/net/ipv4/raw.c

   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		RAW - implementation of IP "raw" sockets.
   8 *
   9 * Authors:	Ross Biro
  10 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11 *
  12 * Fixes:
  13 *		Alan Cox	:	verify_area() fixed up
  14 *		Alan Cox	:	ICMP error handling
  15 *		Alan Cox	:	EMSGSIZE if you send too big a packet
  16 *		Alan Cox	: 	Now uses generic datagrams and shared
  17 *					skbuff library. No more peek crashes,
  18 *					no more backlogs
  19 *		Alan Cox	:	Checks sk->broadcast.
  20 *		Alan Cox	:	Uses skb_free_datagram/skb_copy_datagram
  21 *		Alan Cox	:	Raw passes ip options too
  22 *		Alan Cox	:	Setsocketopt added
  23 *		Alan Cox	:	Fixed error return for broadcasts
  24 *		Alan Cox	:	Removed wake_up calls
  25 *		Alan Cox	:	Use ttl/tos
  26 *		Alan Cox	:	Cleaned up old debugging
  27 *		Alan Cox	:	Use new kernel side addresses
  28 *	Arnt Gulbrandsen	:	Fixed MSG_DONTROUTE in raw sockets.
  29 *		Alan Cox	:	BSD style RAW socket demultiplexing.
  30 *		Alan Cox	:	Beginnings of mrouted support.
  31 *		Alan Cox	:	Added IP_HDRINCL option.
  32 *		Alan Cox	:	Skip broadcast check if BSDism set.
  33 *		David S. Miller	:	New socket lookup architecture.
  34 */
  35
  36#include <linux/types.h>
  37#include <linux/atomic.h>
  38#include <asm/byteorder.h>
  39#include <asm/current.h>
  40#include <linux/uaccess.h>
  41#include <asm/ioctls.h>
  42#include <linux/stddef.h>
  43#include <linux/slab.h>
  44#include <linux/errno.h>
  45#include <linux/kernel.h>
  46#include <linux/export.h>
  47#include <linux/spinlock.h>
  48#include <linux/sockios.h>
  49#include <linux/socket.h>
  50#include <linux/in.h>
  51#include <linux/mroute.h>
  52#include <linux/netdevice.h>
  53#include <linux/in_route.h>
  54#include <linux/route.h>
  55#include <linux/skbuff.h>
  56#include <linux/igmp.h>
  57#include <net/net_namespace.h>
  58#include <net/dst.h>
  59#include <net/sock.h>
  60#include <linux/ip.h>
  61#include <linux/net.h>
  62#include <net/ip.h>
  63#include <net/icmp.h>
  64#include <net/udp.h>
  65#include <net/raw.h>
  66#include <net/snmp.h>
  67#include <net/tcp_states.h>
  68#include <net/inet_common.h>
  69#include <net/checksum.h>
  70#include <net/xfrm.h>
  71#include <linux/rtnetlink.h>
  72#include <linux/proc_fs.h>
  73#include <linux/seq_file.h>
  74#include <linux/netfilter.h>
  75#include <linux/netfilter_ipv4.h>
  76#include <linux/compat.h>
  77#include <linux/uio.h>
  78
  79struct raw_frag_vec {
  80	struct msghdr *msg;
  81	union {
  82		struct icmphdr icmph;
  83		char c[1];
  84	} hdr;
  85	int hlen;
  86};
  87
  88struct raw_hashinfo raw_v4_hashinfo = {
  89	.lock = __RW_LOCK_UNLOCKED(raw_v4_hashinfo.lock),
  90};
  91EXPORT_SYMBOL_GPL(raw_v4_hashinfo);
  92
  93int raw_hash_sk(struct sock *sk)
  94{
  95	struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
  96	struct hlist_head *head;
  97
  98	head = &h->ht[inet_sk(sk)->inet_num & (RAW_HTABLE_SIZE - 1)];
  99
 100	write_lock_bh(&h->lock);
 101	sk_add_node(sk, head);
 102	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 103	write_unlock_bh(&h->lock);
 104
 105	return 0;
 106}
 107EXPORT_SYMBOL_GPL(raw_hash_sk);
 108
 109void raw_unhash_sk(struct sock *sk)
 110{
 111	struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
 112
 113	write_lock_bh(&h->lock);
 114	if (sk_del_node_init(sk))
 115		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 116	write_unlock_bh(&h->lock);
 117}
 118EXPORT_SYMBOL_GPL(raw_unhash_sk);
 119
 120struct sock *__raw_v4_lookup(struct net *net, struct sock *sk,
 121			     unsigned short num, __be32 raddr, __be32 laddr,
 122			     int dif, int sdif)
 123{
 124	sk_for_each_from(sk) {
 125		struct inet_sock *inet = inet_sk(sk);
 126
 127		if (net_eq(sock_net(sk), net) && inet->inet_num == num	&&
 128		    !(inet->inet_daddr && inet->inet_daddr != raddr) 	&&
 129		    !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) &&
 130		    raw_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
 131			goto found; /* gotcha */
 132	}
 133	sk = NULL;
 134found:
 135	return sk;
 136}
 137EXPORT_SYMBOL_GPL(__raw_v4_lookup);
 138
 139/*
 140 *	0 - deliver
 141 *	1 - block
 142 */
 143static int icmp_filter(const struct sock *sk, const struct sk_buff *skb)
 144{
 145	struct icmphdr _hdr;
 146	const struct icmphdr *hdr;
 147
 148	hdr = skb_header_pointer(skb, skb_transport_offset(skb),
 149				 sizeof(_hdr), &_hdr);
 150	if (!hdr)
 151		return 1;
 152
 153	if (hdr->type < 32) {
 154		__u32 data = raw_sk(sk)->filter.data;
 155
 156		return ((1U << hdr->type) & data) != 0;
 157	}
 158
 159	/* Do not block unknown ICMP types */
 160	return 0;
 161}
 162
 163/* IP input processing comes here for RAW socket delivery.
 164 * Caller owns SKB, so we must make clones.
 165 *
 166 * RFC 1122: SHOULD pass TOS value up to the transport layer.
 167 * -> It does. And not only TOS, but all IP header.
 168 */
 169static int raw_v4_input(struct sk_buff *skb, const struct iphdr *iph, int hash)
 170{
 171	int sdif = inet_sdif(skb);
 172	int dif = inet_iif(skb);
 173	struct sock *sk;
 174	struct hlist_head *head;
 175	int delivered = 0;
 176	struct net *net;
 177
 178	read_lock(&raw_v4_hashinfo.lock);
 179	head = &raw_v4_hashinfo.ht[hash];
 180	if (hlist_empty(head))
 181		goto out;
 182
 183	net = dev_net(skb->dev);
 184	sk = __raw_v4_lookup(net, __sk_head(head), iph->protocol,
 185			     iph->saddr, iph->daddr, dif, sdif);
 186
 187	while (sk) {
 188		delivered = 1;
 189		if ((iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) &&
 190		    ip_mc_sf_allow(sk, iph->daddr, iph->saddr,
 191				   skb->dev->ifindex, sdif)) {
 192			struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
 193
 194			/* Not releasing hash table! */
 195			if (clone)
 196				raw_rcv(sk, clone);
 197		}
 198		sk = __raw_v4_lookup(net, sk_next(sk), iph->protocol,
 199				     iph->saddr, iph->daddr,
 200				     dif, sdif);
 201	}
 202out:
 203	read_unlock(&raw_v4_hashinfo.lock);
 204	return delivered;
 205}
 206
 207int raw_local_deliver(struct sk_buff *skb, int protocol)
 208{
 209	int hash;
 210	struct sock *raw_sk;
 211
 212	hash = protocol & (RAW_HTABLE_SIZE - 1);
 213	raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
 214
 215	/* If there maybe a raw socket we must check - if not we
 216	 * don't care less
 217	 */
 218	if (raw_sk && !raw_v4_input(skb, ip_hdr(skb), hash))
 219		raw_sk = NULL;
 220
 221	return raw_sk != NULL;
 222
 223}
 224
 225static void raw_err(struct sock *sk, struct sk_buff *skb, u32 info)
 226{
 227	struct inet_sock *inet = inet_sk(sk);
 228	const int type = icmp_hdr(skb)->type;
 229	const int code = icmp_hdr(skb)->code;
 230	int err = 0;
 231	int harderr = 0;
 232
 233	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 234		ipv4_sk_update_pmtu(skb, sk, info);
 235	else if (type == ICMP_REDIRECT) {
 236		ipv4_sk_redirect(skb, sk);
 237		return;
 238	}
 239
 240	/* Report error on raw socket, if:
 241	   1. User requested ip_recverr.
 242	   2. Socket is connected (otherwise the error indication
 243	      is useless without ip_recverr and error is hard.
 244	 */
 245	if (!inet->recverr && sk->sk_state != TCP_ESTABLISHED)
 246		return;
 247
 248	switch (type) {
 249	default:
 250	case ICMP_TIME_EXCEEDED:
 251		err = EHOSTUNREACH;
 252		break;
 253	case ICMP_SOURCE_QUENCH:
 254		return;
 255	case ICMP_PARAMETERPROB:
 256		err = EPROTO;
 257		harderr = 1;
 258		break;
 259	case ICMP_DEST_UNREACH:
 260		err = EHOSTUNREACH;
 261		if (code > NR_ICMP_UNREACH)
 262			break;
 263		err = icmp_err_convert[code].errno;
 264		harderr = icmp_err_convert[code].fatal;
 265		if (code == ICMP_FRAG_NEEDED) {
 266			harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
 267			err = EMSGSIZE;
 268		}
 269	}
 270
 271	if (inet->recverr) {
 272		const struct iphdr *iph = (const struct iphdr *)skb->data;
 273		u8 *payload = skb->data + (iph->ihl << 2);
 274
 275		if (inet->hdrincl)
 276			payload = skb->data;
 277		ip_icmp_error(sk, skb, err, 0, info, payload);
 278	}
 279
 280	if (inet->recverr || harderr) {
 281		sk->sk_err = err;
 282		sk->sk_error_report(sk);
 283	}
 284}
 285
 286void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
 287{
 288	int hash;
 289	struct sock *raw_sk;
 290	const struct iphdr *iph;
 291	struct net *net;
 292
 293	hash = protocol & (RAW_HTABLE_SIZE - 1);
 294
 295	read_lock(&raw_v4_hashinfo.lock);
 296	raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
 297	if (raw_sk) {
 298		int dif = skb->dev->ifindex;
 299		int sdif = inet_sdif(skb);
 300
 301		iph = (const struct iphdr *)skb->data;
 302		net = dev_net(skb->dev);
 303
 304		while ((raw_sk = __raw_v4_lookup(net, raw_sk, protocol,
 305						iph->daddr, iph->saddr,
 306						dif, sdif)) != NULL) {
 307			raw_err(raw_sk, skb, info);
 308			raw_sk = sk_next(raw_sk);
 309			iph = (const struct iphdr *)skb->data;
 310		}
 311	}
 312	read_unlock(&raw_v4_hashinfo.lock);
 313}
 314
 315static int raw_rcv_skb(struct sock *sk, struct sk_buff *skb)
 316{
 317	/* Charge it to the socket. */
 318
 319	ipv4_pktinfo_prepare(sk, skb);
 320	if (sock_queue_rcv_skb(sk, skb) < 0) {
 321		kfree_skb(skb);
 322		return NET_RX_DROP;
 323	}
 324
 325	return NET_RX_SUCCESS;
 326}
 327
 328int raw_rcv(struct sock *sk, struct sk_buff *skb)
 329{
 330	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
 331		atomic_inc(&sk->sk_drops);
 332		kfree_skb(skb);
 333		return NET_RX_DROP;
 334	}
 335	nf_reset_ct(skb);
 336
 337	skb_push(skb, skb->data - skb_network_header(skb));
 338
 339	raw_rcv_skb(sk, skb);
 340	return 0;
 341}
 342
 343static int raw_send_hdrinc(struct sock *sk, struct flowi4 *fl4,
 344			   struct msghdr *msg, size_t length,
 345			   struct rtable **rtp, unsigned int flags,
 346			   const struct sockcm_cookie *sockc)
 347{
 348	struct inet_sock *inet = inet_sk(sk);
 349	struct net *net = sock_net(sk);
 350	struct iphdr *iph;
 351	struct sk_buff *skb;
 352	unsigned int iphlen;
 353	int err;
 354	struct rtable *rt = *rtp;
 355	int hlen, tlen;
 356
 357	if (length > rt->dst.dev->mtu) {
 358		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
 359			       rt->dst.dev->mtu);
 360		return -EMSGSIZE;
 361	}
 362	if (length < sizeof(struct iphdr))
 363		return -EINVAL;
 364
 365	if (flags&MSG_PROBE)
 366		goto out;
 367
 368	hlen = LL_RESERVED_SPACE(rt->dst.dev);
 369	tlen = rt->dst.dev->needed_tailroom;
 370	skb = sock_alloc_send_skb(sk,
 371				  length + hlen + tlen + 15,
 372				  flags & MSG_DONTWAIT, &err);
 373	if (!skb)
 374		goto error;
 375	skb_reserve(skb, hlen);
 376
 377	skb->priority = sk->sk_priority;
 378	skb->mark = sockc->mark;
 379	skb->tstamp = sockc->transmit_time;
 380	skb_dst_set(skb, &rt->dst);
 381	*rtp = NULL;
 382
 383	skb_reset_network_header(skb);
 384	iph = ip_hdr(skb);
 385	skb_put(skb, length);
 386
 387	skb->ip_summed = CHECKSUM_NONE;
 388
 389	skb_setup_tx_timestamp(skb, sockc->tsflags);
 390
 391	if (flags & MSG_CONFIRM)
 392		skb_set_dst_pending_confirm(skb, 1);
 393
 394	skb->transport_header = skb->network_header;
 395	err = -EFAULT;
 396	if (memcpy_from_msg(iph, msg, length))
 397		goto error_free;
 398
 399	iphlen = iph->ihl * 4;
 400
 401	/*
 402	 * We don't want to modify the ip header, but we do need to
 403	 * be sure that it won't cause problems later along the network
 404	 * stack.  Specifically we want to make sure that iph->ihl is a
 405	 * sane value.  If ihl points beyond the length of the buffer passed
 406	 * in, reject the frame as invalid
 407	 */
 408	err = -EINVAL;
 409	if (iphlen > length)
 410		goto error_free;
 411
 412	if (iphlen >= sizeof(*iph)) {
 413		if (!iph->saddr)
 414			iph->saddr = fl4->saddr;
 415		iph->check   = 0;
 416		iph->tot_len = htons(length);
 417		if (!iph->id)
 418			ip_select_ident(net, skb, NULL);
 419
 420		iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
 421		skb->transport_header += iphlen;
 422		if (iph->protocol == IPPROTO_ICMP &&
 423		    length >= iphlen + sizeof(struct icmphdr))
 424			icmp_out_count(net, ((struct icmphdr *)
 425				skb_transport_header(skb))->type);
 426	}
 427
 428	err = NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
 429		      net, sk, skb, NULL, rt->dst.dev,
 430		      dst_output);
 431	if (err > 0)
 432		err = net_xmit_errno(err);
 433	if (err)
 434		goto error;
 435out:
 436	return 0;
 437
 438error_free:
 439	kfree_skb(skb);
 440error:
 441	IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
 442	if (err == -ENOBUFS && !inet->recverr)
 443		err = 0;
 444	return err;
 445}
 446
 447static int raw_probe_proto_opt(struct raw_frag_vec *rfv, struct flowi4 *fl4)
 448{
 449	int err;
 450
 451	if (fl4->flowi4_proto != IPPROTO_ICMP)
 452		return 0;
 453
 454	/* We only need the first two bytes. */
 455	rfv->hlen = 2;
 456
 457	err = memcpy_from_msg(rfv->hdr.c, rfv->msg, rfv->hlen);
 458	if (err)
 459		return err;
 460
 461	fl4->fl4_icmp_type = rfv->hdr.icmph.type;
 462	fl4->fl4_icmp_code = rfv->hdr.icmph.code;
 463
 464	return 0;
 465}
 466
 467static int raw_getfrag(void *from, char *to, int offset, int len, int odd,
 468		       struct sk_buff *skb)
 469{
 470	struct raw_frag_vec *rfv = from;
 471
 472	if (offset < rfv->hlen) {
 473		int copy = min(rfv->hlen - offset, len);
 474
 475		if (skb->ip_summed == CHECKSUM_PARTIAL)
 476			memcpy(to, rfv->hdr.c + offset, copy);
 477		else
 478			skb->csum = csum_block_add(
 479				skb->csum,
 480				csum_partial_copy_nocheck(rfv->hdr.c + offset,
 481							  to, copy, 0),
 482				odd);
 483
 484		odd = 0;
 485		offset += copy;
 486		to += copy;
 487		len -= copy;
 488
 489		if (!len)
 490			return 0;
 491	}
 492
 493	offset -= rfv->hlen;
 494
 495	return ip_generic_getfrag(rfv->msg, to, offset, len, odd, skb);
 496}
 497
 498static int raw_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 499{
 500	struct inet_sock *inet = inet_sk(sk);
 501	struct net *net = sock_net(sk);
 502	struct ipcm_cookie ipc;
 503	struct rtable *rt = NULL;
 504	struct flowi4 fl4;
 505	int free = 0;
 506	__be32 daddr;
 507	__be32 saddr;
 508	u8  tos;
 509	int err;
 510	struct ip_options_data opt_copy;
 511	struct raw_frag_vec rfv;
 512	int hdrincl;
 513
 514	err = -EMSGSIZE;
 515	if (len > 0xFFFF)
 516		goto out;
 517
 518	/* hdrincl should be READ_ONCE(inet->hdrincl)
 519	 * but READ_ONCE() doesn't work with bit fields.
 520	 * Doing this indirectly yields the same result.
 521	 */
 522	hdrincl = inet->hdrincl;
 523	hdrincl = READ_ONCE(hdrincl);
 524	/*
 525	 *	Check the flags.
 526	 */
 527
 528	err = -EOPNOTSUPP;
 529	if (msg->msg_flags & MSG_OOB)	/* Mirror BSD error message */
 530		goto out;               /* compatibility */
 531
 532	/*
 533	 *	Get and verify the address.
 534	 */
 535
 536	if (msg->msg_namelen) {
 537		DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
 538		err = -EINVAL;
 539		if (msg->msg_namelen < sizeof(*usin))
 540			goto out;
 541		if (usin->sin_family != AF_INET) {
 542			pr_info_once("%s: %s forgot to set AF_INET. Fix it!\n",
 543				     __func__, current->comm);
 544			err = -EAFNOSUPPORT;
 545			if (usin->sin_family)
 546				goto out;
 547		}
 548		daddr = usin->sin_addr.s_addr;
 549		/* ANK: I did not forget to get protocol from port field.
 550		 * I just do not know, who uses this weirdness.
 551		 * IP_HDRINCL is much more convenient.
 552		 */
 553	} else {
 554		err = -EDESTADDRREQ;
 555		if (sk->sk_state != TCP_ESTABLISHED)
 556			goto out;
 557		daddr = inet->inet_daddr;
 558	}
 559
 560	ipcm_init_sk(&ipc, inet);
 561
 562	if (msg->msg_controllen) {
 563		err = ip_cmsg_send(sk, msg, &ipc, false);
 564		if (unlikely(err)) {
 565			kfree(ipc.opt);
 566			goto out;
 567		}
 568		if (ipc.opt)
 569			free = 1;
 570	}
 571
 572	saddr = ipc.addr;
 573	ipc.addr = daddr;
 574
 575	if (!ipc.opt) {
 576		struct ip_options_rcu *inet_opt;
 577
 578		rcu_read_lock();
 579		inet_opt = rcu_dereference(inet->inet_opt);
 580		if (inet_opt) {
 581			memcpy(&opt_copy, inet_opt,
 582			       sizeof(*inet_opt) + inet_opt->opt.optlen);
 583			ipc.opt = &opt_copy.opt;
 584		}
 585		rcu_read_unlock();
 586	}
 587
 588	if (ipc.opt) {
 589		err = -EINVAL;
 590		/* Linux does not mangle headers on raw sockets,
 591		 * so that IP options + IP_HDRINCL is non-sense.
 592		 */
 593		if (hdrincl)
 594			goto done;
 595		if (ipc.opt->opt.srr) {
 596			if (!daddr)
 597				goto done;
 598			daddr = ipc.opt->opt.faddr;
 599		}
 600	}
 601	tos = get_rtconn_flags(&ipc, sk);
 602	if (msg->msg_flags & MSG_DONTROUTE)
 603		tos |= RTO_ONLINK;
 604
 605	if (ipv4_is_multicast(daddr)) {
 606		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
 607			ipc.oif = inet->mc_index;
 608		if (!saddr)
 609			saddr = inet->mc_addr;
 610	} else if (!ipc.oif) {
 611		ipc.oif = inet->uc_index;
 612	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
 613		/* oif is set, packet is to local broadcast and
 614		 * and uc_index is set. oif is most likely set
 615		 * by sk_bound_dev_if. If uc_index != oif check if the
 616		 * oif is an L3 master and uc_index is an L3 slave.
 617		 * If so, we want to allow the send using the uc_index.
 618		 */
 619		if (ipc.oif != inet->uc_index &&
 620		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
 621							      inet->uc_index)) {
 622			ipc.oif = inet->uc_index;
 623		}
 624	}
 625
 626	flowi4_init_output(&fl4, ipc.oif, ipc.sockc.mark, tos,
 627			   RT_SCOPE_UNIVERSE,
 628			   hdrincl ? IPPROTO_RAW : sk->sk_protocol,
 629			   inet_sk_flowi_flags(sk) |
 630			    (hdrincl ? FLOWI_FLAG_KNOWN_NH : 0),
 631			   daddr, saddr, 0, 0, sk->sk_uid);
 632
 633	if (!hdrincl) {
 634		rfv.msg = msg;
 635		rfv.hlen = 0;
 636
 637		err = raw_probe_proto_opt(&rfv, &fl4);
 638		if (err)
 639			goto done;
 640	}
 641
 642	security_sk_classify_flow(sk, flowi4_to_flowi(&fl4));
 643	rt = ip_route_output_flow(net, &fl4, sk);
 644	if (IS_ERR(rt)) {
 645		err = PTR_ERR(rt);
 646		rt = NULL;
 647		goto done;
 648	}
 649
 650	err = -EACCES;
 651	if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
 652		goto done;
 653
 654	if (msg->msg_flags & MSG_CONFIRM)
 655		goto do_confirm;
 656back_from_confirm:
 657
 658	if (hdrincl)
 659		err = raw_send_hdrinc(sk, &fl4, msg, len,
 660				      &rt, msg->msg_flags, &ipc.sockc);
 661
 662	 else {
 663		if (!ipc.addr)
 664			ipc.addr = fl4.daddr;
 665		lock_sock(sk);
 666		err = ip_append_data(sk, &fl4, raw_getfrag,
 667				     &rfv, len, 0,
 668				     &ipc, &rt, msg->msg_flags);
 669		if (err)
 670			ip_flush_pending_frames(sk);
 671		else if (!(msg->msg_flags & MSG_MORE)) {
 672			err = ip_push_pending_frames(sk, &fl4);
 673			if (err == -ENOBUFS && !inet->recverr)
 674				err = 0;
 675		}
 676		release_sock(sk);
 677	}
 678done:
 679	if (free)
 680		kfree(ipc.opt);
 681	ip_rt_put(rt);
 682
 683out:
 684	if (err < 0)
 685		return err;
 686	return len;
 687
 688do_confirm:
 689	if (msg->msg_flags & MSG_PROBE)
 690		dst_confirm_neigh(&rt->dst, &fl4.daddr);
 691	if (!(msg->msg_flags & MSG_PROBE) || len)
 692		goto back_from_confirm;
 693	err = 0;
 694	goto done;
 695}
 696
 697static void raw_close(struct sock *sk, long timeout)
 698{
 699	/*
 700	 * Raw sockets may have direct kernel references. Kill them.
 701	 */
 702	ip_ra_control(sk, 0, NULL);
 703
 704	sk_common_release(sk);
 705}
 706
 707static void raw_destroy(struct sock *sk)
 708{
 709	lock_sock(sk);
 710	ip_flush_pending_frames(sk);
 711	release_sock(sk);
 712}
 713
 714/* This gets rid of all the nasties in af_inet. -DaveM */
 715static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 716{
 717	struct inet_sock *inet = inet_sk(sk);
 718	struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
 719	u32 tb_id = RT_TABLE_LOCAL;
 720	int ret = -EINVAL;
 721	int chk_addr_ret;
 722
 723	if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
 724		goto out;
 725
 726	if (sk->sk_bound_dev_if)
 727		tb_id = l3mdev_fib_table_by_index(sock_net(sk),
 728						 sk->sk_bound_dev_if) ? : tb_id;
 729
 730	chk_addr_ret = inet_addr_type_table(sock_net(sk), addr->sin_addr.s_addr,
 731					    tb_id);
 732
 733	ret = -EADDRNOTAVAIL;
 734	if (addr->sin_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
 735	    chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
 736		goto out;
 737	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
 738	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
 739		inet->inet_saddr = 0;  /* Use device */
 740	sk_dst_reset(sk);
 741	ret = 0;
 742out:	return ret;
 743}
 744
 745/*
 746 *	This should be easy, if there is something there
 747 *	we return it, otherwise we block.
 748 */
 749
 750static int raw_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 751		       int noblock, int flags, int *addr_len)
 752{
 753	struct inet_sock *inet = inet_sk(sk);
 754	size_t copied = 0;
 755	int err = -EOPNOTSUPP;
 756	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
 757	struct sk_buff *skb;
 758
 759	if (flags & MSG_OOB)
 760		goto out;
 761
 762	if (flags & MSG_ERRQUEUE) {
 763		err = ip_recv_error(sk, msg, len, addr_len);
 764		goto out;
 765	}
 766
 767	skb = skb_recv_datagram(sk, flags, noblock, &err);
 768	if (!skb)
 769		goto out;
 770
 771	copied = skb->len;
 772	if (len < copied) {
 773		msg->msg_flags |= MSG_TRUNC;
 774		copied = len;
 775	}
 776
 777	err = skb_copy_datagram_msg(skb, 0, msg, copied);
 778	if (err)
 779		goto done;
 780
 781	sock_recv_ts_and_drops(msg, sk, skb);
 782
 783	/* Copy the address. */
 784	if (sin) {
 785		sin->sin_family = AF_INET;
 786		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
 787		sin->sin_port = 0;
 788		memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
 789		*addr_len = sizeof(*sin);
 790	}
 791	if (inet->cmsg_flags)
 792		ip_cmsg_recv(msg, skb);
 793	if (flags & MSG_TRUNC)
 794		copied = skb->len;
 795done:
 796	skb_free_datagram(sk, skb);
 797out:
 798	if (err)
 799		return err;
 800	return copied;
 801}
 802
 803static int raw_sk_init(struct sock *sk)
 804{
 805	struct raw_sock *rp = raw_sk(sk);
 806
 807	if (inet_sk(sk)->inet_num == IPPROTO_ICMP)
 808		memset(&rp->filter, 0, sizeof(rp->filter));
 809	return 0;
 810}
 811
 812static int raw_seticmpfilter(struct sock *sk, char __user *optval, int optlen)
 813{
 814	if (optlen > sizeof(struct icmp_filter))
 815		optlen = sizeof(struct icmp_filter);
 816	if (copy_from_user(&raw_sk(sk)->filter, optval, optlen))
 817		return -EFAULT;
 818	return 0;
 819}
 820
 821static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
 822{
 823	int len, ret = -EFAULT;
 824
 825	if (get_user(len, optlen))
 826		goto out;
 827	ret = -EINVAL;
 828	if (len < 0)
 829		goto out;
 830	if (len > sizeof(struct icmp_filter))
 831		len = sizeof(struct icmp_filter);
 832	ret = -EFAULT;
 833	if (put_user(len, optlen) ||
 834	    copy_to_user(optval, &raw_sk(sk)->filter, len))
 835		goto out;
 836	ret = 0;
 837out:	return ret;
 838}
 839
 840static int do_raw_setsockopt(struct sock *sk, int level, int optname,
 841			  char __user *optval, unsigned int optlen)
 842{
 843	if (optname == ICMP_FILTER) {
 844		if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
 845			return -EOPNOTSUPP;
 846		else
 847			return raw_seticmpfilter(sk, optval, optlen);
 848	}
 849	return -ENOPROTOOPT;
 850}
 851
 852static int raw_setsockopt(struct sock *sk, int level, int optname,
 853			  char __user *optval, unsigned int optlen)
 854{
 855	if (level != SOL_RAW)
 856		return ip_setsockopt(sk, level, optname, optval, optlen);
 857	return do_raw_setsockopt(sk, level, optname, optval, optlen);
 858}
 859
 860#ifdef CONFIG_COMPAT
 861static int compat_raw_setsockopt(struct sock *sk, int level, int optname,
 862				 char __user *optval, unsigned int optlen)
 863{
 864	if (level != SOL_RAW)
 865		return compat_ip_setsockopt(sk, level, optname, optval, optlen);
 866	return do_raw_setsockopt(sk, level, optname, optval, optlen);
 867}
 868#endif
 869
 870static int do_raw_getsockopt(struct sock *sk, int level, int optname,
 871			  char __user *optval, int __user *optlen)
 872{
 873	if (optname == ICMP_FILTER) {
 874		if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
 875			return -EOPNOTSUPP;
 876		else
 877			return raw_geticmpfilter(sk, optval, optlen);
 878	}
 879	return -ENOPROTOOPT;
 880}
 881
 882static int raw_getsockopt(struct sock *sk, int level, int optname,
 883			  char __user *optval, int __user *optlen)
 884{
 885	if (level != SOL_RAW)
 886		return ip_getsockopt(sk, level, optname, optval, optlen);
 887	return do_raw_getsockopt(sk, level, optname, optval, optlen);
 888}
 889
 890#ifdef CONFIG_COMPAT
 891static int compat_raw_getsockopt(struct sock *sk, int level, int optname,
 892				 char __user *optval, int __user *optlen)
 893{
 894	if (level != SOL_RAW)
 895		return compat_ip_getsockopt(sk, level, optname, optval, optlen);
 896	return do_raw_getsockopt(sk, level, optname, optval, optlen);
 897}
 898#endif
 899
 900static int raw_ioctl(struct sock *sk, int cmd, unsigned long arg)
 901{
 902	switch (cmd) {
 903	case SIOCOUTQ: {
 904		int amount = sk_wmem_alloc_get(sk);
 905
 906		return put_user(amount, (int __user *)arg);
 907	}
 908	case SIOCINQ: {
 909		struct sk_buff *skb;
 910		int amount = 0;
 911
 912		spin_lock_bh(&sk->sk_receive_queue.lock);
 913		skb = skb_peek(&sk->sk_receive_queue);
 914		if (skb)
 915			amount = skb->len;
 916		spin_unlock_bh(&sk->sk_receive_queue.lock);
 917		return put_user(amount, (int __user *)arg);
 918	}
 919
 920	default:
 921#ifdef CONFIG_IP_MROUTE
 922		return ipmr_ioctl(sk, cmd, (void __user *)arg);
 923#else
 924		return -ENOIOCTLCMD;
 925#endif
 926	}
 927}
 928
 929#ifdef CONFIG_COMPAT
 930static int compat_raw_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg)
 931{
 932	switch (cmd) {
 933	case SIOCOUTQ:
 934	case SIOCINQ:
 935		return -ENOIOCTLCMD;
 936	default:
 937#ifdef CONFIG_IP_MROUTE
 938		return ipmr_compat_ioctl(sk, cmd, compat_ptr(arg));
 939#else
 940		return -ENOIOCTLCMD;
 941#endif
 942	}
 943}
 944#endif
 945
 946int raw_abort(struct sock *sk, int err)
 947{
 948	lock_sock(sk);
 949
 950	sk->sk_err = err;
 951	sk->sk_error_report(sk);
 952	__udp_disconnect(sk, 0);
 953
 954	release_sock(sk);
 955
 956	return 0;
 957}
 958EXPORT_SYMBOL_GPL(raw_abort);
 959
 960struct proto raw_prot = {
 961	.name		   = "RAW",
 962	.owner		   = THIS_MODULE,
 963	.close		   = raw_close,
 964	.destroy	   = raw_destroy,
 965	.connect	   = ip4_datagram_connect,
 966	.disconnect	   = __udp_disconnect,
 967	.ioctl		   = raw_ioctl,
 968	.init		   = raw_sk_init,
 969	.setsockopt	   = raw_setsockopt,
 970	.getsockopt	   = raw_getsockopt,
 971	.sendmsg	   = raw_sendmsg,
 972	.recvmsg	   = raw_recvmsg,
 973	.bind		   = raw_bind,
 974	.backlog_rcv	   = raw_rcv_skb,
 975	.release_cb	   = ip4_datagram_release_cb,
 976	.hash		   = raw_hash_sk,
 977	.unhash		   = raw_unhash_sk,
 978	.obj_size	   = sizeof(struct raw_sock),
 979	.useroffset	   = offsetof(struct raw_sock, filter),
 980	.usersize	   = sizeof_field(struct raw_sock, filter),
 981	.h.raw_hash	   = &raw_v4_hashinfo,
 982#ifdef CONFIG_COMPAT
 983	.compat_setsockopt = compat_raw_setsockopt,
 984	.compat_getsockopt = compat_raw_getsockopt,
 985	.compat_ioctl	   = compat_raw_ioctl,
 986#endif
 987	.diag_destroy	   = raw_abort,
 988};
 989
 990#ifdef CONFIG_PROC_FS
 991static struct sock *raw_get_first(struct seq_file *seq)
 992{
 993	struct sock *sk;
 994	struct raw_hashinfo *h = PDE_DATA(file_inode(seq->file));
 995	struct raw_iter_state *state = raw_seq_private(seq);
 996
 997	for (state->bucket = 0; state->bucket < RAW_HTABLE_SIZE;
 998			++state->bucket) {
 999		sk_for_each(sk, &h->ht[state->bucket])
1000			if (sock_net(sk) == seq_file_net(seq))
1001				goto found;
1002	}
1003	sk = NULL;
1004found:
1005	return sk;
1006}
1007
1008static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
1009{
1010	struct raw_hashinfo *h = PDE_DATA(file_inode(seq->file));
1011	struct raw_iter_state *state = raw_seq_private(seq);
1012
1013	do {
1014		sk = sk_next(sk);
1015try_again:
1016		;
1017	} while (sk && sock_net(sk) != seq_file_net(seq));
1018
1019	if (!sk && ++state->bucket < RAW_HTABLE_SIZE) {
1020		sk = sk_head(&h->ht[state->bucket]);
1021		goto try_again;
1022	}
1023	return sk;
1024}
1025
1026static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
1027{
1028	struct sock *sk = raw_get_first(seq);
1029
1030	if (sk)
1031		while (pos && (sk = raw_get_next(seq, sk)) != NULL)
1032			--pos;
1033	return pos ? NULL : sk;
1034}
1035
1036void *raw_seq_start(struct seq_file *seq, loff_t *pos)
1037	__acquires(&h->lock)
1038{
1039	struct raw_hashinfo *h = PDE_DATA(file_inode(seq->file));
1040
1041	read_lock(&h->lock);
1042	return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1043}
1044EXPORT_SYMBOL_GPL(raw_seq_start);
1045
1046void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1047{
1048	struct sock *sk;
1049
1050	if (v == SEQ_START_TOKEN)
1051		sk = raw_get_first(seq);
1052	else
1053		sk = raw_get_next(seq, v);
1054	++*pos;
1055	return sk;
1056}
1057EXPORT_SYMBOL_GPL(raw_seq_next);
1058
1059void raw_seq_stop(struct seq_file *seq, void *v)
1060	__releases(&h->lock)
1061{
1062	struct raw_hashinfo *h = PDE_DATA(file_inode(seq->file));
1063
1064	read_unlock(&h->lock);
1065}
1066EXPORT_SYMBOL_GPL(raw_seq_stop);
1067
1068static void raw_sock_seq_show(struct seq_file *seq, struct sock *sp, int i)
1069{
1070	struct inet_sock *inet = inet_sk(sp);
1071	__be32 dest = inet->inet_daddr,
1072	       src = inet->inet_rcv_saddr;
1073	__u16 destp = 0,
1074	      srcp  = inet->inet_num;
1075
1076	seq_printf(seq, "%4d: %08X:%04X %08X:%04X"
1077		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u\n",
1078		i, src, srcp, dest, destp, sp->sk_state,
1079		sk_wmem_alloc_get(sp),
1080		sk_rmem_alloc_get(sp),
1081		0, 0L, 0,
1082		from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1083		0, sock_i_ino(sp),
1084		refcount_read(&sp->sk_refcnt), sp, atomic_read(&sp->sk_drops));
1085}
1086
1087static int raw_seq_show(struct seq_file *seq, void *v)
1088{
1089	if (v == SEQ_START_TOKEN)
1090		seq_printf(seq, "  sl  local_address rem_address   st tx_queue "
1091				"rx_queue tr tm->when retrnsmt   uid  timeout "
1092				"inode ref pointer drops\n");
1093	else
1094		raw_sock_seq_show(seq, v, raw_seq_private(seq)->bucket);
1095	return 0;
1096}
1097
1098static const struct seq_operations raw_seq_ops = {
1099	.start = raw_seq_start,
1100	.next  = raw_seq_next,
1101	.stop  = raw_seq_stop,
1102	.show  = raw_seq_show,
1103};
1104
1105static __net_init int raw_init_net(struct net *net)
1106{
1107	if (!proc_create_net_data("raw", 0444, net->proc_net, &raw_seq_ops,
1108			sizeof(struct raw_iter_state), &raw_v4_hashinfo))
1109		return -ENOMEM;
1110
1111	return 0;
1112}
1113
1114static __net_exit void raw_exit_net(struct net *net)
1115{
1116	remove_proc_entry("raw", net->proc_net);
1117}
1118
1119static __net_initdata struct pernet_operations raw_net_ops = {
1120	.init = raw_init_net,
1121	.exit = raw_exit_net,
1122};
1123
1124int __init raw_proc_init(void)
1125{
1126	return register_pernet_subsys(&raw_net_ops);
1127}
1128
1129void __init raw_proc_exit(void)
1130{
1131	unregister_pernet_subsys(&raw_net_ops);
1132}
1133#endif /* CONFIG_PROC_FS */
1134
1135static void raw_sysctl_init_net(struct net *net)
1136{
1137#ifdef CONFIG_NET_L3_MASTER_DEV
1138	net->ipv4.sysctl_raw_l3mdev_accept = 1;
1139#endif
1140}
1141
1142static int __net_init raw_sysctl_init(struct net *net)
1143{
1144	raw_sysctl_init_net(net);
1145	return 0;
1146}
1147
1148static struct pernet_operations __net_initdata raw_sysctl_ops = {
1149	.init	= raw_sysctl_init,
1150};
1151
1152void __init raw_init(void)
1153{
1154	raw_sysctl_init_net(&init_net);
1155	if (register_pernet_subsys(&raw_sysctl_ops))
1156		panic("RAW: failed to init sysctl parameters.\n");
1157}