	rcu_read_lock();
 246	/* Check if the address belongs to the host. */
 247	if (addr_type != IPV6_ADDR_ANY) {
 248		struct net_device *dev = NULL;
 249
 250		if (__ipv6_addr_needs_scope_id(addr_type)) {
 251			if (addr_len >= sizeof(struct sockaddr_in6) &&
 252			    addr->sin6_scope_id) {
 253				/* Override any existing binding, if another
 254				 * one is supplied by user.
 255				 */
 256				sk->sk_bound_dev_if = addr->sin6_scope_id;
 257			}
 258
 259			/* Binding to link-local address requires an interface */
 260			if (!sk->sk_bound_dev_if)
 261				goto out_unlock;
 262		}
 263
 264		if (sk->sk_bound_dev_if) {
 265			err = -ENODEV;
 266			dev = dev_get_by_index_rcu(sock_net(sk),
 267						   sk->sk_bound_dev_if);
 268			if (!dev)
 269				goto out_unlock;
 270		}
 271
 272		/* ipv4 addr of the socket is invalid.  Only the
 273		 * unspecified and mapped address have a v4 equivalent.
 274		 */
 275		v4addr = LOOPBACK4_IPV6;
 276		if (!(addr_type & IPV6_ADDR_MULTICAST) &&
 277		    !ipv6_can_nonlocal_bind(sock_net(sk), inet)) {
 278			err = -EADDRNOTAVAIL;
 279			if (!ipv6_chk_addr(sock_net(sk), &addr->sin6_addr,
 280					   dev, 0)) {
 281				goto out_unlock;
 282			}
 283		}
 284	}
 285
 286	inet->inet_rcv_saddr = inet->inet_saddr = v4addr;
 287	sk->sk_v6_rcv_saddr = addr->sin6_addr;
 288	if (!(addr_type & IPV6_ADDR_MULTICAST))
 289		np->saddr = addr->sin6_addr;
 290	err = 0;
 291out_unlock:
 292	rcu_read_unlock();
 293out:
 294	release_sock(sk);
 295	return err;
 296}
 297
 298static void rawv6_err(struct sock *sk, struct sk_buff *skb,
 299		      u8 type, u8 code, int offset, __be32 info)
 300{
 301	bool recverr = inet6_test_bit(RECVERR6, sk);
 302	struct ipv6_pinfo *np = inet6_sk(sk);
 303	int err;
 304	int harderr;
 305
 306	/* Report error on raw socket, if:
 307	   1. User requested recverr.
 308	   2. Socket is connected (otherwise the error indication
 309	      is useless without recverr and error is hard.
 310	 */
 311	if (!recverr && sk->sk_state != TCP_ESTABLISHED)
 312		return;
 313
 314	harderr = icmpv6_err_convert(type, code, &err);
 315	if (type == ICMPV6_PKT_TOOBIG) {
 316		ip6_sk_update_pmtu(skb, sk, info);
 317		harderr = (READ_ONCE(np->pmtudisc) == IPV6_PMTUDISC_DO);
 318	}
 319	if (type == NDISC_REDIRECT) {
 320		ip6_sk_redirect(skb, sk);
 321		return;
 322	}
 323	if (recverr) {
 324		u8 *payload = skb->data;
 325		if (!inet_test_bit(HDRINCL, sk))
 326			payload += offset;
 327		ipv6_icmp_error(sk, skb, err, 0, ntohl(info), payload);
 328	}
 329
 330	if (recverr || harderr) {
 331		sk->sk_err = err;
 332		sk_error_report(sk);
 333	}
 334}
 335
 336void raw6_icmp_error(struct sk_buff *skb, int nexthdr,
 337		u8 type, u8 code, int inner_offset, __be32 info)
 338{
 339	struct net *net = dev_net(skb->dev);
 340	struct hlist_head *hlist;
 341	struct sock *sk;
 342	int hash;
 343
 344	hash = raw_hashfunc(net, nexthdr);
 345	hlist = &raw_v6_hashinfo.ht[hash];
 346	rcu_read_lock();
 347	sk_for_each_rcu(sk, hlist) {
 348		/* Note: ipv6_hdr(skb) != skb->data */
 349		const struct ipv6hdr *ip6h = (const struct ipv6hdr *)skb->data;
 350
 351		if (!raw_v6_match(net, sk, nexthdr, &ip6h->saddr, &ip6h->daddr,
 352				  inet6_iif(skb), inet6_iif(skb)))
 353			continue;
 354		rawv6_err(sk, skb, type, code, inner_offset, info);
 355	}
 356	rcu_read_unlock();
 357}
 358
 359static inline int rawv6_rcv_skb(struct sock *sk, struct sk_buff *skb)
 360{
 361	enum skb_drop_reason reason;
 362
 363	if ((raw6_sk(sk)->checksum || rcu_access_pointer(sk->sk_filter)) &&
 364	    skb_checksum_complete(skb)) {
 365		sk_drops_inc(sk);
 366		sk_skb_reason_drop(sk, skb, SKB_DROP_REASON_SKB_CSUM);
 367		return NET_RX_DROP;
 368	}
 369
 370	/* Charge it to the socket. */
 371	skb_dst_drop(skb);
 372	if (sock_queue_rcv_skb_reason(sk, skb, &reason) < 0) {
 373		sk_skb_reason_drop(sk, skb, reason);
 374		return NET_RX_DROP;
 375	}
 376
 377	return 0;
 378}
 379
 380/*
 381 *	This is next to useless...
 382 *	if we demultiplex in network layer we don't need the extra call
 383 *	just to queue the skb...
 384 *	maybe we could have the network decide upon a hint if it
 385 *	should call raw_rcv for demultiplexing
 386 */
 387int rawv6_rcv(struct sock *sk, struct sk_buff *skb)
 388{
 389	struct inet_sock *inet = inet_sk(sk);
 390	struct raw6_sock *rp = raw6_sk(sk);
 391
 392	if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
 393		sk_drops_inc(sk);
 394		sk_skb_reason_drop(sk, skb, SKB_DROP_REASON_XFRM_POLICY);
 395		return NET_RX_DROP;
 396	}
 397	nf_reset_ct(skb);
 398
 399	if (!rp->checksum)
 400		skb->ip_summed = CHECKSUM_UNNECESSARY;
 401
 402	if (skb->ip_summed == CHECKSUM_COMPLETE) {
 403		skb_postpull_rcsum(skb, skb_network_header(skb),
 404				   skb_network_header_len(skb));
 405		if (!csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 406				     &ipv6_hdr(skb)->daddr,
 407				     skb->len, inet->inet_num, skb->csum))
 408			skb->ip_summed = CHECKSUM_UNNECESSARY;
 409	}
 410	if (!skb_csum_unnecessary(skb))
 411		skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 412							 &ipv6_hdr(skb)->daddr,
 413							 skb->len,
 414							 inet->inet_num, 0));
 415
 416	if (inet_test_bit(HDRINCL, sk)) {
 417		if (skb_checksum_complete(skb)) {
 418			sk_drops_inc(sk);
 419			sk_skb_reason_drop(sk, skb, SKB_DROP_REASON_SKB_CSUM);
 420			return NET_RX_DROP;
 421		}
 422	}
 423
 424	rawv6_rcv_skb(sk, skb);
 425	return 0;
 426}
 427
 428
 429/*
 430 *	This should be easy, if there is something there
 431 *	we return it, otherwise we block.
 432 */
 433
 434static int rawv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 435			 int flags, int *addr_len)
 436{
 437	struct ipv6_pinfo *np = inet6_sk(sk);
 438	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
 439	struct sk_buff *skb;
 440	size_t copied;
 441	int err;
 442
 443	if (flags & MSG_OOB)
 444		return -EOPNOTSUPP;
 445
 446	if (flags & MSG_ERRQUEUE)
 447		return ipv6_recv_error(sk, msg, len, addr_len);
 448
 449	if (np->rxopt.bits.rxpmtu && READ_ONCE(np->rxpmtu))
 450		return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
 451
 452	skb = skb_recv_datagram(sk, flags, &err);
 453	if (!skb)
 454		goto out;
 455
 456	copied = skb->len;
 457	if (copied > len) {
 458		copied = len;
 459		msg->msg_flags |= MSG_TRUNC;
 460	}
 461
 462	if (skb_csum_unnecessary(skb)) {
 463		err = skb_copy_datagram_msg(skb, 0, msg, copied);
 464	} else if (msg->msg_flags&MSG_TRUNC) {
 465		if (__skb_checksum_complete(skb))
 466			goto csum_copy_err;
 467		err = skb_copy_datagram_msg(skb, 0, msg, copied);
 468	} else {
 469		err = skb_copy_and_csum_datagram_msg(skb, 0, msg);
 470		if (err == -EINVAL)
 471			goto csum_copy_err;
 472	}
 473	if (err)
 474		goto out_free;
 475
 476	/* Copy the address. */
 477	if (sin6) {
 478		sin6->sin6_family = AF_INET6;
 479		sin6->sin6_port = 0;
 480		sin6->sin6_addr = ipv6_hdr(skb)->saddr;
 481		sin6->sin6_flowinfo = 0;
 482		sin6->sin6_scope_id = ipv6_iface_scope_id(&sin6->sin6_addr,
 483							  inet6_iif(skb));
 484		*addr_len = sizeof(*sin6);
 485	}
 486
 487	sock_recv_cmsgs(msg, sk, skb);
 488
 489	if (np->rxopt.all)
 490		ip6_datagram_recv_ctl(sk, msg, skb);
 491
 492	err = copied;
 493	if (flags & MSG_TRUNC)
 494		err = skb->len;
 495
 496out_free:
 497	skb_free_datagram(sk, skb);
 498out:
 499	return err;
 500
 501csum_copy_err:
 502	skb_kill_datagram(sk, skb, flags);
 503
 504	/* Error for blocking case is chosen to masquerade
 505	   as some normal condition.
 506	 */
 507	err = (flags&MSG_DONTWAIT) ? -EAGAIN : -EHOSTUNREACH;
 508	goto out;
 509}
 510
 511static int rawv6_push_pending_frames(struct sock *sk, struct flowi6 *fl6,
 512				     struct raw6_sock *rp)
 513{
 514	struct ipv6_txoptions *opt;
 515	struct sk_buff *skb;
 516	int err = 0;
 517	int offset;
 518	int len;
 519	int total_len;
 520	__wsum tmp_csum;
 521	__sum16 csum;
 522
 523	if (!rp->checksum)
 524		goto send;
 525
 526	skb = skb_peek(&sk->sk_write_queue);
 527	if (!skb)
 528		goto out;
 529
 530	offset = rp->offset;
 531	total_len = inet_sk(sk)->cork.base.length;
 532	opt = inet6_sk(sk)->cork.opt;
 533	total_len -= opt ? opt->opt_flen : 0;
 534
 535	if (offset >= total_len - 1) {
 536		err = -EINVAL;
 537		ip6_flush_pending_frames(sk);
 538		goto out;
 539	}
 540
 541	/* should be check HW csum miyazawa */
 542	if (skb_queue_len(&sk->sk_write_queue) == 1) {
 543		/*
 544		 * Only one fragment on the socket.
 545		 */
 546		tmp_csum = skb->csum;
 547	} else {
 548		struct sk_buff *csum_skb = NULL;
 549		tmp_csum = 0;
 550
 551		skb_queue_walk(&sk->sk_write_queue, skb) {
 552			tmp_csum = csum_add(tmp_csum, skb->csum);
 553
 554			if (csum_skb)
 555				continue;
 556
 557			len = skb->len - skb_transport_offset(skb);
 558			if (offset >= len) {
 559				offset -= len;
 560				continue;
 561			}
 562
 563			csum_skb = skb;
 564		}
 565
 566		skb = csum_skb;
 567	}
 568
 569	offset += skb_transport_offset(skb);
 570	err = skb_copy_bits(skb, offset, &csum, 2);
 571	if (err < 0) {
 572		ip6_flush_pending_frames(sk);
 573		goto out;
 574	}
 575
 576	/* in case cksum was not initialized */
 577	if (unlikely(csum))
 578		tmp_csum = csum_sub(tmp_csum, csum_unfold(csum));
 579
 580	csum = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
 581			       total_len, fl6->flowi6_proto, tmp_csum);
 582
 583	if (csum == 0 && fl6->flowi6_proto == IPPROTO_UDP)
 584		csum = CSUM_MANGLED_0;
 585
 586	BUG_ON(skb_store_bits(skb, offset, &csum, 2));
 587
 588send:
 589	err = ip6_push_pending_frames(sk);
 590out:
 591	return err;
 592}
 593
 594static int rawv6_send_hdrinc(struct sock *sk, struct msghdr *msg, int length,
 595			struct flowi6 *fl6, struct dst_entry **dstp,
 596			unsigned int flags, const struct sockcm_cookie *sockc)
 597{
 598	struct net *net = sock_net(sk);
 599	struct ipv6hdr *iph;
 600	struct sk_buff *skb;
 601	int err;
 602	struct rt6_info *rt = dst_rt6_info(*dstp);
 603	int hlen = LL_RESERVED_SPACE(rt->dst.dev);
 604	int tlen = rt->dst.dev->needed_tailroom;
 605
 606	if (length > rt->dst.dev->mtu) {
 607		ipv6_local_error(sk, EMSGSIZE, fl6, rt->dst.dev->mtu);
 608		return -EMSGSIZE;
 609	}
 610	if (length < sizeof(struct ipv6hdr))
 611		return -EINVAL;
 612	if (flags&MSG_PROBE)
 613		goto out;
 614
 615	skb = sock_alloc_send_skb(sk,
 616				  length + hlen + tlen + 15,
 617				  flags & MSG_DONTWAIT, &err);
 618	if (!skb)
 619		goto error;
 620	skb_reserve(skb, hlen);
 621
 622	skb->protocol = htons(ETH_P_IPV6);
 623	skb->priority = sockc->priority;
 624	skb->mark = sockc->mark;
 625	skb_set_delivery_type_by_clockid(skb, sockc->transmit_time, sk->sk_clockid);
 626
 627	skb_put(skb, length);
 628	skb_reset_network_header(skb);
 629	iph = ipv6_hdr(skb);
 630
 631	skb->ip_summed = CHECKSUM_NONE;
 632
 633	skb_setup_tx_timestamp(skb, sockc);
 634
 635	if (flags & MSG_CONFIRM)
 636		skb_set_dst_pending_confirm(skb, 1);
 637
 638	skb->transport_header = skb->network_header;
 639	err = memcpy_from_msg(iph, msg, length);
 640	if (err) {
 641		err = -EFAULT;
 642		kfree_skb(skb);
 643		goto error;
 644	}
 645
 646	skb_dst_set(skb, &rt->dst);
 647	*dstp = NULL;
 648
 649	/* if egress device is enslaved to an L3 master device pass the
 650	 * skb to its handler for processing
 651	 */
 652	skb = l3mdev_ip6_out(sk, skb);
 653	if (unlikely(!skb))
 654		return 0;
 655
 656	/* Acquire rcu_read_lock() in case we need to use rt->rt6i_idev
 657	 * in the error path. Since skb has been freed, the dst could
 658	 * have been queued for deletion.
 659	 */
 660	rcu_read_lock();
 661	IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
 662	err = NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, net, sk, skb,
 663		      NULL, rt->dst.dev, dst_output);
 664	if (err > 0)
 665		err = net_xmit_errno(err);
 666	if (err) {
 667		IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
 668		rcu_read_unlock();
 669		goto error_check;
 670	}
 671	rcu_read_unlock();
 672out:
 673	return 0;
 674
 675error:
 676	IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
 677error_check:
 678	if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk))
 679		err = 0;
 680	return err;
 681}
 682
 683struct raw6_frag_vec {
 684	struct msghdr *msg;
 685	int hlen;
 686	char c[4];
 687};
 688
 689static int rawv6_probe_proto_opt(struct raw6_frag_vec *rfv, struct flowi6 *fl6)
 690{
 691	int err = 0;
 692	switch (fl6->flowi6_proto) {
 693	case IPPROTO_ICMPV6:
 694		rfv->hlen = 2;
 695		err = memcpy_from_msg(rfv->c, rfv->msg, rfv->hlen);
 696		if (!err) {
 697			fl6->fl6_icmp_type = rfv->c[0];
 698			fl6->fl6_icmp_code = rfv->c[1];
 699		}
 700		break;
 701	case IPPROTO_MH:
 702		rfv->hlen = 4;
 703		err = memcpy_from_msg(rfv->c, rfv->msg, rfv->hlen);
 704		if (!err)
 705			fl6->fl6_mh_type = rfv->c[2];
 706	}
 707	return err;
 708}
 709
 710static int raw6_getfrag(void *from, char *to, int offset, int len, int odd,
 711		       struct sk_buff *skb)
 712{
 713	struct raw6_frag_vec *rfv = from;
 714
 715	if (offset < rfv->hlen) {
 716		int copy = min(rfv->hlen - offset, len);
 717
 718		if (skb->ip_summed == CHECKSUM_PARTIAL)
 719			memcpy(to, rfv->c + offset, copy);
 720		else
 721			skb->csum = csum_block_add(
 722				skb->csum,
 723				csum_partial_copy_nocheck(rfv->c + offset,
 724							  to, copy),
 725				odd);
 726
 727		odd = 0;
 728		offset += copy;
 729		to += copy;
 730		len -= copy;
 731
 732		if (!len)
 733			return 0;
 734	}
 735
 736	offset -= rfv->hlen;
 737
 738	return ip_generic_getfrag(rfv->msg, to, offset, len, odd, skb);
 739}
 740
 741static int rawv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 742{
 743	struct ipv6_txoptions *opt_to_free = NULL;
 744	struct ipv6_txoptions opt_space;
 745	DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
 746	struct in6_addr *daddr, *final_p, final;
 747	struct inet_sock *inet = inet_sk(sk);
 748	struct ipv6_pinfo *np = inet6_sk(sk);
 749	struct raw6_sock *rp = raw6_sk(sk);
 750	struct ipv6_txoptions *opt = NULL;
 751	struct ip6_flowlabel *flowlabel = NULL;
 752	struct dst_entry *dst = NULL;
 753	struct raw6_frag_vec rfv;
 754	struct flowi6 fl6;
 755	struct ipcm6_cookie ipc6;
 756	int addr_len = msg->msg_namelen;
 757	int hdrincl;
 758	u16 proto;
 759	int err;
 760
 761	/* Rough check on arithmetic overflow,
 762	   better check is made in ip6_append_data().
 763	 */
 764	if (len > INT_MAX)
 765		return -EMSGSIZE;
 766
 767	/* Mirror BSD error message compatibility */
 768	if (msg->msg_flags & MSG_OOB)
 769		return -EOPNOTSUPP;
 770
 771	hdrincl = inet_test_bit(HDRINCL, sk);
 772
 773	ipcm6_init_sk(&ipc6, sk);
 774
 775	/*
 776	 *	Get and verify the address.
 777	 */
 778	memset(&fl6, 0, sizeof(fl6));
 779
 780	fl6.flowi6_mark = ipc6.sockc.mark;
 781	fl6.flowi6_uid = sk_uid(sk);
 782
 783	if (sin6) {
 784		if (addr_len < SIN6_LEN_RFC2133)
 785			return -EINVAL;
 786
 787		if (sin6->sin6_family && sin6->sin6_family != AF_INET6)
 788			return -EAFNOSUPPORT;
 789
 790		/* port is the proto value [0..255] carried in nexthdr */
 791		proto = ntohs(sin6->sin6_port);
 792
 793		if (!proto)
 794			proto = inet->inet_num;
 795		else if (proto != inet->inet_num &&
 796			 inet->inet_num != IPPROTO_RAW)
 797			return -EINVAL;
 798
 799		if (proto > 255)
 800			return -EINVAL;
 801
 802		daddr = &sin6->sin6_addr;
 803		if (inet6_test_bit(SNDFLOW, sk)) {
 804			fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
 805			if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
 806				flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 807				if (IS_ERR(flowlabel))
 808					return -EINVAL;
 809			}
 810		}
 811
 812		/*
 813		 * Otherwise it will be difficult to maintain
 814		 * sk->sk_dst_cache.
 815		 */
 816		if (sk->sk_state == TCP_ESTABLISHED &&
 817		    ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
 818			daddr = &sk->sk_v6_daddr;
 819
 820		if (addr_len >= sizeof(struct sockaddr_in6) &&
 821		    sin6->sin6_scope_id &&
 822		    __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
 823			fl6.flowi6_oif = sin6->sin6_scope_id;
 824	} else {
 825		if (sk->sk_state != TCP_ESTABLISHED)
 826			return -EDESTADDRREQ;
 827
 828		proto = inet->inet_num;
 829		daddr = &sk->sk_v6_daddr;
 830		fl6.flowlabel = np->flow_label;
 831	}
 832
 833	if (fl6.flowi6_oif == 0)
 834		fl6.flowi6_oif = sk->sk_bound_dev_if;
 835
 836	if (msg->msg_controllen) {
 837		opt = &opt_space;
 838		memset(opt, 0, sizeof(struct ipv6_txoptions));
 839		opt->tot_len = sizeof(struct ipv6_txoptions);
 840		ipc6.opt = opt;
 841
 842		err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6);
 843		if (err < 0) {
 844			fl6_sock_release(flowlabel);
 845			return err;
 846		}
 847		if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
 848			flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
 849			if (IS_ERR(flowlabel))
 850				return -EINVAL;
 851		}
 852		if (!(opt->opt_nflen|opt->opt_flen))
 853			opt = NULL;
 854	}
 855	if (!opt) {
 856		opt = txopt_get(np);
 857		opt_to_free = opt;
 858	}
 859	if (flowlabel)
 860		opt = fl6_merge_options(&opt_space, flowlabel, opt);
 861	opt = ipv6_fixup_options(&opt_space, opt);
 862
 863	fl6.flowi6_proto = proto;
 864	fl6.flowi6_mark = ipc6.sockc.mark;
 865
 866	if (!hdrincl) {
 867		rfv.msg = msg;
 868		rfv.hlen = 0;
 869		err = rawv6_probe_proto_opt(&rfv, &fl6);
 870		if (err)
 871			goto out;
 872	}
 873
 874	if (!ipv6_addr_any(daddr))
 875		fl6.daddr = *daddr;
 876	else
 877		fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
 878	if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
 879		fl6.saddr = np->saddr;
 880
 881	final_p = fl6_update_dst(&fl6, opt, &final);
 882
 883	if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr))
 884		fl6.flowi6_oif = READ_ONCE(np->mcast_oif);
 885	else if (!fl6.flowi6_oif)
 886		fl6.flowi6_oif = READ_ONCE(np->ucast_oif);
 887	security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6));
 888
 889	if (hdrincl)
 890		fl6.flowi6_flags |= FLOWI_FLAG_KNOWN_NH;
 891
 892	fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
 893
 894	dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p);
 895	if (IS_ERR(dst)) {
 896		err = PTR_ERR(dst);
 897		goto out;
 898	}
 899	if (ipc6.hlimit < 0)
 900		ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
 901
 902	if (msg->msg_flags&MSG_CONFIRM)
 903		goto do_confirm;
 904
 905back_from_confirm:
 906	if (hdrincl)
 907		err = rawv6_send_hdrinc(sk, msg, len, &fl6, &dst,
 908					msg->msg_flags, &ipc6.sockc);
 909	else {
 910		ipc6.opt = opt;
 911		lock_sock(sk);
 912		err = ip6_append_data(sk, raw6_getfrag, &rfv,
 913			len, 0, &ipc6, &fl6, dst_rt6_info(dst),
 914			msg->msg_flags);
 915
 916		if (err)
 917			ip6_flush_pending_frames(sk);
 918		else if (!(msg->msg_flags & MSG_MORE))
 919			err = rawv6_push_pending_frames(sk, &fl6, rp);
 920		release_sock(sk);
 921	}
 922done:
 923	dst_release(dst);
 924out:
 925	fl6_sock_release(flowlabel);
 926	txopt_put(opt_to_free);
 927	return err < 0 ? err : len;
 928do_confirm:
 929	if (msg->msg_flags & MSG_PROBE)
 930		dst_confirm_neigh(dst, &fl6.daddr);
 931	if (!(msg->msg_flags & MSG_PROBE) || len)
 932		goto back_from_confirm;
 933	err = 0;
 934	goto done;
 935}
 936
 937static int rawv6_seticmpfilter(struct sock *sk, int optname,
 938			       sockptr_t optval, int optlen)
 939{
 940	switch (optname) {
 941	case ICMPV6_FILTER:
 942		if (optlen > sizeof(struct icmp6_filter))
 943			optlen = sizeof(struct icmp6_filter);
 944		if (copy_from_sockptr(&raw6_sk(sk)->filter, optval, optlen))
 945			return -EFAULT;
 946		return 0;
 947	default:
 948		return -ENOPROTOOPT;
 949	}
 950
 951	return 0;
 952}
 953
 954static int rawv6_geticmpfilter(struct sock *sk, int optname,
 955			       char __user *optval, int __user *optlen)
 956{
 957	int len;
 958
 959	switch (optname) {
 960	case ICMPV6_FILTER:
 961		if (get_user(len, optlen))
 962			return -EFAULT;
 963		if (len < 0)
 964			return -EINVAL;
 965		if (len > sizeof(struct icmp6_filter))
 966			len = sizeof(struct icmp6_filter);
 967		if (put_user(len, optlen))
 968			return -EFAULT;
 969		if (copy_to_user(optval, &raw6_sk(sk)->filter, len))
 970			return -EFAULT;
 971		return 0;
 972	default:
 973		return -ENOPROTOOPT;
 974	}
 975
 976	return 0;
 977}
 978
 979
 980static int do_rawv6_setsockopt(struct sock *sk, int level, int optname,
 981			       sockptr_t optval, unsigned int optlen)
 982{
 983	struct raw6_sock *rp = raw6_sk(sk);
 984	int val;
 985
 986	if (optlen < sizeof(val))
 987		return -EINVAL;
 988
 989	if (copy_from_sockptr(&val, optval, sizeof(val)))
 990		return -EFAULT;
 991
 992	switch (optname) {
 993	case IPV6_HDRINCL:
 994		if (sk->sk_type != SOCK_RAW)
 995			return -EINVAL;
 996		inet_assign_bit(HDRINCL, sk, val);
 997		return 0;
 998	case IPV6_CHECKSUM:
 999		if (inet_sk(sk)->inet_num == IPPROTO_ICMPV6 &&
1000		    level == IPPROTO_IPV6) {
1001			/*
1002			 * RFC3542 tells that IPV6_CHECKSUM socket
1003			 * option in the IPPROTO_IPV6 level is not
1004			 * allowed on ICMPv6 sockets.
1005			 * If you want to set it, use IPPROTO_RAW
1006			 * level IPV6_CHECKSUM socket option
1007			 * (Linux extension).
1008			 */
1009			return -EINVAL;
1010		}
1011
1012		/* You may get strange result with a positive odd offset;
1013		   RFC2292bis agrees with me. */
1014		if (val > 0 && (val&1))
1015			return -EINVAL;
1016		if (val < 0) {
1017			rp->checksum = 0;
1018		} else {
1019			rp->checksum = 1;
1020			rp->offset = val;
1021		}
1022
1023		return 0;
1024
1025	default:
1026		return -ENOPROTOOPT;
1027	}
1028}
1029
1030static int rawv6_setsockopt(struct sock *sk, int level, int optname,
1031			    sockptr_t optval, unsigned int optlen)
1032{
1033	switch (level) {
1034	case SOL_RAW:
1035		break;
1036
1037	case SOL_ICMPV6:
1038		if (inet_sk(sk)->inet_num != IPPROTO_ICMPV6)
1039			return -EOPNOTSUPP;
1040		return rawv6_seticmpfilter(sk, optname, optval, optlen);
1041	case SOL_IPV6:
1042		if (optname == IPV6_CHECKSUM ||
1043		    optname == IPV6_HDRINCL)
1044			break;
1045		fallthrough;
1046	default:
1047		return ipv6_setsockopt(sk, level, optname, optval, optlen);
1048	}
1049
1050	return do_rawv6_setsockopt(sk, level, optname, optval, optlen);
1051}
1052
1053static int do_rawv6_getsockopt(struct sock *sk, int level, int optname,
1054			    char __user *optval, int __user *optlen)
1055{
1056	struct raw6_sock *rp = raw6_sk(sk);
1057	int val, len;
1058
1059	if (get_user(len, optlen))
1060		return -EFAULT;
1061
1062	switch (optname) {
1063	case IPV6_HDRINCL:
1064		val = inet_test_bit(HDRINCL, sk);
1065		break;
1066	case IPV6_CHECKSUM:
1067		/*
1068		 * We allow getsockopt() for IPPROTO_IPV6-level
1069		 * IPV6_CHECKSUM socket option on ICMPv6 sockets
1070		 * since RFC3542 is silent about it.
1071		 */
1072		if (rp->checksum == 0)
1073			val = -1;
1074		else
1075			val = rp->offset;
1076		break;
1077
1078	default:
1079		return -ENOPROTOOPT;
1080	}
1081
1082	len = min_t(unsigned int, sizeof(int), len);
1083
1084	if (put_user(len, optlen))
1085		return -EFAULT;
1086	if (copy_to_user(optval, &val, len))
1087		return -EFAULT;
1088	return 0;
1089}
1090
1091static int rawv6_getsockopt(struct sock *sk, int level, int optname,
1092			  char __user *optval, int __user *optlen)
1093{
1094	switch (level) {
1095	case SOL_RAW:
1096		break;
1097
1098	case SOL_ICMPV6:
1099		if (inet_sk(sk)->inet_num != IPPROTO_ICMPV6)
1100			return -EOPNOTSUPP;
1101		return rawv6_geticmpfilter(sk, optname, optval, optlen);
1102	case SOL_IPV6:
1103		if (optname == IPV6_CHECKSUM ||
1104		    optname == IPV6_HDRINCL)
1105			break;
1106		fallthrough;
1107	default:
1108		return ipv6_getsockopt(sk, level, optname, optval, optlen);
1109	}
1110
1111	return do_rawv6_getsockopt(sk, level, optname, optval, optlen);
1112}
1113
1114static int rawv6_ioctl(struct sock *sk, int cmd, int *karg)
1115{
1116	switch (cmd) {
1117	case SIOCOUTQ: {
1118		*karg = sk_wmem_alloc_get(sk);
1119		return 0;
1120	}
1121	case SIOCINQ: {
1122		struct sk_buff *skb;
1123
1124		spin_lock_bh(&sk->sk_receive_queue.lock);
1125		skb = skb_peek(&sk->sk_receive_queue);
1126		if (skb)
1127			*karg = skb->len;
1128		else
1129			*karg = 0;
1130		spin_unlock_bh(&sk->sk_receive_queue.lock);
1131		return 0;
1132	}
1133
1134	default:
1135#ifdef CONFIG_IPV6_MROUTE
1136		return ip6mr_ioctl(sk, cmd, karg);
1137#else
1138		return -ENOIOCTLCMD;
1139#endif
1140	}
1141}
1142
1143#ifdef CONFIG_COMPAT
1144static int compat_rawv6_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg)
1145{
1146	switch (cmd) {
1147	case SIOCOUTQ:
1148	case SIOCINQ:
1149		return -ENOIOCTLCMD;
1150	default:
1151#ifdef CONFIG_IPV6_MROUTE
1152		return ip6mr_compat_ioctl(sk, cmd, compat_ptr(arg));
1153#else
1154		return -ENOIOCTLCMD;
1155#endif
1156	}
1157}
1158#endif
1159
1160static void rawv6_close(struct sock *sk, long timeout)
1161{
1162	if (inet_sk(sk)->inet_num == IPPROTO_RAW)
1163		ip6_ra_control(sk, -1);
1164	ip6mr_sk_done(sk);
1165	sk_common_release(sk);
1166}
1167
1168static void raw6_destroy(struct sock *sk)
1169{
1170	lock_sock(sk);
1171	ip6_flush_pending_frames(sk);
1172	release_sock(sk);
1173}
1174
1175static int rawv6_init_sk(struct sock *sk)
1176{
1177	struct raw6_sock *rp = raw6_sk(sk);
1178
1179	sk->sk_drop_counters = &rp->drop_counters;
1180	switch (inet_sk(sk)->inet_num) {
1181	case IPPROTO_ICMPV6:
1182		rp->checksum = 1;
1183		rp->offset   = 2;
1184		break;
1185	case IPPROTO_MH:
1186		rp->checksum = 1;
1187		rp->offset   = 4;
1188		break;
1189	default:
1190		break;
1191	}
1192	return 0;
1193}
1194
1195struct proto rawv6_prot = {
1196	.name		   = "RAWv6",
1197	.owner		   = THIS_MODULE,
1198	.close		   = rawv6_close,
1199	.destroy	   = raw6_destroy,
1200	.connect	   = ip6_datagram_connect_v6_only,
1201	.disconnect	   = __udp_disconnect,
1202	.ioctl		   = rawv6_ioctl,
1203	.init		   = rawv6_init_sk,
1204	.setsockopt	   = rawv6_setsockopt,
1205	.getsockopt	   = rawv6_getsockopt,
1206	.sendmsg	   = rawv6_sendmsg,
1207	.recvmsg	   = rawv6_recvmsg,
1208	.bind		   = rawv6_bind,
1209	.backlog_rcv	   = rawv6_rcv_skb,
1210	.hash		   = raw_hash_sk,
1211	.unhash		   = raw_unhash_sk,
1212	.obj_size	   = sizeof(struct raw6_sock),
1213	.ipv6_pinfo_offset = offsetof(struct raw6_sock, inet6),
1214	.useroffset	   = offsetof(struct raw6_sock, filter),
1215	.usersize	   = sizeof_field(struct raw6_sock, filter),
1216	.h.raw_hash	   = &raw_v6_hashinfo,
1217#ifdef CONFIG_COMPAT
1218	.compat_ioctl	   = compat_rawv6_ioctl,
1219#endif
1220	.diag_destroy	   = raw_abort,
1221};
1222
1223#ifdef CONFIG_PROC_FS
1224static int raw6_seq_show(struct seq_file *seq, void *v)
1225{
1226	if (v == SEQ_START_TOKEN) {
1227		seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1228	} else {
1229		struct sock *sp = v;
1230		__u16 srcp  = inet_sk(sp)->inet_num;
1231		ip6_dgram_sock_seq_show(seq, v, srcp, 0,
1232					raw_seq_private(seq)->bucket);
1233	}
1234	return 0;
1235}
1236
1237static const struct seq_operations raw6_seq_ops = {
1238	.start =	raw_seq_start,
1239	.next =		raw_seq_next,
1240	.stop =		raw_seq_stop,
1241	.show =		raw6_seq_show,
1242};
1243
1244static int __net_init raw6_init_net(struct net *net)
1245{
1246	if (!proc_create_net_data("raw6", 0444, net->proc_net, &raw6_seq_ops,
1247			sizeof(struct raw_iter_state), &raw_v6_hashinfo))
1248		return -ENOMEM;
1249
1250	return 0;
1251}
1252
1253static void __net_exit raw6_exit_net(struct net *net)
1254{
1255	remove_proc_entry("raw6", net->proc_net);
1256}
1257
1258static struct pernet_operations raw6_net_ops = {
1259	.init = raw6_init_net,
1260	.exit = raw6_exit_net,
1261};
1262
1263int __init raw6_proc_init(void)
1264{
1265	return register_pernet_subsys(&raw6_net_ops);
1266}
1267
1268void raw6_proc_exit(void)
1269{
1270	unregister_pernet_subsys(&raw6_net_ops);
1271}
1272#endif	/* CONFIG_PROC_FS */
1273
1274/* Same as inet6_dgram_ops, sans udp_poll.  */
1275const struct proto_ops inet6_sockraw_ops = {
1276	.family		   = PF_INET6,
1277	.owner		   = THIS_MODULE,
1278	.release	   = inet6_release,
1279	.bind		   = inet6_bind,
1280	.connect	   = inet_dgram_connect,	/* ok		*/
1281	.socketpair	   = sock_no_socketpair,	/* a do nothing	*/
1282	.accept		   = sock_no_accept,		/* a do nothing	*/
1283	.getname	   = inet6_getname,
1284	.poll		   = datagram_poll,		/* ok		*/
1285	.ioctl		   = inet6_ioctl,		/* must change  */
1286	.gettstamp	   = sock_gettstamp,
1287	.listen		   = sock_no_listen,		/* ok		*/
1288	.shutdown	   = inet_shutdown,		/* ok		*/
1289	.setsockopt	   = sock_common_setsockopt,	/* ok		*/
1290	.getsockopt	   = sock_common_getsockopt,	/* ok		*/
1291	.sendmsg	   = inet_sendmsg,		/* ok		*/
1292	.recvmsg	   = sock_common_recvmsg,	/* ok		*/
1293	.mmap		   = sock_no_mmap,
1294#ifdef CONFIG_COMPAT
1295	.compat_ioctl	   = inet6_compat_ioctl,
1296#endif
1297};
1298
1299static struct inet_protosw rawv6_protosw = {
1300	.type		= SOCK_RAW,
1301	.protocol	= IPPROTO_IP,	/* wild card */
1302	.prot		= &rawv6_prot,
1303	.ops		= &inet6_sockraw_ops,
1304	.flags		= INET_PROTOSW_REUSE,
1305};
1306
1307int __init rawv6_init(void)
1308{
1309	return inet6_register_protosw(&rawv6_protosw);
1310}
1311
1312void rawv6_exit(void)
1313{
1314	inet6_unregister_protosw(&rawv6_protosw);
1315}