Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *	NET3:	Implementation of the ICMP protocol layer.
   4 *
   5 *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
   6 *
 
 
 
 
 
   7 *	Some of the function names and the icmp unreach table for this
   8 *	module were derived from [icmp.c 1.0.11 06/02/93] by
   9 *	Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
  10 *	Other than that this module is a complete rewrite.
  11 *
  12 *	Fixes:
  13 *	Clemens Fruhwirth	:	introduce global icmp rate limiting
  14 *					with icmp type masking ability instead
  15 *					of broken per type icmp timeouts.
  16 *		Mike Shaver	:	RFC1122 checks.
  17 *		Alan Cox	:	Multicast ping reply as self.
  18 *		Alan Cox	:	Fix atomicity lockup in ip_build_xmit
  19 *					call.
  20 *		Alan Cox	:	Added 216,128 byte paths to the MTU
  21 *					code.
  22 *		Martin Mares	:	RFC1812 checks.
  23 *		Martin Mares	:	Can be configured to follow redirects
  24 *					if acting as a router _without_ a
  25 *					routing protocol (RFC 1812).
  26 *		Martin Mares	:	Echo requests may be configured to
  27 *					be ignored (RFC 1812).
  28 *		Martin Mares	:	Limitation of ICMP error message
  29 *					transmit rate (RFC 1812).
  30 *		Martin Mares	:	TOS and Precedence set correctly
  31 *					(RFC 1812).
  32 *		Martin Mares	:	Now copying as much data from the
  33 *					original packet as we can without
  34 *					exceeding 576 bytes (RFC 1812).
  35 *	Willy Konynenberg	:	Transparent proxying support.
  36 *		Keith Owens	:	RFC1191 correction for 4.2BSD based
  37 *					path MTU bug.
  38 *		Thomas Quinot	:	ICMP Dest Unreach codes up to 15 are
  39 *					valid (RFC 1812).
  40 *		Andi Kleen	:	Check all packet lengths properly
  41 *					and moved all kfree_skb() up to
  42 *					icmp_rcv.
  43 *		Andi Kleen	:	Move the rate limit bookkeeping
  44 *					into the dest entry and use a token
  45 *					bucket filter (thanks to ANK). Make
  46 *					the rates sysctl configurable.
  47 *		Yu Tianli	:	Fixed two ugly bugs in icmp_send
  48 *					- IP option length was accounted wrongly
  49 *					- ICMP header length was not accounted
  50 *					  at all.
  51 *              Tristan Greaves :       Added sysctl option to ignore bogus
  52 *              			broadcast responses from broken routers.
  53 *
  54 * To Fix:
  55 *
  56 *	- Should use skb_pull() instead of all the manual checking.
  57 *	  This would also greatly simply some upper layer error handlers. --AK
 
  58 */
  59
  60#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  61
  62#include <linux/module.h>
  63#include <linux/types.h>
  64#include <linux/jiffies.h>
  65#include <linux/kernel.h>
  66#include <linux/fcntl.h>
  67#include <linux/socket.h>
  68#include <linux/in.h>
  69#include <linux/inet.h>
  70#include <linux/inetdevice.h>
  71#include <linux/netdevice.h>
  72#include <linux/string.h>
  73#include <linux/netfilter_ipv4.h>
  74#include <linux/slab.h>
  75#include <net/snmp.h>
  76#include <net/ip.h>
  77#include <net/route.h>
  78#include <net/protocol.h>
  79#include <net/icmp.h>
  80#include <net/tcp.h>
  81#include <net/udp.h>
  82#include <net/raw.h>
  83#include <net/ping.h>
  84#include <linux/skbuff.h>
  85#include <net/sock.h>
  86#include <linux/errno.h>
  87#include <linux/timer.h>
  88#include <linux/init.h>
  89#include <linux/uaccess.h>
  90#include <net/checksum.h>
  91#include <net/xfrm.h>
  92#include <net/inet_common.h>
  93#include <net/ip_fib.h>
  94#include <net/l3mdev.h>
  95
  96/*
  97 *	Build xmit assembly blocks
  98 */
  99
 100struct icmp_bxm {
 101	struct sk_buff *skb;
 102	int offset;
 103	int data_len;
 104
 105	struct {
 106		struct icmphdr icmph;
 107		__be32	       times[3];
 108	} data;
 109	int head_len;
 110	struct ip_options_data replyopts;
 111};
 112
 113/* An array of errno for error messages from dest unreach. */
 114/* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
 115
 116const struct icmp_err icmp_err_convert[] = {
 117	{
 118		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
 119		.fatal = 0,
 120	},
 121	{
 122		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
 123		.fatal = 0,
 124	},
 125	{
 126		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
 127		.fatal = 1,
 128	},
 129	{
 130		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
 131		.fatal = 1,
 132	},
 133	{
 134		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
 135		.fatal = 0,
 136	},
 137	{
 138		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
 139		.fatal = 0,
 140	},
 141	{
 142		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
 143		.fatal = 1,
 144	},
 145	{
 146		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
 147		.fatal = 1,
 148	},
 149	{
 150		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
 151		.fatal = 1,
 152	},
 153	{
 154		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
 155		.fatal = 1,
 156	},
 157	{
 158		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
 159		.fatal = 1,
 160	},
 161	{
 162		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
 163		.fatal = 0,
 164	},
 165	{
 166		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
 167		.fatal = 0,
 168	},
 169	{
 170		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
 171		.fatal = 1,
 172	},
 173	{
 174		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
 175		.fatal = 1,
 176	},
 177	{
 178		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
 179		.fatal = 1,
 180	},
 181};
 182EXPORT_SYMBOL(icmp_err_convert);
 183
 184/*
 185 *	ICMP control array. This specifies what to do with each ICMP.
 186 */
 187
 188struct icmp_control {
 189	bool (*handler)(struct sk_buff *skb);
 190	short   error;		/* This ICMP is classed as an error message */
 191};
 192
 193static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 194
 195/*
 196 *	The ICMP socket(s). This is the most convenient way to flow control
 197 *	our ICMP output as well as maintain a clean interface throughout
 198 *	all layers. All Socketless IP sends will soon be gone.
 199 *
 200 *	On SMP we have one ICMP socket per-cpu.
 201 */
 202static struct sock *icmp_sk(struct net *net)
 203{
 204	return this_cpu_read(*net->ipv4.icmp_sk);
 205}
 206
 207/* Called with BH disabled */
 208static inline struct sock *icmp_xmit_lock(struct net *net)
 209{
 210	struct sock *sk;
 211
 212	sk = icmp_sk(net);
 213
 214	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 215		/* This can happen if the output path signals a
 216		 * dst_link_failure() for an outgoing ICMP packet.
 217		 */
 218		return NULL;
 219	}
 220	return sk;
 221}
 222
 223static inline void icmp_xmit_unlock(struct sock *sk)
 224{
 225	spin_unlock(&sk->sk_lock.slock);
 226}
 227
 228int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
 229int sysctl_icmp_msgs_burst __read_mostly = 50;
 230
 231static struct {
 232	spinlock_t	lock;
 233	u32		credit;
 234	u32		stamp;
 235} icmp_global = {
 236	.lock		= __SPIN_LOCK_UNLOCKED(icmp_global.lock),
 237};
 238
 239/**
 240 * icmp_global_allow - Are we allowed to send one more ICMP message ?
 241 *
 242 * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec.
 243 * Returns false if we reached the limit and can not send another packet.
 244 * Note: called with BH disabled
 245 */
 246bool icmp_global_allow(void)
 247{
 248	u32 credit, delta, incr = 0, now = (u32)jiffies;
 249	bool rc = false;
 250
 251	/* Check if token bucket is empty and cannot be refilled
 252	 * without taking the spinlock.
 253	 */
 254	if (!icmp_global.credit) {
 255		delta = min_t(u32, now - icmp_global.stamp, HZ);
 256		if (delta < HZ / 50)
 257			return false;
 258	}
 259
 260	spin_lock(&icmp_global.lock);
 261	delta = min_t(u32, now - icmp_global.stamp, HZ);
 262	if (delta >= HZ / 50) {
 263		incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
 264		if (incr)
 265			icmp_global.stamp = now;
 266	}
 267	credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
 268	if (credit) {
 269		credit--;
 270		rc = true;
 271	}
 272	icmp_global.credit = credit;
 273	spin_unlock(&icmp_global.lock);
 274	return rc;
 275}
 276EXPORT_SYMBOL(icmp_global_allow);
 277
 278static bool icmpv4_mask_allow(struct net *net, int type, int code)
 279{
 280	if (type > NR_ICMP_TYPES)
 281		return true;
 282
 283	/* Don't limit PMTU discovery. */
 284	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 285		return true;
 286
 287	/* Limit if icmp type is enabled in ratemask. */
 288	if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
 289		return true;
 290
 291	return false;
 292}
 293
 294static bool icmpv4_global_allow(struct net *net, int type, int code)
 295{
 296	if (icmpv4_mask_allow(net, type, code))
 297		return true;
 298
 299	if (icmp_global_allow())
 300		return true;
 301
 302	return false;
 303}
 304
 305/*
 306 *	Send an ICMP frame.
 307 */
 308
 309static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 310			       struct flowi4 *fl4, int type, int code)
 311{
 312	struct dst_entry *dst = &rt->dst;
 313	struct inet_peer *peer;
 314	bool rc = true;
 315	int vif;
 316
 317	if (icmpv4_mask_allow(net, type, code))
 318		goto out;
 319
 320	/* No rate limit on loopback */
 321	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 322		goto out;
 323
 324	vif = l3mdev_master_ifindex(dst->dev);
 325	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
 326	rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
 327	if (peer)
 328		inet_putpeer(peer);
 329out:
 330	return rc;
 331}
 332
 333/*
 334 *	Maintain the counters used in the SNMP statistics for outgoing ICMP
 335 */
 336void icmp_out_count(struct net *net, unsigned char type)
 337{
 338	ICMPMSGOUT_INC_STATS(net, type);
 339	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 340}
 341
 342/*
 343 *	Checksum each fragment, and on the first include the headers and final
 344 *	checksum.
 345 */
 346static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 347			  struct sk_buff *skb)
 348{
 349	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
 350	__wsum csum;
 351
 352	csum = skb_copy_and_csum_bits(icmp_param->skb,
 353				      icmp_param->offset + offset,
 354				      to, len, 0);
 355
 356	skb->csum = csum_block_add(skb->csum, csum, odd);
 357	if (icmp_pointers[icmp_param->data.icmph.type].error)
 358		nf_ct_attach(skb, icmp_param->skb);
 359	return 0;
 360}
 361
 362static void icmp_push_reply(struct icmp_bxm *icmp_param,
 363			    struct flowi4 *fl4,
 364			    struct ipcm_cookie *ipc, struct rtable **rt)
 365{
 366	struct sock *sk;
 367	struct sk_buff *skb;
 368
 369	sk = icmp_sk(dev_net((*rt)->dst.dev));
 370	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
 371			   icmp_param->data_len+icmp_param->head_len,
 372			   icmp_param->head_len,
 373			   ipc, rt, MSG_DONTWAIT) < 0) {
 374		__ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
 375		ip_flush_pending_frames(sk);
 376	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 377		struct icmphdr *icmph = icmp_hdr(skb);
 378		__wsum csum = 0;
 379		struct sk_buff *skb1;
 380
 381		skb_queue_walk(&sk->sk_write_queue, skb1) {
 382			csum = csum_add(csum, skb1->csum);
 383		}
 384		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 385						 (char *)icmph,
 386						 icmp_param->head_len, csum);
 387		icmph->checksum = csum_fold(csum);
 388		skb->ip_summed = CHECKSUM_NONE;
 389		ip_push_pending_frames(sk, fl4);
 390	}
 391}
 392
 393/*
 394 *	Driving logic for building and sending ICMP messages.
 395 */
 396
 397static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 398{
 399	struct ipcm_cookie ipc;
 400	struct rtable *rt = skb_rtable(skb);
 401	struct net *net = dev_net(rt->dst.dev);
 402	struct flowi4 fl4;
 403	struct sock *sk;
 404	struct inet_sock *inet;
 405	__be32 daddr, saddr;
 406	u32 mark = IP4_REPLY_MARK(net, skb->mark);
 407	int type = icmp_param->data.icmph.type;
 408	int code = icmp_param->data.icmph.code;
 409
 410	if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
 411		return;
 412
 413	/* Needed by both icmp_global_allow and icmp_xmit_lock */
 414	local_bh_disable();
 415
 416	/* global icmp_msgs_per_sec */
 417	if (!icmpv4_global_allow(net, type, code))
 418		goto out_bh_enable;
 419
 420	sk = icmp_xmit_lock(net);
 421	if (!sk)
 422		goto out_bh_enable;
 423	inet = inet_sk(sk);
 424
 425	icmp_param->data.icmph.checksum = 0;
 426
 427	ipcm_init(&ipc);
 428	inet->tos = ip_hdr(skb)->tos;
 429	sk->sk_mark = mark;
 430	daddr = ipc.addr = ip_hdr(skb)->saddr;
 431	saddr = fib_compute_spec_dst(skb);
 
 
 
 
 432
 433	if (icmp_param->replyopts.opt.opt.optlen) {
 434		ipc.opt = &icmp_param->replyopts.opt;
 435		if (ipc.opt->opt.srr)
 436			daddr = icmp_param->replyopts.opt.opt.faddr;
 437	}
 438	memset(&fl4, 0, sizeof(fl4));
 439	fl4.daddr = daddr;
 440	fl4.saddr = saddr;
 441	fl4.flowi4_mark = mark;
 442	fl4.flowi4_uid = sock_net_uid(net, NULL);
 443	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 444	fl4.flowi4_proto = IPPROTO_ICMP;
 445	fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
 446	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 447	rt = ip_route_output_key(net, &fl4);
 448	if (IS_ERR(rt))
 449		goto out_unlock;
 450	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 451		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
 452	ip_rt_put(rt);
 453out_unlock:
 454	icmp_xmit_unlock(sk);
 455out_bh_enable:
 456	local_bh_enable();
 457}
 458
 459static struct rtable *icmp_route_lookup(struct net *net,
 460					struct flowi4 *fl4,
 461					struct sk_buff *skb_in,
 462					const struct iphdr *iph,
 463					__be32 saddr, u8 tos, u32 mark,
 464					int type, int code,
 465					struct icmp_bxm *param)
 466{
 467	struct rtable *rt, *rt2;
 468	struct flowi4 fl4_dec;
 469	int err;
 470
 471	memset(fl4, 0, sizeof(*fl4));
 472	fl4->daddr = (param->replyopts.opt.opt.srr ?
 473		      param->replyopts.opt.opt.faddr : iph->saddr);
 474	fl4->saddr = saddr;
 475	fl4->flowi4_mark = mark;
 476	fl4->flowi4_uid = sock_net_uid(net, NULL);
 477	fl4->flowi4_tos = RT_TOS(tos);
 478	fl4->flowi4_proto = IPPROTO_ICMP;
 479	fl4->fl4_icmp_type = type;
 480	fl4->fl4_icmp_code = code;
 481	fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
 482
 483	security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
 484	rt = ip_route_output_key_hash(net, fl4, skb_in);
 485	if (IS_ERR(rt))
 486		return rt;
 487
 488	/* No need to clone since we're just using its address. */
 489	rt2 = rt;
 490
 491	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 492					   flowi4_to_flowi(fl4), NULL, 0);
 493	if (!IS_ERR(rt)) {
 494		if (rt != rt2)
 495			return rt;
 496	} else if (PTR_ERR(rt) == -EPERM) {
 497		rt = NULL;
 498	} else
 499		return rt;
 500
 501	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 502	if (err)
 503		goto relookup_failed;
 504
 505	if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
 506				     fl4_dec.saddr) == RTN_LOCAL) {
 507		rt2 = __ip_route_output_key(net, &fl4_dec);
 508		if (IS_ERR(rt2))
 509			err = PTR_ERR(rt2);
 510	} else {
 511		struct flowi4 fl4_2 = {};
 512		unsigned long orefdst;
 513
 514		fl4_2.daddr = fl4_dec.saddr;
 515		rt2 = ip_route_output_key(net, &fl4_2);
 516		if (IS_ERR(rt2)) {
 517			err = PTR_ERR(rt2);
 518			goto relookup_failed;
 519		}
 520		/* Ugh! */
 521		orefdst = skb_in->_skb_refdst; /* save old refdst */
 522		skb_dst_set(skb_in, NULL);
 523		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 524				     RT_TOS(tos), rt2->dst.dev);
 525
 526		dst_release(&rt2->dst);
 527		rt2 = skb_rtable(skb_in);
 528		skb_in->_skb_refdst = orefdst; /* restore old refdst */
 529	}
 530
 531	if (err)
 532		goto relookup_failed;
 533
 534	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 535					    flowi4_to_flowi(&fl4_dec), NULL,
 536					    XFRM_LOOKUP_ICMP);
 537	if (!IS_ERR(rt2)) {
 538		dst_release(&rt->dst);
 539		memcpy(fl4, &fl4_dec, sizeof(*fl4));
 540		rt = rt2;
 541	} else if (PTR_ERR(rt2) == -EPERM) {
 542		if (rt)
 543			dst_release(&rt->dst);
 544		return rt2;
 545	} else {
 546		err = PTR_ERR(rt2);
 547		goto relookup_failed;
 548	}
 549	return rt;
 550
 551relookup_failed:
 552	if (rt)
 553		return rt;
 554	return ERR_PTR(err);
 555}
 556
 557/*
 558 *	Send an ICMP message in response to a situation
 559 *
 560 *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
 561 *		  MAY send more (we do).
 562 *			MUST NOT change this header information.
 563 *			MUST NOT reply to a multicast/broadcast IP address.
 564 *			MUST NOT reply to a multicast/broadcast MAC address.
 565 *			MUST reply to only the first fragment.
 566 */
 567
 568void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
 569		 const struct ip_options *opt)
 570{
 571	struct iphdr *iph;
 572	int room;
 573	struct icmp_bxm icmp_param;
 574	struct rtable *rt = skb_rtable(skb_in);
 575	struct ipcm_cookie ipc;
 576	struct flowi4 fl4;
 577	__be32 saddr;
 578	u8  tos;
 579	u32 mark;
 580	struct net *net;
 581	struct sock *sk;
 582
 583	if (!rt)
 584		goto out;
 585
 586	if (rt->dst.dev)
 587		net = dev_net(rt->dst.dev);
 588	else if (skb_in->dev)
 589		net = dev_net(skb_in->dev);
 590	else
 591		goto out;
 592
 593	/*
 594	 *	Find the original header. It is expected to be valid, of course.
 595	 *	Check this, icmp_send is called from the most obscure devices
 596	 *	sometimes.
 597	 */
 598	iph = ip_hdr(skb_in);
 599
 600	if ((u8 *)iph < skb_in->head ||
 601	    (skb_network_header(skb_in) + sizeof(*iph)) >
 602	    skb_tail_pointer(skb_in))
 603		goto out;
 604
 605	/*
 606	 *	No replies to physical multicast/broadcast
 607	 */
 608	if (skb_in->pkt_type != PACKET_HOST)
 609		goto out;
 610
 611	/*
 612	 *	Now check at the protocol level
 613	 */
 614	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 615		goto out;
 616
 617	/*
 618	 *	Only reply to fragment 0. We byte re-order the constant
 619	 *	mask for efficiency.
 620	 */
 621	if (iph->frag_off & htons(IP_OFFSET))
 622		goto out;
 623
 624	/*
 625	 *	If we send an ICMP error to an ICMP error a mess would result..
 626	 */
 627	if (icmp_pointers[type].error) {
 628		/*
 629		 *	We are an error, check if we are replying to an
 630		 *	ICMP error
 631		 */
 632		if (iph->protocol == IPPROTO_ICMP) {
 633			u8 _inner_type, *itp;
 634
 635			itp = skb_header_pointer(skb_in,
 636						 skb_network_header(skb_in) +
 637						 (iph->ihl << 2) +
 638						 offsetof(struct icmphdr,
 639							  type) -
 640						 skb_in->data,
 641						 sizeof(_inner_type),
 642						 &_inner_type);
 643			if (!itp)
 644				goto out;
 645
 646			/*
 647			 *	Assume any unknown ICMP type is an error. This
 648			 *	isn't specified by the RFC, but think about it..
 649			 */
 650			if (*itp > NR_ICMP_TYPES ||
 651			    icmp_pointers[*itp].error)
 652				goto out;
 653		}
 654	}
 655
 656	/* Needed by both icmp_global_allow and icmp_xmit_lock */
 657	local_bh_disable();
 658
 659	/* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
 660	 * incoming dev is loopback.  If outgoing dev change to not be
 661	 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
 662	 */
 663	if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
 664	      !icmpv4_global_allow(net, type, code))
 665		goto out_bh_enable;
 666
 667	sk = icmp_xmit_lock(net);
 668	if (!sk)
 669		goto out_bh_enable;
 670
 671	/*
 672	 *	Construct source address and options.
 673	 */
 674
 675	saddr = iph->daddr;
 676	if (!(rt->rt_flags & RTCF_LOCAL)) {
 677		struct net_device *dev = NULL;
 678
 679		rcu_read_lock();
 680		if (rt_is_input_route(rt) &&
 681		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 682			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 683
 684		if (dev)
 685			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
 686		else
 687			saddr = 0;
 688		rcu_read_unlock();
 689	}
 690
 691	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 692					   IPTOS_PREC_INTERNETCONTROL) :
 693					  iph->tos;
 694	mark = IP4_REPLY_MARK(net, skb_in->mark);
 695
 696	if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
 697		goto out_unlock;
 698
 699
 700	/*
 701	 *	Prepare data for ICMP header.
 702	 */
 703
 704	icmp_param.data.icmph.type	 = type;
 705	icmp_param.data.icmph.code	 = code;
 706	icmp_param.data.icmph.un.gateway = info;
 707	icmp_param.data.icmph.checksum	 = 0;
 708	icmp_param.skb	  = skb_in;
 709	icmp_param.offset = skb_network_offset(skb_in);
 710	inet_sk(sk)->tos = tos;
 711	sk->sk_mark = mark;
 712	ipcm_init(&ipc);
 713	ipc.addr = iph->saddr;
 714	ipc.opt = &icmp_param.replyopts.opt;
 
 
 
 715
 716	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
 717			       type, code, &icmp_param);
 718	if (IS_ERR(rt))
 719		goto out_unlock;
 720
 721	/* peer icmp_ratelimit */
 722	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 723		goto ende;
 724
 725	/* RFC says return as much as we can without exceeding 576 bytes. */
 726
 727	room = dst_mtu(&rt->dst);
 728	if (room > 576)
 729		room = 576;
 730	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 731	room -= sizeof(struct icmphdr);
 732
 733	icmp_param.data_len = skb_in->len - icmp_param.offset;
 734	if (icmp_param.data_len > room)
 735		icmp_param.data_len = room;
 736	icmp_param.head_len = sizeof(struct icmphdr);
 737
 738	icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
 739ende:
 740	ip_rt_put(rt);
 741out_unlock:
 742	icmp_xmit_unlock(sk);
 743out_bh_enable:
 744	local_bh_enable();
 745out:;
 746}
 747EXPORT_SYMBOL(__icmp_send);
 748
 749
 750static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 751{
 752	const struct iphdr *iph = (const struct iphdr *) skb->data;
 753	const struct net_protocol *ipprot;
 754	int protocol = iph->protocol;
 755
 756	/* Checkin full IP header plus 8 bytes of protocol to
 757	 * avoid additional coding at protocol handlers.
 758	 */
 759	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 760		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 761		return;
 762	}
 763
 764	raw_icmp_error(skb, protocol, info);
 765
 766	ipprot = rcu_dereference(inet_protos[protocol]);
 767	if (ipprot && ipprot->err_handler)
 768		ipprot->err_handler(skb, info);
 769}
 770
 771static bool icmp_tag_validation(int proto)
 772{
 773	bool ok;
 774
 775	rcu_read_lock();
 776	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 777	rcu_read_unlock();
 778	return ok;
 779}
 780
 781/*
 782 *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 783 *	ICMP_PARAMETERPROB.
 784 */
 785
 786static bool icmp_unreach(struct sk_buff *skb)
 787{
 788	const struct iphdr *iph;
 789	struct icmphdr *icmph;
 790	struct net *net;
 791	u32 info = 0;
 792
 793	net = dev_net(skb_dst(skb)->dev);
 794
 795	/*
 796	 *	Incomplete header ?
 797	 * 	Only checks for the IP header, there should be an
 798	 *	additional check for longer headers in upper levels.
 799	 */
 800
 801	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 802		goto out_err;
 803
 804	icmph = icmp_hdr(skb);
 805	iph   = (const struct iphdr *)skb->data;
 806
 807	if (iph->ihl < 5) /* Mangled header, drop. */
 808		goto out_err;
 809
 810	switch (icmph->type) {
 811	case ICMP_DEST_UNREACH:
 812		switch (icmph->code & 15) {
 813		case ICMP_NET_UNREACH:
 814		case ICMP_HOST_UNREACH:
 815		case ICMP_PROT_UNREACH:
 816		case ICMP_PORT_UNREACH:
 817			break;
 818		case ICMP_FRAG_NEEDED:
 819			/* for documentation of the ip_no_pmtu_disc
 820			 * values please see
 821			 * Documentation/networking/ip-sysctl.txt
 822			 */
 823			switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
 824			default:
 825				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 826						    &iph->daddr);
 827				break;
 828			case 2:
 829				goto out;
 830			case 3:
 831				if (!icmp_tag_validation(iph->protocol))
 832					goto out;
 833				/* fall through */
 834			case 0:
 835				info = ntohs(icmph->un.frag.mtu);
 836			}
 837			break;
 838		case ICMP_SR_FAILED:
 839			net_dbg_ratelimited("%pI4: Source Route Failed\n",
 840					    &iph->daddr);
 841			break;
 842		default:
 843			break;
 844		}
 845		if (icmph->code > NR_ICMP_UNREACH)
 846			goto out;
 847		break;
 848	case ICMP_PARAMETERPROB:
 849		info = ntohl(icmph->un.gateway) >> 24;
 850		break;
 851	case ICMP_TIME_EXCEEDED:
 852		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 853		if (icmph->code == ICMP_EXC_FRAGTIME)
 854			goto out;
 855		break;
 856	}
 857
 858	/*
 859	 *	Throw it at our lower layers
 860	 *
 861	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 862	 *		  header.
 863	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 864	 *		  transport layer.
 865	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 866	 *		  transport layer.
 867	 */
 868
 869	/*
 870	 *	Check the other end isn't violating RFC 1122. Some routers send
 871	 *	bogus responses to broadcast frames. If you see this message
 872	 *	first check your netmask matches at both ends, if it does then
 873	 *	get the other vendor to fix their kit.
 874	 */
 875
 876	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 877	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 878		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 879				     &ip_hdr(skb)->saddr,
 880				     icmph->type, icmph->code,
 881				     &iph->daddr, skb->dev->name);
 882		goto out;
 883	}
 884
 885	icmp_socket_deliver(skb, info);
 886
 887out:
 888	return true;
 889out_err:
 890	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 891	return false;
 892}
 893
 894
 895/*
 896 *	Handle ICMP_REDIRECT.
 897 */
 898
 899static bool icmp_redirect(struct sk_buff *skb)
 900{
 901	if (skb->len < sizeof(struct iphdr)) {
 902		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 903		return false;
 904	}
 905
 906	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 907		/* there aught to be a stat */
 908		return false;
 909	}
 910
 911	icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
 912	return true;
 913}
 914
 915/*
 916 *	Handle ICMP_ECHO ("ping") requests.
 917 *
 918 *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 919 *		  requests.
 920 *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 921 *		  included in the reply.
 922 *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 923 *		  echo requests, MUST have default=NOT.
 924 *	See also WRT handling of options once they are done and working.
 925 */
 926
 927static bool icmp_echo(struct sk_buff *skb)
 928{
 929	struct net *net;
 930
 931	net = dev_net(skb_dst(skb)->dev);
 932	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
 933		struct icmp_bxm icmp_param;
 934
 935		icmp_param.data.icmph	   = *icmp_hdr(skb);
 936		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 937		icmp_param.skb		   = skb;
 938		icmp_param.offset	   = 0;
 939		icmp_param.data_len	   = skb->len;
 940		icmp_param.head_len	   = sizeof(struct icmphdr);
 941		icmp_reply(&icmp_param, skb);
 942	}
 943	/* should there be an ICMP stat for ignored echos? */
 944	return true;
 945}
 946
 947/*
 948 *	Handle ICMP Timestamp requests.
 949 *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 950 *		  SHOULD be in the kernel for minimum random latency.
 951 *		  MUST be accurate to a few minutes.
 952 *		  MUST be updated at least at 15Hz.
 953 */
 954static bool icmp_timestamp(struct sk_buff *skb)
 955{
 956	struct icmp_bxm icmp_param;
 957	/*
 958	 *	Too short.
 959	 */
 960	if (skb->len < 4)
 961		goto out_err;
 962
 963	/*
 964	 *	Fill in the current time as ms since midnight UT:
 965	 */
 966	icmp_param.data.times[1] = inet_current_timestamp();
 967	icmp_param.data.times[2] = icmp_param.data.times[1];
 968
 969	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
 970
 971	icmp_param.data.icmph	   = *icmp_hdr(skb);
 972	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
 973	icmp_param.data.icmph.code = 0;
 974	icmp_param.skb		   = skb;
 975	icmp_param.offset	   = 0;
 976	icmp_param.data_len	   = 0;
 977	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
 978	icmp_reply(&icmp_param, skb);
 979	return true;
 980
 981out_err:
 982	__ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
 983	return false;
 984}
 985
 986static bool icmp_discard(struct sk_buff *skb)
 987{
 988	/* pretend it was a success */
 989	return true;
 990}
 991
 992/*
 993 *	Deal with incoming ICMP packets.
 994 */
 995int icmp_rcv(struct sk_buff *skb)
 996{
 997	struct icmphdr *icmph;
 998	struct rtable *rt = skb_rtable(skb);
 999	struct net *net = dev_net(rt->dst.dev);
1000	bool success;
1001
1002	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1003		struct sec_path *sp = skb_sec_path(skb);
1004		int nh;
1005
1006		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1007				 XFRM_STATE_ICMP))
1008			goto drop;
1009
1010		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1011			goto drop;
1012
1013		nh = skb_network_offset(skb);
1014		skb_set_network_header(skb, sizeof(*icmph));
1015
1016		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1017			goto drop;
1018
1019		skb_set_network_header(skb, nh);
1020	}
1021
1022	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1023
1024	if (skb_checksum_simple_validate(skb))
1025		goto csum_error;
1026
1027	if (!pskb_pull(skb, sizeof(*icmph)))
1028		goto error;
1029
1030	icmph = icmp_hdr(skb);
1031
1032	ICMPMSGIN_INC_STATS(net, icmph->type);
1033	/*
1034	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1035	 *
1036	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1037	 *		  discarded.
1038	 */
1039	if (icmph->type > NR_ICMP_TYPES)
1040		goto error;
1041
1042
1043	/*
1044	 *	Parse the ICMP message
1045	 */
1046
1047	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1048		/*
1049		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1050		 *	  silently ignored (we let user decide with a sysctl).
1051		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1052		 *	  discarded if to broadcast/multicast.
1053		 */
1054		if ((icmph->type == ICMP_ECHO ||
1055		     icmph->type == ICMP_TIMESTAMP) &&
1056		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1057			goto error;
1058		}
1059		if (icmph->type != ICMP_ECHO &&
1060		    icmph->type != ICMP_TIMESTAMP &&
1061		    icmph->type != ICMP_ADDRESS &&
1062		    icmph->type != ICMP_ADDRESSREPLY) {
1063			goto error;
1064		}
1065	}
1066
1067	success = icmp_pointers[icmph->type].handler(skb);
1068
1069	if (success)  {
1070		consume_skb(skb);
1071		return NET_RX_SUCCESS;
1072	}
1073
1074drop:
1075	kfree_skb(skb);
1076	return NET_RX_DROP;
1077csum_error:
1078	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1079error:
1080	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1081	goto drop;
1082}
1083
1084int icmp_err(struct sk_buff *skb, u32 info)
1085{
1086	struct iphdr *iph = (struct iphdr *)skb->data;
1087	int offset = iph->ihl<<2;
1088	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1089	int type = icmp_hdr(skb)->type;
1090	int code = icmp_hdr(skb)->code;
1091	struct net *net = dev_net(skb->dev);
1092
1093	/*
1094	 * Use ping_err to handle all icmp errors except those
1095	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1096	 */
1097	if (icmph->type != ICMP_ECHOREPLY) {
1098		ping_err(skb, offset, info);
1099		return 0;
1100	}
1101
1102	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1103		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1104	else if (type == ICMP_REDIRECT)
1105		ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1106
1107	return 0;
1108}
1109
1110/*
1111 *	This table is the definition of how we handle ICMP.
1112 */
1113static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1114	[ICMP_ECHOREPLY] = {
1115		.handler = ping_rcv,
1116	},
1117	[1] = {
1118		.handler = icmp_discard,
1119		.error = 1,
1120	},
1121	[2] = {
1122		.handler = icmp_discard,
1123		.error = 1,
1124	},
1125	[ICMP_DEST_UNREACH] = {
1126		.handler = icmp_unreach,
1127		.error = 1,
1128	},
1129	[ICMP_SOURCE_QUENCH] = {
1130		.handler = icmp_unreach,
1131		.error = 1,
1132	},
1133	[ICMP_REDIRECT] = {
1134		.handler = icmp_redirect,
1135		.error = 1,
1136	},
1137	[6] = {
1138		.handler = icmp_discard,
1139		.error = 1,
1140	},
1141	[7] = {
1142		.handler = icmp_discard,
1143		.error = 1,
1144	},
1145	[ICMP_ECHO] = {
1146		.handler = icmp_echo,
1147	},
1148	[9] = {
1149		.handler = icmp_discard,
1150		.error = 1,
1151	},
1152	[10] = {
1153		.handler = icmp_discard,
1154		.error = 1,
1155	},
1156	[ICMP_TIME_EXCEEDED] = {
1157		.handler = icmp_unreach,
1158		.error = 1,
1159	},
1160	[ICMP_PARAMETERPROB] = {
1161		.handler = icmp_unreach,
1162		.error = 1,
1163	},
1164	[ICMP_TIMESTAMP] = {
1165		.handler = icmp_timestamp,
1166	},
1167	[ICMP_TIMESTAMPREPLY] = {
1168		.handler = icmp_discard,
1169	},
1170	[ICMP_INFO_REQUEST] = {
1171		.handler = icmp_discard,
1172	},
1173	[ICMP_INFO_REPLY] = {
1174		.handler = icmp_discard,
1175	},
1176	[ICMP_ADDRESS] = {
1177		.handler = icmp_discard,
1178	},
1179	[ICMP_ADDRESSREPLY] = {
1180		.handler = icmp_discard,
1181	},
1182};
1183
1184static void __net_exit icmp_sk_exit(struct net *net)
1185{
1186	int i;
1187
1188	for_each_possible_cpu(i)
1189		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1190	free_percpu(net->ipv4.icmp_sk);
1191	net->ipv4.icmp_sk = NULL;
1192}
1193
1194static int __net_init icmp_sk_init(struct net *net)
1195{
1196	int i, err;
1197
1198	net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1199	if (!net->ipv4.icmp_sk)
1200		return -ENOMEM;
1201
1202	for_each_possible_cpu(i) {
1203		struct sock *sk;
1204
1205		err = inet_ctl_sock_create(&sk, PF_INET,
1206					   SOCK_RAW, IPPROTO_ICMP, net);
1207		if (err < 0)
1208			goto fail;
1209
1210		*per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1211
1212		/* Enough space for 2 64K ICMP packets, including
1213		 * sk_buff/skb_shared_info struct overhead.
1214		 */
1215		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1216
1217		/*
1218		 * Speedup sock_wfree()
1219		 */
1220		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1221		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1222	}
1223
1224	/* Control parameters for ECHO replies. */
1225	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1226	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1227
1228	/* Control parameter - ignore bogus broadcast responses? */
1229	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1230
1231	/*
1232	 * 	Configurable global rate limit.
1233	 *
1234	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1235	 *	bucket ratemask defines which icmp types are ratelimited by
1236	 *	setting	it's bit position.
1237	 *
1238	 *	default:
1239	 *	dest unreachable (3), source quench (4),
1240	 *	time exceeded (11), parameter problem (12)
1241	 */
1242
1243	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1244	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1245	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1246
1247	return 0;
1248
1249fail:
1250	icmp_sk_exit(net);
 
 
1251	return err;
1252}
1253
1254static struct pernet_operations __net_initdata icmp_sk_ops = {
1255       .init = icmp_sk_init,
1256       .exit = icmp_sk_exit,
1257};
1258
1259int __init icmp_init(void)
1260{
1261	return register_pernet_subsys(&icmp_sk_ops);
1262}
v4.17
 
   1/*
   2 *	NET3:	Implementation of the ICMP protocol layer.
   3 *
   4 *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5 *
   6 *	This program is free software; you can redistribute it and/or
   7 *	modify it under the terms of the GNU General Public License
   8 *	as published by the Free Software Foundation; either version
   9 *	2 of the License, or (at your option) any later version.
  10 *
  11 *	Some of the function names and the icmp unreach table for this
  12 *	module were derived from [icmp.c 1.0.11 06/02/93] by
  13 *	Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
  14 *	Other than that this module is a complete rewrite.
  15 *
  16 *	Fixes:
  17 *	Clemens Fruhwirth	:	introduce global icmp rate limiting
  18 *					with icmp type masking ability instead
  19 *					of broken per type icmp timeouts.
  20 *		Mike Shaver	:	RFC1122 checks.
  21 *		Alan Cox	:	Multicast ping reply as self.
  22 *		Alan Cox	:	Fix atomicity lockup in ip_build_xmit
  23 *					call.
  24 *		Alan Cox	:	Added 216,128 byte paths to the MTU
  25 *					code.
  26 *		Martin Mares	:	RFC1812 checks.
  27 *		Martin Mares	:	Can be configured to follow redirects
  28 *					if acting as a router _without_ a
  29 *					routing protocol (RFC 1812).
  30 *		Martin Mares	:	Echo requests may be configured to
  31 *					be ignored (RFC 1812).
  32 *		Martin Mares	:	Limitation of ICMP error message
  33 *					transmit rate (RFC 1812).
  34 *		Martin Mares	:	TOS and Precedence set correctly
  35 *					(RFC 1812).
  36 *		Martin Mares	:	Now copying as much data from the
  37 *					original packet as we can without
  38 *					exceeding 576 bytes (RFC 1812).
  39 *	Willy Konynenberg	:	Transparent proxying support.
  40 *		Keith Owens	:	RFC1191 correction for 4.2BSD based
  41 *					path MTU bug.
  42 *		Thomas Quinot	:	ICMP Dest Unreach codes up to 15 are
  43 *					valid (RFC 1812).
  44 *		Andi Kleen	:	Check all packet lengths properly
  45 *					and moved all kfree_skb() up to
  46 *					icmp_rcv.
  47 *		Andi Kleen	:	Move the rate limit bookkeeping
  48 *					into the dest entry and use a token
  49 *					bucket filter (thanks to ANK). Make
  50 *					the rates sysctl configurable.
  51 *		Yu Tianli	:	Fixed two ugly bugs in icmp_send
  52 *					- IP option length was accounted wrongly
  53 *					- ICMP header length was not accounted
  54 *					  at all.
  55 *              Tristan Greaves :       Added sysctl option to ignore bogus
  56 *              			broadcast responses from broken routers.
  57 *
  58 * To Fix:
  59 *
  60 *	- Should use skb_pull() instead of all the manual checking.
  61 *	  This would also greatly simply some upper layer error handlers. --AK
  62 *
  63 */
  64
  65#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  66
  67#include <linux/module.h>
  68#include <linux/types.h>
  69#include <linux/jiffies.h>
  70#include <linux/kernel.h>
  71#include <linux/fcntl.h>
  72#include <linux/socket.h>
  73#include <linux/in.h>
  74#include <linux/inet.h>
  75#include <linux/inetdevice.h>
  76#include <linux/netdevice.h>
  77#include <linux/string.h>
  78#include <linux/netfilter_ipv4.h>
  79#include <linux/slab.h>
  80#include <net/snmp.h>
  81#include <net/ip.h>
  82#include <net/route.h>
  83#include <net/protocol.h>
  84#include <net/icmp.h>
  85#include <net/tcp.h>
  86#include <net/udp.h>
  87#include <net/raw.h>
  88#include <net/ping.h>
  89#include <linux/skbuff.h>
  90#include <net/sock.h>
  91#include <linux/errno.h>
  92#include <linux/timer.h>
  93#include <linux/init.h>
  94#include <linux/uaccess.h>
  95#include <net/checksum.h>
  96#include <net/xfrm.h>
  97#include <net/inet_common.h>
  98#include <net/ip_fib.h>
  99#include <net/l3mdev.h>
 100
 101/*
 102 *	Build xmit assembly blocks
 103 */
 104
 105struct icmp_bxm {
 106	struct sk_buff *skb;
 107	int offset;
 108	int data_len;
 109
 110	struct {
 111		struct icmphdr icmph;
 112		__be32	       times[3];
 113	} data;
 114	int head_len;
 115	struct ip_options_data replyopts;
 116};
 117
 118/* An array of errno for error messages from dest unreach. */
 119/* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
 120
 121const struct icmp_err icmp_err_convert[] = {
 122	{
 123		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
 124		.fatal = 0,
 125	},
 126	{
 127		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
 128		.fatal = 0,
 129	},
 130	{
 131		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
 132		.fatal = 1,
 133	},
 134	{
 135		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
 136		.fatal = 1,
 137	},
 138	{
 139		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
 140		.fatal = 0,
 141	},
 142	{
 143		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
 144		.fatal = 0,
 145	},
 146	{
 147		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
 148		.fatal = 1,
 149	},
 150	{
 151		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
 152		.fatal = 1,
 153	},
 154	{
 155		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
 156		.fatal = 1,
 157	},
 158	{
 159		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
 160		.fatal = 1,
 161	},
 162	{
 163		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
 164		.fatal = 1,
 165	},
 166	{
 167		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
 168		.fatal = 0,
 169	},
 170	{
 171		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
 172		.fatal = 0,
 173	},
 174	{
 175		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
 176		.fatal = 1,
 177	},
 178	{
 179		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
 180		.fatal = 1,
 181	},
 182	{
 183		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
 184		.fatal = 1,
 185	},
 186};
 187EXPORT_SYMBOL(icmp_err_convert);
 188
 189/*
 190 *	ICMP control array. This specifies what to do with each ICMP.
 191 */
 192
 193struct icmp_control {
 194	bool (*handler)(struct sk_buff *skb);
 195	short   error;		/* This ICMP is classed as an error message */
 196};
 197
 198static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 199
 200/*
 201 *	The ICMP socket(s). This is the most convenient way to flow control
 202 *	our ICMP output as well as maintain a clean interface throughout
 203 *	all layers. All Socketless IP sends will soon be gone.
 204 *
 205 *	On SMP we have one ICMP socket per-cpu.
 206 */
 207static struct sock *icmp_sk(struct net *net)
 208{
 209	return *this_cpu_ptr(net->ipv4.icmp_sk);
 210}
 211
 212/* Called with BH disabled */
 213static inline struct sock *icmp_xmit_lock(struct net *net)
 214{
 215	struct sock *sk;
 216
 217	sk = icmp_sk(net);
 218
 219	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 220		/* This can happen if the output path signals a
 221		 * dst_link_failure() for an outgoing ICMP packet.
 222		 */
 223		return NULL;
 224	}
 225	return sk;
 226}
 227
 228static inline void icmp_xmit_unlock(struct sock *sk)
 229{
 230	spin_unlock(&sk->sk_lock.slock);
 231}
 232
 233int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
 234int sysctl_icmp_msgs_burst __read_mostly = 50;
 235
 236static struct {
 237	spinlock_t	lock;
 238	u32		credit;
 239	u32		stamp;
 240} icmp_global = {
 241	.lock		= __SPIN_LOCK_UNLOCKED(icmp_global.lock),
 242};
 243
 244/**
 245 * icmp_global_allow - Are we allowed to send one more ICMP message ?
 246 *
 247 * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec.
 248 * Returns false if we reached the limit and can not send another packet.
 249 * Note: called with BH disabled
 250 */
 251bool icmp_global_allow(void)
 252{
 253	u32 credit, delta, incr = 0, now = (u32)jiffies;
 254	bool rc = false;
 255
 256	/* Check if token bucket is empty and cannot be refilled
 257	 * without taking the spinlock.
 258	 */
 259	if (!icmp_global.credit) {
 260		delta = min_t(u32, now - icmp_global.stamp, HZ);
 261		if (delta < HZ / 50)
 262			return false;
 263	}
 264
 265	spin_lock(&icmp_global.lock);
 266	delta = min_t(u32, now - icmp_global.stamp, HZ);
 267	if (delta >= HZ / 50) {
 268		incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
 269		if (incr)
 270			icmp_global.stamp = now;
 271	}
 272	credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
 273	if (credit) {
 274		credit--;
 275		rc = true;
 276	}
 277	icmp_global.credit = credit;
 278	spin_unlock(&icmp_global.lock);
 279	return rc;
 280}
 281EXPORT_SYMBOL(icmp_global_allow);
 282
 283static bool icmpv4_mask_allow(struct net *net, int type, int code)
 284{
 285	if (type > NR_ICMP_TYPES)
 286		return true;
 287
 288	/* Don't limit PMTU discovery. */
 289	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 290		return true;
 291
 292	/* Limit if icmp type is enabled in ratemask. */
 293	if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
 294		return true;
 295
 296	return false;
 297}
 298
 299static bool icmpv4_global_allow(struct net *net, int type, int code)
 300{
 301	if (icmpv4_mask_allow(net, type, code))
 302		return true;
 303
 304	if (icmp_global_allow())
 305		return true;
 306
 307	return false;
 308}
 309
 310/*
 311 *	Send an ICMP frame.
 312 */
 313
 314static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 315			       struct flowi4 *fl4, int type, int code)
 316{
 317	struct dst_entry *dst = &rt->dst;
 318	struct inet_peer *peer;
 319	bool rc = true;
 320	int vif;
 321
 322	if (icmpv4_mask_allow(net, type, code))
 323		goto out;
 324
 325	/* No rate limit on loopback */
 326	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 327		goto out;
 328
 329	vif = l3mdev_master_ifindex(dst->dev);
 330	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
 331	rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
 332	if (peer)
 333		inet_putpeer(peer);
 334out:
 335	return rc;
 336}
 337
 338/*
 339 *	Maintain the counters used in the SNMP statistics for outgoing ICMP
 340 */
 341void icmp_out_count(struct net *net, unsigned char type)
 342{
 343	ICMPMSGOUT_INC_STATS(net, type);
 344	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 345}
 346
 347/*
 348 *	Checksum each fragment, and on the first include the headers and final
 349 *	checksum.
 350 */
 351static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 352			  struct sk_buff *skb)
 353{
 354	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
 355	__wsum csum;
 356
 357	csum = skb_copy_and_csum_bits(icmp_param->skb,
 358				      icmp_param->offset + offset,
 359				      to, len, 0);
 360
 361	skb->csum = csum_block_add(skb->csum, csum, odd);
 362	if (icmp_pointers[icmp_param->data.icmph.type].error)
 363		nf_ct_attach(skb, icmp_param->skb);
 364	return 0;
 365}
 366
 367static void icmp_push_reply(struct icmp_bxm *icmp_param,
 368			    struct flowi4 *fl4,
 369			    struct ipcm_cookie *ipc, struct rtable **rt)
 370{
 371	struct sock *sk;
 372	struct sk_buff *skb;
 373
 374	sk = icmp_sk(dev_net((*rt)->dst.dev));
 375	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
 376			   icmp_param->data_len+icmp_param->head_len,
 377			   icmp_param->head_len,
 378			   ipc, rt, MSG_DONTWAIT) < 0) {
 379		__ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
 380		ip_flush_pending_frames(sk);
 381	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 382		struct icmphdr *icmph = icmp_hdr(skb);
 383		__wsum csum = 0;
 384		struct sk_buff *skb1;
 385
 386		skb_queue_walk(&sk->sk_write_queue, skb1) {
 387			csum = csum_add(csum, skb1->csum);
 388		}
 389		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 390						 (char *)icmph,
 391						 icmp_param->head_len, csum);
 392		icmph->checksum = csum_fold(csum);
 393		skb->ip_summed = CHECKSUM_NONE;
 394		ip_push_pending_frames(sk, fl4);
 395	}
 396}
 397
 398/*
 399 *	Driving logic for building and sending ICMP messages.
 400 */
 401
 402static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 403{
 404	struct ipcm_cookie ipc;
 405	struct rtable *rt = skb_rtable(skb);
 406	struct net *net = dev_net(rt->dst.dev);
 407	struct flowi4 fl4;
 408	struct sock *sk;
 409	struct inet_sock *inet;
 410	__be32 daddr, saddr;
 411	u32 mark = IP4_REPLY_MARK(net, skb->mark);
 412	int type = icmp_param->data.icmph.type;
 413	int code = icmp_param->data.icmph.code;
 414
 415	if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
 416		return;
 417
 418	/* Needed by both icmp_global_allow and icmp_xmit_lock */
 419	local_bh_disable();
 420
 421	/* global icmp_msgs_per_sec */
 422	if (!icmpv4_global_allow(net, type, code))
 423		goto out_bh_enable;
 424
 425	sk = icmp_xmit_lock(net);
 426	if (!sk)
 427		goto out_bh_enable;
 428	inet = inet_sk(sk);
 429
 430	icmp_param->data.icmph.checksum = 0;
 431
 
 432	inet->tos = ip_hdr(skb)->tos;
 433	sk->sk_mark = mark;
 434	daddr = ipc.addr = ip_hdr(skb)->saddr;
 435	saddr = fib_compute_spec_dst(skb);
 436	ipc.opt = NULL;
 437	ipc.tx_flags = 0;
 438	ipc.ttl = 0;
 439	ipc.tos = -1;
 440
 441	if (icmp_param->replyopts.opt.opt.optlen) {
 442		ipc.opt = &icmp_param->replyopts.opt;
 443		if (ipc.opt->opt.srr)
 444			daddr = icmp_param->replyopts.opt.opt.faddr;
 445	}
 446	memset(&fl4, 0, sizeof(fl4));
 447	fl4.daddr = daddr;
 448	fl4.saddr = saddr;
 449	fl4.flowi4_mark = mark;
 450	fl4.flowi4_uid = sock_net_uid(net, NULL);
 451	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 452	fl4.flowi4_proto = IPPROTO_ICMP;
 453	fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
 454	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 455	rt = ip_route_output_key(net, &fl4);
 456	if (IS_ERR(rt))
 457		goto out_unlock;
 458	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 459		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
 460	ip_rt_put(rt);
 461out_unlock:
 462	icmp_xmit_unlock(sk);
 463out_bh_enable:
 464	local_bh_enable();
 465}
 466
 467static struct rtable *icmp_route_lookup(struct net *net,
 468					struct flowi4 *fl4,
 469					struct sk_buff *skb_in,
 470					const struct iphdr *iph,
 471					__be32 saddr, u8 tos, u32 mark,
 472					int type, int code,
 473					struct icmp_bxm *param)
 474{
 475	struct rtable *rt, *rt2;
 476	struct flowi4 fl4_dec;
 477	int err;
 478
 479	memset(fl4, 0, sizeof(*fl4));
 480	fl4->daddr = (param->replyopts.opt.opt.srr ?
 481		      param->replyopts.opt.opt.faddr : iph->saddr);
 482	fl4->saddr = saddr;
 483	fl4->flowi4_mark = mark;
 484	fl4->flowi4_uid = sock_net_uid(net, NULL);
 485	fl4->flowi4_tos = RT_TOS(tos);
 486	fl4->flowi4_proto = IPPROTO_ICMP;
 487	fl4->fl4_icmp_type = type;
 488	fl4->fl4_icmp_code = code;
 489	fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
 490
 491	security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
 492	rt = ip_route_output_key_hash(net, fl4, skb_in);
 493	if (IS_ERR(rt))
 494		return rt;
 495
 496	/* No need to clone since we're just using its address. */
 497	rt2 = rt;
 498
 499	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 500					   flowi4_to_flowi(fl4), NULL, 0);
 501	if (!IS_ERR(rt)) {
 502		if (rt != rt2)
 503			return rt;
 504	} else if (PTR_ERR(rt) == -EPERM) {
 505		rt = NULL;
 506	} else
 507		return rt;
 508
 509	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 510	if (err)
 511		goto relookup_failed;
 512
 513	if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
 514				     fl4_dec.saddr) == RTN_LOCAL) {
 515		rt2 = __ip_route_output_key(net, &fl4_dec);
 516		if (IS_ERR(rt2))
 517			err = PTR_ERR(rt2);
 518	} else {
 519		struct flowi4 fl4_2 = {};
 520		unsigned long orefdst;
 521
 522		fl4_2.daddr = fl4_dec.saddr;
 523		rt2 = ip_route_output_key(net, &fl4_2);
 524		if (IS_ERR(rt2)) {
 525			err = PTR_ERR(rt2);
 526			goto relookup_failed;
 527		}
 528		/* Ugh! */
 529		orefdst = skb_in->_skb_refdst; /* save old refdst */
 530		skb_dst_set(skb_in, NULL);
 531		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 532				     RT_TOS(tos), rt2->dst.dev);
 533
 534		dst_release(&rt2->dst);
 535		rt2 = skb_rtable(skb_in);
 536		skb_in->_skb_refdst = orefdst; /* restore old refdst */
 537	}
 538
 539	if (err)
 540		goto relookup_failed;
 541
 542	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 543					    flowi4_to_flowi(&fl4_dec), NULL,
 544					    XFRM_LOOKUP_ICMP);
 545	if (!IS_ERR(rt2)) {
 546		dst_release(&rt->dst);
 547		memcpy(fl4, &fl4_dec, sizeof(*fl4));
 548		rt = rt2;
 549	} else if (PTR_ERR(rt2) == -EPERM) {
 550		if (rt)
 551			dst_release(&rt->dst);
 552		return rt2;
 553	} else {
 554		err = PTR_ERR(rt2);
 555		goto relookup_failed;
 556	}
 557	return rt;
 558
 559relookup_failed:
 560	if (rt)
 561		return rt;
 562	return ERR_PTR(err);
 563}
 564
 565/*
 566 *	Send an ICMP message in response to a situation
 567 *
 568 *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
 569 *		  MAY send more (we do).
 570 *			MUST NOT change this header information.
 571 *			MUST NOT reply to a multicast/broadcast IP address.
 572 *			MUST NOT reply to a multicast/broadcast MAC address.
 573 *			MUST reply to only the first fragment.
 574 */
 575
 576void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 
 577{
 578	struct iphdr *iph;
 579	int room;
 580	struct icmp_bxm icmp_param;
 581	struct rtable *rt = skb_rtable(skb_in);
 582	struct ipcm_cookie ipc;
 583	struct flowi4 fl4;
 584	__be32 saddr;
 585	u8  tos;
 586	u32 mark;
 587	struct net *net;
 588	struct sock *sk;
 589
 590	if (!rt)
 591		goto out;
 592	net = dev_net(rt->dst.dev);
 
 
 
 
 
 
 593
 594	/*
 595	 *	Find the original header. It is expected to be valid, of course.
 596	 *	Check this, icmp_send is called from the most obscure devices
 597	 *	sometimes.
 598	 */
 599	iph = ip_hdr(skb_in);
 600
 601	if ((u8 *)iph < skb_in->head ||
 602	    (skb_network_header(skb_in) + sizeof(*iph)) >
 603	    skb_tail_pointer(skb_in))
 604		goto out;
 605
 606	/*
 607	 *	No replies to physical multicast/broadcast
 608	 */
 609	if (skb_in->pkt_type != PACKET_HOST)
 610		goto out;
 611
 612	/*
 613	 *	Now check at the protocol level
 614	 */
 615	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 616		goto out;
 617
 618	/*
 619	 *	Only reply to fragment 0. We byte re-order the constant
 620	 *	mask for efficiency.
 621	 */
 622	if (iph->frag_off & htons(IP_OFFSET))
 623		goto out;
 624
 625	/*
 626	 *	If we send an ICMP error to an ICMP error a mess would result..
 627	 */
 628	if (icmp_pointers[type].error) {
 629		/*
 630		 *	We are an error, check if we are replying to an
 631		 *	ICMP error
 632		 */
 633		if (iph->protocol == IPPROTO_ICMP) {
 634			u8 _inner_type, *itp;
 635
 636			itp = skb_header_pointer(skb_in,
 637						 skb_network_header(skb_in) +
 638						 (iph->ihl << 2) +
 639						 offsetof(struct icmphdr,
 640							  type) -
 641						 skb_in->data,
 642						 sizeof(_inner_type),
 643						 &_inner_type);
 644			if (!itp)
 645				goto out;
 646
 647			/*
 648			 *	Assume any unknown ICMP type is an error. This
 649			 *	isn't specified by the RFC, but think about it..
 650			 */
 651			if (*itp > NR_ICMP_TYPES ||
 652			    icmp_pointers[*itp].error)
 653				goto out;
 654		}
 655	}
 656
 657	/* Needed by both icmp_global_allow and icmp_xmit_lock */
 658	local_bh_disable();
 659
 660	/* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
 661	 * incoming dev is loopback.  If outgoing dev change to not be
 662	 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
 663	 */
 664	if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
 665	      !icmpv4_global_allow(net, type, code))
 666		goto out_bh_enable;
 667
 668	sk = icmp_xmit_lock(net);
 669	if (!sk)
 670		goto out_bh_enable;
 671
 672	/*
 673	 *	Construct source address and options.
 674	 */
 675
 676	saddr = iph->daddr;
 677	if (!(rt->rt_flags & RTCF_LOCAL)) {
 678		struct net_device *dev = NULL;
 679
 680		rcu_read_lock();
 681		if (rt_is_input_route(rt) &&
 682		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
 683			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 684
 685		if (dev)
 686			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
 687		else
 688			saddr = 0;
 689		rcu_read_unlock();
 690	}
 691
 692	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
 693					   IPTOS_PREC_INTERNETCONTROL) :
 694					  iph->tos;
 695	mark = IP4_REPLY_MARK(net, skb_in->mark);
 696
 697	if (ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in))
 698		goto out_unlock;
 699
 700
 701	/*
 702	 *	Prepare data for ICMP header.
 703	 */
 704
 705	icmp_param.data.icmph.type	 = type;
 706	icmp_param.data.icmph.code	 = code;
 707	icmp_param.data.icmph.un.gateway = info;
 708	icmp_param.data.icmph.checksum	 = 0;
 709	icmp_param.skb	  = skb_in;
 710	icmp_param.offset = skb_network_offset(skb_in);
 711	inet_sk(sk)->tos = tos;
 712	sk->sk_mark = mark;
 
 713	ipc.addr = iph->saddr;
 714	ipc.opt = &icmp_param.replyopts.opt;
 715	ipc.tx_flags = 0;
 716	ipc.ttl = 0;
 717	ipc.tos = -1;
 718
 719	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
 720			       type, code, &icmp_param);
 721	if (IS_ERR(rt))
 722		goto out_unlock;
 723
 724	/* peer icmp_ratelimit */
 725	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 726		goto ende;
 727
 728	/* RFC says return as much as we can without exceeding 576 bytes. */
 729
 730	room = dst_mtu(&rt->dst);
 731	if (room > 576)
 732		room = 576;
 733	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 734	room -= sizeof(struct icmphdr);
 735
 736	icmp_param.data_len = skb_in->len - icmp_param.offset;
 737	if (icmp_param.data_len > room)
 738		icmp_param.data_len = room;
 739	icmp_param.head_len = sizeof(struct icmphdr);
 740
 741	icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
 742ende:
 743	ip_rt_put(rt);
 744out_unlock:
 745	icmp_xmit_unlock(sk);
 746out_bh_enable:
 747	local_bh_enable();
 748out:;
 749}
 750EXPORT_SYMBOL(icmp_send);
 751
 752
 753static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 754{
 755	const struct iphdr *iph = (const struct iphdr *) skb->data;
 756	const struct net_protocol *ipprot;
 757	int protocol = iph->protocol;
 758
 759	/* Checkin full IP header plus 8 bytes of protocol to
 760	 * avoid additional coding at protocol handlers.
 761	 */
 762	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 763		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 764		return;
 765	}
 766
 767	raw_icmp_error(skb, protocol, info);
 768
 769	ipprot = rcu_dereference(inet_protos[protocol]);
 770	if (ipprot && ipprot->err_handler)
 771		ipprot->err_handler(skb, info);
 772}
 773
 774static bool icmp_tag_validation(int proto)
 775{
 776	bool ok;
 777
 778	rcu_read_lock();
 779	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 780	rcu_read_unlock();
 781	return ok;
 782}
 783
 784/*
 785 *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 786 *	ICMP_PARAMETERPROB.
 787 */
 788
 789static bool icmp_unreach(struct sk_buff *skb)
 790{
 791	const struct iphdr *iph;
 792	struct icmphdr *icmph;
 793	struct net *net;
 794	u32 info = 0;
 795
 796	net = dev_net(skb_dst(skb)->dev);
 797
 798	/*
 799	 *	Incomplete header ?
 800	 * 	Only checks for the IP header, there should be an
 801	 *	additional check for longer headers in upper levels.
 802	 */
 803
 804	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 805		goto out_err;
 806
 807	icmph = icmp_hdr(skb);
 808	iph   = (const struct iphdr *)skb->data;
 809
 810	if (iph->ihl < 5) /* Mangled header, drop. */
 811		goto out_err;
 812
 813	switch (icmph->type) {
 814	case ICMP_DEST_UNREACH:
 815		switch (icmph->code & 15) {
 816		case ICMP_NET_UNREACH:
 817		case ICMP_HOST_UNREACH:
 818		case ICMP_PROT_UNREACH:
 819		case ICMP_PORT_UNREACH:
 820			break;
 821		case ICMP_FRAG_NEEDED:
 822			/* for documentation of the ip_no_pmtu_disc
 823			 * values please see
 824			 * Documentation/networking/ip-sysctl.txt
 825			 */
 826			switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
 827			default:
 828				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 829						    &iph->daddr);
 830				break;
 831			case 2:
 832				goto out;
 833			case 3:
 834				if (!icmp_tag_validation(iph->protocol))
 835					goto out;
 836				/* fall through */
 837			case 0:
 838				info = ntohs(icmph->un.frag.mtu);
 839			}
 840			break;
 841		case ICMP_SR_FAILED:
 842			net_dbg_ratelimited("%pI4: Source Route Failed\n",
 843					    &iph->daddr);
 844			break;
 845		default:
 846			break;
 847		}
 848		if (icmph->code > NR_ICMP_UNREACH)
 849			goto out;
 850		break;
 851	case ICMP_PARAMETERPROB:
 852		info = ntohl(icmph->un.gateway) >> 24;
 853		break;
 854	case ICMP_TIME_EXCEEDED:
 855		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 856		if (icmph->code == ICMP_EXC_FRAGTIME)
 857			goto out;
 858		break;
 859	}
 860
 861	/*
 862	 *	Throw it at our lower layers
 863	 *
 864	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 865	 *		  header.
 866	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 867	 *		  transport layer.
 868	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 869	 *		  transport layer.
 870	 */
 871
 872	/*
 873	 *	Check the other end isn't violating RFC 1122. Some routers send
 874	 *	bogus responses to broadcast frames. If you see this message
 875	 *	first check your netmask matches at both ends, if it does then
 876	 *	get the other vendor to fix their kit.
 877	 */
 878
 879	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
 880	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 881		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 882				     &ip_hdr(skb)->saddr,
 883				     icmph->type, icmph->code,
 884				     &iph->daddr, skb->dev->name);
 885		goto out;
 886	}
 887
 888	icmp_socket_deliver(skb, info);
 889
 890out:
 891	return true;
 892out_err:
 893	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 894	return false;
 895}
 896
 897
 898/*
 899 *	Handle ICMP_REDIRECT.
 900 */
 901
 902static bool icmp_redirect(struct sk_buff *skb)
 903{
 904	if (skb->len < sizeof(struct iphdr)) {
 905		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 906		return false;
 907	}
 908
 909	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 910		/* there aught to be a stat */
 911		return false;
 912	}
 913
 914	icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
 915	return true;
 916}
 917
 918/*
 919 *	Handle ICMP_ECHO ("ping") requests.
 920 *
 921 *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 922 *		  requests.
 923 *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 924 *		  included in the reply.
 925 *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 926 *		  echo requests, MUST have default=NOT.
 927 *	See also WRT handling of options once they are done and working.
 928 */
 929
 930static bool icmp_echo(struct sk_buff *skb)
 931{
 932	struct net *net;
 933
 934	net = dev_net(skb_dst(skb)->dev);
 935	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
 936		struct icmp_bxm icmp_param;
 937
 938		icmp_param.data.icmph	   = *icmp_hdr(skb);
 939		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
 940		icmp_param.skb		   = skb;
 941		icmp_param.offset	   = 0;
 942		icmp_param.data_len	   = skb->len;
 943		icmp_param.head_len	   = sizeof(struct icmphdr);
 944		icmp_reply(&icmp_param, skb);
 945	}
 946	/* should there be an ICMP stat for ignored echos? */
 947	return true;
 948}
 949
 950/*
 951 *	Handle ICMP Timestamp requests.
 952 *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
 953 *		  SHOULD be in the kernel for minimum random latency.
 954 *		  MUST be accurate to a few minutes.
 955 *		  MUST be updated at least at 15Hz.
 956 */
 957static bool icmp_timestamp(struct sk_buff *skb)
 958{
 959	struct icmp_bxm icmp_param;
 960	/*
 961	 *	Too short.
 962	 */
 963	if (skb->len < 4)
 964		goto out_err;
 965
 966	/*
 967	 *	Fill in the current time as ms since midnight UT:
 968	 */
 969	icmp_param.data.times[1] = inet_current_timestamp();
 970	icmp_param.data.times[2] = icmp_param.data.times[1];
 971
 972	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
 973
 974	icmp_param.data.icmph	   = *icmp_hdr(skb);
 975	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
 976	icmp_param.data.icmph.code = 0;
 977	icmp_param.skb		   = skb;
 978	icmp_param.offset	   = 0;
 979	icmp_param.data_len	   = 0;
 980	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
 981	icmp_reply(&icmp_param, skb);
 982	return true;
 983
 984out_err:
 985	__ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
 986	return false;
 987}
 988
 989static bool icmp_discard(struct sk_buff *skb)
 990{
 991	/* pretend it was a success */
 992	return true;
 993}
 994
 995/*
 996 *	Deal with incoming ICMP packets.
 997 */
 998int icmp_rcv(struct sk_buff *skb)
 999{
1000	struct icmphdr *icmph;
1001	struct rtable *rt = skb_rtable(skb);
1002	struct net *net = dev_net(rt->dst.dev);
1003	bool success;
1004
1005	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1006		struct sec_path *sp = skb_sec_path(skb);
1007		int nh;
1008
1009		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1010				 XFRM_STATE_ICMP))
1011			goto drop;
1012
1013		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1014			goto drop;
1015
1016		nh = skb_network_offset(skb);
1017		skb_set_network_header(skb, sizeof(*icmph));
1018
1019		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1020			goto drop;
1021
1022		skb_set_network_header(skb, nh);
1023	}
1024
1025	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1026
1027	if (skb_checksum_simple_validate(skb))
1028		goto csum_error;
1029
1030	if (!pskb_pull(skb, sizeof(*icmph)))
1031		goto error;
1032
1033	icmph = icmp_hdr(skb);
1034
1035	ICMPMSGIN_INC_STATS(net, icmph->type);
1036	/*
1037	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1038	 *
1039	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1040	 *		  discarded.
1041	 */
1042	if (icmph->type > NR_ICMP_TYPES)
1043		goto error;
1044
1045
1046	/*
1047	 *	Parse the ICMP message
1048	 */
1049
1050	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1051		/*
1052		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1053		 *	  silently ignored (we let user decide with a sysctl).
1054		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1055		 *	  discarded if to broadcast/multicast.
1056		 */
1057		if ((icmph->type == ICMP_ECHO ||
1058		     icmph->type == ICMP_TIMESTAMP) &&
1059		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1060			goto error;
1061		}
1062		if (icmph->type != ICMP_ECHO &&
1063		    icmph->type != ICMP_TIMESTAMP &&
1064		    icmph->type != ICMP_ADDRESS &&
1065		    icmph->type != ICMP_ADDRESSREPLY) {
1066			goto error;
1067		}
1068	}
1069
1070	success = icmp_pointers[icmph->type].handler(skb);
1071
1072	if (success)  {
1073		consume_skb(skb);
1074		return NET_RX_SUCCESS;
1075	}
1076
1077drop:
1078	kfree_skb(skb);
1079	return NET_RX_DROP;
1080csum_error:
1081	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1082error:
1083	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1084	goto drop;
1085}
1086
1087void icmp_err(struct sk_buff *skb, u32 info)
1088{
1089	struct iphdr *iph = (struct iphdr *)skb->data;
1090	int offset = iph->ihl<<2;
1091	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1092	int type = icmp_hdr(skb)->type;
1093	int code = icmp_hdr(skb)->code;
1094	struct net *net = dev_net(skb->dev);
1095
1096	/*
1097	 * Use ping_err to handle all icmp errors except those
1098	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1099	 */
1100	if (icmph->type != ICMP_ECHOREPLY) {
1101		ping_err(skb, offset, info);
1102		return;
1103	}
1104
1105	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1106		ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1107	else if (type == ICMP_REDIRECT)
1108		ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
 
 
1109}
1110
1111/*
1112 *	This table is the definition of how we handle ICMP.
1113 */
1114static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1115	[ICMP_ECHOREPLY] = {
1116		.handler = ping_rcv,
1117	},
1118	[1] = {
1119		.handler = icmp_discard,
1120		.error = 1,
1121	},
1122	[2] = {
1123		.handler = icmp_discard,
1124		.error = 1,
1125	},
1126	[ICMP_DEST_UNREACH] = {
1127		.handler = icmp_unreach,
1128		.error = 1,
1129	},
1130	[ICMP_SOURCE_QUENCH] = {
1131		.handler = icmp_unreach,
1132		.error = 1,
1133	},
1134	[ICMP_REDIRECT] = {
1135		.handler = icmp_redirect,
1136		.error = 1,
1137	},
1138	[6] = {
1139		.handler = icmp_discard,
1140		.error = 1,
1141	},
1142	[7] = {
1143		.handler = icmp_discard,
1144		.error = 1,
1145	},
1146	[ICMP_ECHO] = {
1147		.handler = icmp_echo,
1148	},
1149	[9] = {
1150		.handler = icmp_discard,
1151		.error = 1,
1152	},
1153	[10] = {
1154		.handler = icmp_discard,
1155		.error = 1,
1156	},
1157	[ICMP_TIME_EXCEEDED] = {
1158		.handler = icmp_unreach,
1159		.error = 1,
1160	},
1161	[ICMP_PARAMETERPROB] = {
1162		.handler = icmp_unreach,
1163		.error = 1,
1164	},
1165	[ICMP_TIMESTAMP] = {
1166		.handler = icmp_timestamp,
1167	},
1168	[ICMP_TIMESTAMPREPLY] = {
1169		.handler = icmp_discard,
1170	},
1171	[ICMP_INFO_REQUEST] = {
1172		.handler = icmp_discard,
1173	},
1174	[ICMP_INFO_REPLY] = {
1175		.handler = icmp_discard,
1176	},
1177	[ICMP_ADDRESS] = {
1178		.handler = icmp_discard,
1179	},
1180	[ICMP_ADDRESSREPLY] = {
1181		.handler = icmp_discard,
1182	},
1183};
1184
1185static void __net_exit icmp_sk_exit(struct net *net)
1186{
1187	int i;
1188
1189	for_each_possible_cpu(i)
1190		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1191	free_percpu(net->ipv4.icmp_sk);
1192	net->ipv4.icmp_sk = NULL;
1193}
1194
1195static int __net_init icmp_sk_init(struct net *net)
1196{
1197	int i, err;
1198
1199	net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1200	if (!net->ipv4.icmp_sk)
1201		return -ENOMEM;
1202
1203	for_each_possible_cpu(i) {
1204		struct sock *sk;
1205
1206		err = inet_ctl_sock_create(&sk, PF_INET,
1207					   SOCK_RAW, IPPROTO_ICMP, net);
1208		if (err < 0)
1209			goto fail;
1210
1211		*per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1212
1213		/* Enough space for 2 64K ICMP packets, including
1214		 * sk_buff/skb_shared_info struct overhead.
1215		 */
1216		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1217
1218		/*
1219		 * Speedup sock_wfree()
1220		 */
1221		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1222		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1223	}
1224
1225	/* Control parameters for ECHO replies. */
1226	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1227	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1228
1229	/* Control parameter - ignore bogus broadcast responses? */
1230	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1231
1232	/*
1233	 * 	Configurable global rate limit.
1234	 *
1235	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1236	 *	bucket ratemask defines which icmp types are ratelimited by
1237	 *	setting	it's bit position.
1238	 *
1239	 *	default:
1240	 *	dest unreachable (3), source quench (4),
1241	 *	time exceeded (11), parameter problem (12)
1242	 */
1243
1244	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1245	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1246	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1247
1248	return 0;
1249
1250fail:
1251	for_each_possible_cpu(i)
1252		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1253	free_percpu(net->ipv4.icmp_sk);
1254	return err;
1255}
1256
1257static struct pernet_operations __net_initdata icmp_sk_ops = {
1258       .init = icmp_sk_init,
1259       .exit = icmp_sk_exit,
1260};
1261
1262int __init icmp_init(void)
1263{
1264	return register_pernet_subsys(&icmp_sk_ops);
1265}