Linux Audio

Check our new training course

Loading...
v6.13.7
   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#include <net/addrconf.h>
  96#include <net/inet_dscp.h>
  97#define CREATE_TRACE_POINTS
  98#include <trace/events/icmp.h>
  99
 100/*
 101 *	Build xmit assembly blocks
 102 */
 103
 104struct icmp_bxm {
 105	struct sk_buff *skb;
 106	int offset;
 107	int data_len;
 108
 109	struct {
 110		struct icmphdr icmph;
 111		__be32	       times[3];
 112	} data;
 113	int head_len;
 114	struct ip_options_data replyopts;
 115};
 116
 117/* An array of errno for error messages from dest unreach. */
 118/* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
 119
 120const struct icmp_err icmp_err_convert[] = {
 121	{
 122		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
 123		.fatal = 0,
 124	},
 125	{
 126		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
 127		.fatal = 0,
 128	},
 129	{
 130		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
 131		.fatal = 1,
 132	},
 133	{
 134		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
 135		.fatal = 1,
 136	},
 137	{
 138		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
 139		.fatal = 0,
 140	},
 141	{
 142		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
 143		.fatal = 0,
 144	},
 145	{
 146		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
 147		.fatal = 1,
 148	},
 149	{
 150		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
 151		.fatal = 1,
 152	},
 153	{
 154		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
 155		.fatal = 1,
 156	},
 157	{
 158		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
 159		.fatal = 1,
 160	},
 161	{
 162		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
 163		.fatal = 1,
 164	},
 165	{
 166		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
 167		.fatal = 0,
 168	},
 169	{
 170		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
 171		.fatal = 0,
 172	},
 173	{
 174		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
 175		.fatal = 1,
 176	},
 177	{
 178		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
 179		.fatal = 1,
 180	},
 181	{
 182		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
 183		.fatal = 1,
 184	},
 185};
 186EXPORT_SYMBOL(icmp_err_convert);
 187
 188/*
 189 *	ICMP control array. This specifies what to do with each ICMP.
 190 */
 191
 192struct icmp_control {
 193	enum skb_drop_reason (*handler)(struct sk_buff *skb);
 194	short   error;		/* This ICMP is classed as an error message */
 195};
 196
 197static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 198
 199static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
 200
 201/* Called with BH disabled */
 202static inline struct sock *icmp_xmit_lock(struct net *net)
 203{
 204	struct sock *sk;
 205
 206	sk = this_cpu_read(ipv4_icmp_sk);
 207
 208	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 209		/* This can happen if the output path signals a
 210		 * dst_link_failure() for an outgoing ICMP packet.
 211		 */
 212		return NULL;
 213	}
 214	sock_net_set(sk, net);
 215	return sk;
 216}
 217
 218static inline void icmp_xmit_unlock(struct sock *sk)
 219{
 220	sock_net_set(sk, &init_net);
 221	spin_unlock(&sk->sk_lock.slock);
 222}
 223
 
 
 
 
 
 
 
 
 
 
 
 224/**
 225 * icmp_global_allow - Are we allowed to send one more ICMP message ?
 226 * @net: network namespace
 227 *
 228 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
 229 * Returns false if we reached the limit and can not send another packet.
 230 * Works in tandem with icmp_global_consume().
 231 */
 232bool icmp_global_allow(struct net *net)
 233{
 234	u32 delta, now, oldstamp;
 235	int incr, new, old;
 236
 237	/* Note: many cpus could find this condition true.
 238	 * Then later icmp_global_consume() could consume more credits,
 239	 * this is an acceptable race.
 240	 */
 241	if (atomic_read(&net->ipv4.icmp_global_credit) > 0)
 242		return true;
 243
 244	now = jiffies;
 245	oldstamp = READ_ONCE(net->ipv4.icmp_global_stamp);
 246	delta = min_t(u32, now - oldstamp, HZ);
 247	if (delta < HZ / 50)
 248		return false;
 249
 250	incr = READ_ONCE(net->ipv4.sysctl_icmp_msgs_per_sec) * delta / HZ;
 251	if (!incr)
 252		return false;
 253
 254	if (cmpxchg(&net->ipv4.icmp_global_stamp, oldstamp, now) == oldstamp) {
 255		old = atomic_read(&net->ipv4.icmp_global_credit);
 256		do {
 257			new = min(old + incr, READ_ONCE(net->ipv4.sysctl_icmp_msgs_burst));
 258		} while (!atomic_try_cmpxchg(&net->ipv4.icmp_global_credit, &old, new));
 
 
 
 
 
 
 
 
 
 
 259	}
 260	return true;
 
 
 261}
 262EXPORT_SYMBOL(icmp_global_allow);
 263
 264void icmp_global_consume(struct net *net)
 265{
 266	int credits = get_random_u32_below(3);
 267
 268	/* Note: this might make icmp_global.credit negative. */
 269	if (credits)
 270		atomic_sub(credits, &net->ipv4.icmp_global_credit);
 271}
 272EXPORT_SYMBOL(icmp_global_consume);
 273
 274static bool icmpv4_mask_allow(struct net *net, int type, int code)
 275{
 276	if (type > NR_ICMP_TYPES)
 277		return true;
 278
 279	/* Don't limit PMTU discovery. */
 280	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 281		return true;
 282
 283	/* Limit if icmp type is enabled in ratemask. */
 284	if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
 285		return true;
 286
 287	return false;
 288}
 289
 290static bool icmpv4_global_allow(struct net *net, int type, int code,
 291				bool *apply_ratelimit)
 292{
 293	if (icmpv4_mask_allow(net, type, code))
 294		return true;
 295
 296	if (icmp_global_allow(net)) {
 297		*apply_ratelimit = true;
 298		return true;
 299	}
 300	__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
 301	return false;
 302}
 303
 304/*
 305 *	Send an ICMP frame.
 306 */
 307
 308static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 309			       struct flowi4 *fl4, int type, int code,
 310			       bool apply_ratelimit)
 311{
 312	struct dst_entry *dst = &rt->dst;
 313	struct inet_peer *peer;
 314	bool rc = true;
 
 315
 316	if (!apply_ratelimit)
 317		return true;
 318
 319	/* No rate limit on loopback */
 320	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 321		goto out;
 322
 323	rcu_read_lock();
 324	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr,
 325			       l3mdev_master_ifindex_rcu(dst->dev));
 326	rc = inet_peer_xrlim_allow(peer,
 327				   READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
 328	rcu_read_unlock();
 
 329out:
 330	if (!rc)
 331		__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
 332	else
 333		icmp_global_consume(net);
 334	return rc;
 335}
 336
 337/*
 338 *	Maintain the counters used in the SNMP statistics for outgoing ICMP
 339 */
 340void icmp_out_count(struct net *net, unsigned char type)
 341{
 342	ICMPMSGOUT_INC_STATS(net, type);
 343	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 344}
 345
 346/*
 347 *	Checksum each fragment, and on the first include the headers and final
 348 *	checksum.
 349 */
 350static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 351			  struct sk_buff *skb)
 352{
 353	struct icmp_bxm *icmp_param = from;
 354	__wsum csum;
 355
 356	csum = skb_copy_and_csum_bits(icmp_param->skb,
 357				      icmp_param->offset + offset,
 358				      to, len);
 359
 360	skb->csum = csum_block_add(skb->csum, csum, odd);
 361	if (icmp_pointers[icmp_param->data.icmph.type].error)
 362		nf_ct_attach(skb, icmp_param->skb);
 363	return 0;
 364}
 365
 366static void icmp_push_reply(struct sock *sk,
 367			    struct icmp_bxm *icmp_param,
 368			    struct flowi4 *fl4,
 369			    struct ipcm_cookie *ipc, struct rtable **rt)
 370{
 371	struct sk_buff *skb;
 372
 373	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
 374			   icmp_param->data_len+icmp_param->head_len,
 375			   icmp_param->head_len,
 376			   ipc, rt, MSG_DONTWAIT) < 0) {
 377		__ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
 378		ip_flush_pending_frames(sk);
 379	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
 380		struct icmphdr *icmph = icmp_hdr(skb);
 381		__wsum csum;
 382		struct sk_buff *skb1;
 383
 384		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 385						 (char *)icmph,
 386						 icmp_param->head_len);
 387		skb_queue_walk(&sk->sk_write_queue, skb1) {
 388			csum = csum_add(csum, skb1->csum);
 389		}
 390		icmph->checksum = csum_fold(csum);
 391		skb->ip_summed = CHECKSUM_NONE;
 392		ip_push_pending_frames(sk, fl4);
 393	}
 394}
 395
 396/*
 397 *	Driving logic for building and sending ICMP messages.
 398 */
 399
 400static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 401{
 402	struct rtable *rt = skb_rtable(skb);
 403	struct net *net = dev_net_rcu(rt->dst.dev);
 404	bool apply_ratelimit = false;
 405	struct ipcm_cookie ipc;
 
 
 406	struct flowi4 fl4;
 407	struct sock *sk;
 408	struct inet_sock *inet;
 409	__be32 daddr, saddr;
 410	u32 mark = IP4_REPLY_MARK(net, skb->mark);
 411	int type = icmp_param->data.icmph.type;
 412	int code = icmp_param->data.icmph.code;
 413
 414	if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
 415		return;
 416
 417	/* Needed by both icmpv4_global_allow and icmp_xmit_lock */
 418	local_bh_disable();
 419
 420	/* is global icmp_msgs_per_sec exhausted ? */
 421	if (!icmpv4_global_allow(net, type, code, &apply_ratelimit))
 422		goto out_bh_enable;
 423
 424	sk = icmp_xmit_lock(net);
 425	if (!sk)
 426		goto out_bh_enable;
 427	inet = inet_sk(sk);
 428
 429	icmp_param->data.icmph.checksum = 0;
 430
 431	ipcm_init(&ipc);
 432	inet->tos = ip_hdr(skb)->tos;
 433	ipc.sockc.mark = mark;
 434	daddr = ipc.addr = ip_hdr(skb)->saddr;
 435	saddr = fib_compute_spec_dst(skb);
 436
 437	if (icmp_param->replyopts.opt.opt.optlen) {
 438		ipc.opt = &icmp_param->replyopts.opt;
 439		if (ipc.opt->opt.srr)
 440			daddr = icmp_param->replyopts.opt.opt.faddr;
 441	}
 442	memset(&fl4, 0, sizeof(fl4));
 443	fl4.daddr = daddr;
 444	fl4.saddr = saddr;
 445	fl4.flowi4_mark = mark;
 446	fl4.flowi4_uid = sock_net_uid(net, NULL);
 447	fl4.flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(ip_hdr(skb)));
 448	fl4.flowi4_proto = IPPROTO_ICMP;
 449	fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
 450	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
 451	rt = ip_route_output_key(net, &fl4);
 452	if (IS_ERR(rt))
 453		goto out_unlock;
 454	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
 455		icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
 456	ip_rt_put(rt);
 457out_unlock:
 458	icmp_xmit_unlock(sk);
 459out_bh_enable:
 460	local_bh_enable();
 461}
 462
 463/*
 464 * The device used for looking up which routing table to use for sending an ICMP
 465 * error is preferably the source whenever it is set, which should ensure the
 466 * icmp error can be sent to the source host, else lookup using the routing
 467 * table of the destination device, else use the main routing table (index 0).
 468 */
 469static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
 470{
 471	struct net_device *route_lookup_dev = NULL;
 472
 473	if (skb->dev)
 474		route_lookup_dev = skb->dev;
 475	else if (skb_dst(skb))
 476		route_lookup_dev = skb_dst(skb)->dev;
 477	return route_lookup_dev;
 478}
 479
 480static struct rtable *icmp_route_lookup(struct net *net, struct flowi4 *fl4,
 
 481					struct sk_buff *skb_in,
 482					const struct iphdr *iph, __be32 saddr,
 483					dscp_t dscp, u32 mark, int type,
 484					int code, struct icmp_bxm *param)
 
 485{
 486	struct net_device *route_lookup_dev;
 487	struct dst_entry *dst, *dst2;
 488	struct rtable *rt, *rt2;
 489	struct flowi4 fl4_dec;
 490	int err;
 491
 492	memset(fl4, 0, sizeof(*fl4));
 493	fl4->daddr = (param->replyopts.opt.opt.srr ?
 494		      param->replyopts.opt.opt.faddr : iph->saddr);
 495	fl4->saddr = saddr;
 496	fl4->flowi4_mark = mark;
 497	fl4->flowi4_uid = sock_net_uid(net, NULL);
 498	fl4->flowi4_tos = inet_dscp_to_dsfield(dscp);
 499	fl4->flowi4_proto = IPPROTO_ICMP;
 500	fl4->fl4_icmp_type = type;
 501	fl4->fl4_icmp_code = code;
 502	route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
 503	fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
 504
 505	security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
 506	rt = ip_route_output_key_hash(net, fl4, skb_in);
 507	if (IS_ERR(rt))
 508		return rt;
 509
 510	/* No need to clone since we're just using its address. */
 511	rt2 = rt;
 512
 513	dst = xfrm_lookup(net, &rt->dst,
 514			  flowi4_to_flowi(fl4), NULL, 0);
 515	rt = dst_rtable(dst);
 516	if (!IS_ERR(dst)) {
 517		if (rt != rt2)
 518			return rt;
 519		if (inet_addr_type_dev_table(net, route_lookup_dev,
 520					     fl4->daddr) == RTN_LOCAL)
 521			return rt;
 522	} else if (PTR_ERR(dst) == -EPERM) {
 523		rt = NULL;
 524	} else {
 525		return rt;
 526	}
 527	err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 528	if (err)
 529		goto relookup_failed;
 530
 531	if (inet_addr_type_dev_table(net, route_lookup_dev,
 532				     fl4_dec.saddr) == RTN_LOCAL) {
 533		rt2 = __ip_route_output_key(net, &fl4_dec);
 534		if (IS_ERR(rt2))
 535			err = PTR_ERR(rt2);
 536	} else {
 537		struct flowi4 fl4_2 = {};
 538		unsigned long orefdst;
 539
 540		fl4_2.daddr = fl4_dec.saddr;
 541		rt2 = ip_route_output_key(net, &fl4_2);
 542		if (IS_ERR(rt2)) {
 543			err = PTR_ERR(rt2);
 544			goto relookup_failed;
 545		}
 546		/* Ugh! */
 547		orefdst = skb_in->_skb_refdst; /* save old refdst */
 548		skb_dst_set(skb_in, NULL);
 549		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 550				     dscp, rt2->dst.dev) ? -EINVAL : 0;
 551
 552		dst_release(&rt2->dst);
 553		rt2 = skb_rtable(skb_in);
 554		skb_in->_skb_refdst = orefdst; /* restore old refdst */
 555	}
 556
 557	if (err)
 558		goto relookup_failed;
 559
 560	dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL,
 561			   XFRM_LOOKUP_ICMP);
 562	rt2 = dst_rtable(dst2);
 563	if (!IS_ERR(dst2)) {
 564		dst_release(&rt->dst);
 565		memcpy(fl4, &fl4_dec, sizeof(*fl4));
 566		rt = rt2;
 567	} else if (PTR_ERR(dst2) == -EPERM) {
 568		if (rt)
 569			dst_release(&rt->dst);
 570		return rt2;
 571	} else {
 572		err = PTR_ERR(dst2);
 573		goto relookup_failed;
 574	}
 575	return rt;
 576
 577relookup_failed:
 578	if (rt)
 579		return rt;
 580	return ERR_PTR(err);
 581}
 582
 583/*
 584 *	Send an ICMP message in response to a situation
 585 *
 586 *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
 587 *		  MAY send more (we do).
 588 *			MUST NOT change this header information.
 589 *			MUST NOT reply to a multicast/broadcast IP address.
 590 *			MUST NOT reply to a multicast/broadcast MAC address.
 591 *			MUST reply to only the first fragment.
 592 */
 593
 594void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
 595		 const struct ip_options *opt)
 596{
 597	struct iphdr *iph;
 598	int room;
 599	struct icmp_bxm icmp_param;
 600	struct rtable *rt = skb_rtable(skb_in);
 601	bool apply_ratelimit = false;
 602	struct ipcm_cookie ipc;
 603	struct flowi4 fl4;
 604	__be32 saddr;
 605	u8  tos;
 606	u32 mark;
 607	struct net *net;
 608	struct sock *sk;
 609
 610	if (!rt)
 611		return;
 612
 613	rcu_read_lock();
 614
 615	if (rt->dst.dev)
 616		net = dev_net_rcu(rt->dst.dev);
 617	else if (skb_in->dev)
 618		net = dev_net_rcu(skb_in->dev);
 619	else
 620		goto out;
 621
 622	/*
 623	 *	Find the original header. It is expected to be valid, of course.
 624	 *	Check this, icmp_send is called from the most obscure devices
 625	 *	sometimes.
 626	 */
 627	iph = ip_hdr(skb_in);
 628
 629	if ((u8 *)iph < skb_in->head ||
 630	    (skb_network_header(skb_in) + sizeof(*iph)) >
 631	    skb_tail_pointer(skb_in))
 632		goto out;
 633
 634	/*
 635	 *	No replies to physical multicast/broadcast
 636	 */
 637	if (skb_in->pkt_type != PACKET_HOST)
 638		goto out;
 639
 640	/*
 641	 *	Now check at the protocol level
 642	 */
 643	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 644		goto out;
 645
 646	/*
 647	 *	Only reply to fragment 0. We byte re-order the constant
 648	 *	mask for efficiency.
 649	 */
 650	if (iph->frag_off & htons(IP_OFFSET))
 651		goto out;
 652
 653	/*
 654	 *	If we send an ICMP error to an ICMP error a mess would result..
 655	 */
 656	if (icmp_pointers[type].error) {
 657		/*
 658		 *	We are an error, check if we are replying to an
 659		 *	ICMP error
 660		 */
 661		if (iph->protocol == IPPROTO_ICMP) {
 662			u8 _inner_type, *itp;
 663
 664			itp = skb_header_pointer(skb_in,
 665						 skb_network_header(skb_in) +
 666						 (iph->ihl << 2) +
 667						 offsetof(struct icmphdr,
 668							  type) -
 669						 skb_in->data,
 670						 sizeof(_inner_type),
 671						 &_inner_type);
 672			if (!itp)
 673				goto out;
 674
 675			/*
 676			 *	Assume any unknown ICMP type is an error. This
 677			 *	isn't specified by the RFC, but think about it..
 678			 */
 679			if (*itp > NR_ICMP_TYPES ||
 680			    icmp_pointers[*itp].error)
 681				goto out;
 682		}
 683	}
 684
 685	/* Needed by both icmpv4_global_allow and icmp_xmit_lock */
 686	local_bh_disable();
 687
 688	/* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
 689	 * incoming dev is loopback.  If outgoing dev change to not be
 690	 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
 691	 */
 692	if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
 693	      !icmpv4_global_allow(net, type, code, &apply_ratelimit))
 694		goto out_bh_enable;
 695
 696	sk = icmp_xmit_lock(net);
 697	if (!sk)
 698		goto out_bh_enable;
 699
 700	/*
 701	 *	Construct source address and options.
 702	 */
 703
 704	saddr = iph->daddr;
 705	if (!(rt->rt_flags & RTCF_LOCAL)) {
 706		struct net_device *dev = NULL;
 707
 708		rcu_read_lock();
 709		if (rt_is_input_route(rt) &&
 710		    READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
 711			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 712
 713		if (dev)
 714			saddr = inet_select_addr(dev, iph->saddr,
 715						 RT_SCOPE_LINK);
 716		else
 717			saddr = 0;
 718		rcu_read_unlock();
 719	}
 720
 721	tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
 722					   IPTOS_PREC_INTERNETCONTROL) :
 723					   iph->tos;
 724	mark = IP4_REPLY_MARK(net, skb_in->mark);
 725
 726	if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
 727		goto out_unlock;
 728
 729
 730	/*
 731	 *	Prepare data for ICMP header.
 732	 */
 733
 734	icmp_param.data.icmph.type	 = type;
 735	icmp_param.data.icmph.code	 = code;
 736	icmp_param.data.icmph.un.gateway = info;
 737	icmp_param.data.icmph.checksum	 = 0;
 738	icmp_param.skb	  = skb_in;
 739	icmp_param.offset = skb_network_offset(skb_in);
 740	inet_sk(sk)->tos = tos;
 741	ipcm_init(&ipc);
 742	ipc.addr = iph->saddr;
 743	ipc.opt = &icmp_param.replyopts.opt;
 744	ipc.sockc.mark = mark;
 745
 746	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr,
 747			       inet_dsfield_to_dscp(tos), mark, type, code,
 748			       &icmp_param);
 749	if (IS_ERR(rt))
 750		goto out_unlock;
 751
 752	/* peer icmp_ratelimit */
 753	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
 754		goto ende;
 755
 756	/* RFC says return as much as we can without exceeding 576 bytes. */
 757
 758	room = dst_mtu(&rt->dst);
 759	if (room > 576)
 760		room = 576;
 761	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 762	room -= sizeof(struct icmphdr);
 763	/* Guard against tiny mtu. We need to include at least one
 764	 * IP network header for this message to make any sense.
 765	 */
 766	if (room <= (int)sizeof(struct iphdr))
 767		goto ende;
 768
 769	icmp_param.data_len = skb_in->len - icmp_param.offset;
 770	if (icmp_param.data_len > room)
 771		icmp_param.data_len = room;
 772	icmp_param.head_len = sizeof(struct icmphdr);
 773
 774	/* if we don't have a source address at this point, fall back to the
 775	 * dummy address instead of sending out a packet with a source address
 776	 * of 0.0.0.0
 777	 */
 778	if (!fl4.saddr)
 779		fl4.saddr = htonl(INADDR_DUMMY);
 780
 781	trace_icmp_send(skb_in, type, code);
 782
 783	icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
 784ende:
 785	ip_rt_put(rt);
 786out_unlock:
 787	icmp_xmit_unlock(sk);
 788out_bh_enable:
 789	local_bh_enable();
 790out:
 791	rcu_read_unlock();
 792}
 793EXPORT_SYMBOL(__icmp_send);
 794
 795#if IS_ENABLED(CONFIG_NF_NAT)
 796#include <net/netfilter/nf_conntrack.h>
 797void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 798{
 799	struct sk_buff *cloned_skb = NULL;
 800	struct ip_options opts = { 0 };
 801	enum ip_conntrack_info ctinfo;
 802	struct nf_conn *ct;
 803	__be32 orig_ip;
 804
 805	ct = nf_ct_get(skb_in, &ctinfo);
 806	if (!ct || !(ct->status & IPS_SRC_NAT)) {
 807		__icmp_send(skb_in, type, code, info, &opts);
 808		return;
 809	}
 810
 811	if (skb_shared(skb_in))
 812		skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
 813
 814	if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
 815	    (skb_network_header(skb_in) + sizeof(struct iphdr)) >
 816	    skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
 817	    skb_network_offset(skb_in) + sizeof(struct iphdr))))
 818		goto out;
 819
 820	orig_ip = ip_hdr(skb_in)->saddr;
 821	ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
 822	__icmp_send(skb_in, type, code, info, &opts);
 823	ip_hdr(skb_in)->saddr = orig_ip;
 824out:
 825	consume_skb(cloned_skb);
 826}
 827EXPORT_SYMBOL(icmp_ndo_send);
 828#endif
 829
 830static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 831{
 832	const struct iphdr *iph = (const struct iphdr *)skb->data;
 833	const struct net_protocol *ipprot;
 834	int protocol = iph->protocol;
 835
 836	/* Checkin full IP header plus 8 bytes of protocol to
 837	 * avoid additional coding at protocol handlers.
 838	 */
 839	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 840		__ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
 841		return;
 842	}
 843
 844	raw_icmp_error(skb, protocol, info);
 845
 846	ipprot = rcu_dereference(inet_protos[protocol]);
 847	if (ipprot && ipprot->err_handler)
 848		ipprot->err_handler(skb, info);
 849}
 850
 851static bool icmp_tag_validation(int proto)
 852{
 853	bool ok;
 854
 855	rcu_read_lock();
 856	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 857	rcu_read_unlock();
 858	return ok;
 859}
 860
 861/*
 862 *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 863 *	ICMP_PARAMETERPROB.
 864 */
 865
 866static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
 867{
 868	enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
 869	const struct iphdr *iph;
 870	struct icmphdr *icmph;
 871	struct net *net;
 872	u32 info = 0;
 873
 874	net = dev_net_rcu(skb_dst(skb)->dev);
 875
 876	/*
 877	 *	Incomplete header ?
 878	 * 	Only checks for the IP header, there should be an
 879	 *	additional check for longer headers in upper levels.
 880	 */
 881
 882	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 883		goto out_err;
 884
 885	icmph = icmp_hdr(skb);
 886	iph   = (const struct iphdr *)skb->data;
 887
 888	if (iph->ihl < 5)  { /* Mangled header, drop. */
 889		reason = SKB_DROP_REASON_IP_INHDR;
 890		goto out_err;
 891	}
 892
 893	switch (icmph->type) {
 894	case ICMP_DEST_UNREACH:
 895		switch (icmph->code & 15) {
 896		case ICMP_NET_UNREACH:
 897		case ICMP_HOST_UNREACH:
 898		case ICMP_PROT_UNREACH:
 899		case ICMP_PORT_UNREACH:
 900			break;
 901		case ICMP_FRAG_NEEDED:
 902			/* for documentation of the ip_no_pmtu_disc
 903			 * values please see
 904			 * Documentation/networking/ip-sysctl.rst
 905			 */
 906			switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
 907			default:
 908				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 909						    &iph->daddr);
 910				break;
 911			case 2:
 912				goto out;
 913			case 3:
 914				if (!icmp_tag_validation(iph->protocol))
 915					goto out;
 916				fallthrough;
 917			case 0:
 918				info = ntohs(icmph->un.frag.mtu);
 919			}
 920			break;
 921		case ICMP_SR_FAILED:
 922			net_dbg_ratelimited("%pI4: Source Route Failed\n",
 923					    &iph->daddr);
 924			break;
 925		default:
 926			break;
 927		}
 928		if (icmph->code > NR_ICMP_UNREACH)
 929			goto out;
 930		break;
 931	case ICMP_PARAMETERPROB:
 932		info = ntohl(icmph->un.gateway) >> 24;
 933		break;
 934	case ICMP_TIME_EXCEEDED:
 935		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 936		if (icmph->code == ICMP_EXC_FRAGTIME)
 937			goto out;
 938		break;
 939	}
 940
 941	/*
 942	 *	Throw it at our lower layers
 943	 *
 944	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 945	 *		  header.
 946	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 947	 *		  transport layer.
 948	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 949	 *		  transport layer.
 950	 */
 951
 952	/*
 953	 *	Check the other end isn't violating RFC 1122. Some routers send
 954	 *	bogus responses to broadcast frames. If you see this message
 955	 *	first check your netmask matches at both ends, if it does then
 956	 *	get the other vendor to fix their kit.
 957	 */
 958
 959	if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
 960	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 961		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 962				     &ip_hdr(skb)->saddr,
 963				     icmph->type, icmph->code,
 964				     &iph->daddr, skb->dev->name);
 965		goto out;
 966	}
 967
 968	icmp_socket_deliver(skb, info);
 969
 970out:
 971	return reason;
 972out_err:
 973	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 974	return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
 975}
 976
 977
 978/*
 979 *	Handle ICMP_REDIRECT.
 980 */
 981
 982static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
 983{
 984	if (skb->len < sizeof(struct iphdr)) {
 985		__ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
 986		return SKB_DROP_REASON_PKT_TOO_SMALL;
 987	}
 988
 989	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 990		/* there aught to be a stat */
 991		return SKB_DROP_REASON_NOMEM;
 992	}
 993
 994	icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
 995	return SKB_NOT_DROPPED_YET;
 996}
 997
 998/*
 999 *	Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
1000 *
1001 *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
1002 *		  requests.
1003 *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
1004 *		  included in the reply.
1005 *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
1006 *		  echo requests, MUST have default=NOT.
1007 *	RFC 8335: 8 MUST have a config option to enable/disable ICMP
1008 *		  Extended Echo Functionality, MUST be disabled by default
1009 *	See also WRT handling of options once they are done and working.
1010 */
1011
1012static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1013{
1014	struct icmp_bxm icmp_param;
1015	struct net *net;
1016
1017	net = dev_net_rcu(skb_dst(skb)->dev);
1018	/* should there be an ICMP stat for ignored echos? */
1019	if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1020		return SKB_NOT_DROPPED_YET;
1021
1022	icmp_param.data.icmph	   = *icmp_hdr(skb);
1023	icmp_param.skb		   = skb;
1024	icmp_param.offset	   = 0;
1025	icmp_param.data_len	   = skb->len;
1026	icmp_param.head_len	   = sizeof(struct icmphdr);
1027
1028	if (icmp_param.data.icmph.type == ICMP_ECHO)
1029		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1030	else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1031		return SKB_NOT_DROPPED_YET;
1032
1033	icmp_reply(&icmp_param, skb);
1034	return SKB_NOT_DROPPED_YET;
1035}
1036
1037/*	Helper for icmp_echo and icmpv6_echo_reply.
1038 *	Searches for net_device that matches PROBE interface identifier
1039 *		and builds PROBE reply message in icmphdr.
1040 *
1041 *	Returns false if PROBE responses are disabled via sysctl
1042 */
1043
1044bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1045{
1046	struct net *net = dev_net_rcu(skb->dev);
1047	struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1048	struct icmp_ext_echo_iio *iio, _iio;
1049	struct inet6_dev *in6_dev;
1050	struct in_device *in_dev;
1051	struct net_device *dev;
1052	char buff[IFNAMSIZ];
1053	u16 ident_len;
1054	u8 status;
1055
1056	if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1057		return false;
1058
1059	/* We currently only support probing interfaces on the proxy node
1060	 * Check to ensure L-bit is set
1061	 */
1062	if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1063		return false;
1064	/* Clear status bits in reply message */
1065	icmphdr->un.echo.sequence &= htons(0xFF00);
1066	if (icmphdr->type == ICMP_EXT_ECHO)
1067		icmphdr->type = ICMP_EXT_ECHOREPLY;
1068	else
1069		icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1070	ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1071	/* Size of iio is class_type dependent.
1072	 * Only check header here and assign length based on ctype in the switch statement
1073	 */
1074	iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1075	if (!ext_hdr || !iio)
1076		goto send_mal_query;
1077	if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1078	    ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1079		goto send_mal_query;
1080	ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1081	iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1082				 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1083	if (!iio)
1084		goto send_mal_query;
1085
1086	status = 0;
1087	dev = NULL;
1088	switch (iio->extobj_hdr.class_type) {
1089	case ICMP_EXT_ECHO_CTYPE_NAME:
1090		if (ident_len >= IFNAMSIZ)
1091			goto send_mal_query;
1092		memset(buff, 0, sizeof(buff));
1093		memcpy(buff, &iio->ident.name, ident_len);
1094		dev = dev_get_by_name(net, buff);
1095		break;
1096	case ICMP_EXT_ECHO_CTYPE_INDEX:
1097		if (ident_len != sizeof(iio->ident.ifindex))
1098			goto send_mal_query;
1099		dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1100		break;
1101	case ICMP_EXT_ECHO_CTYPE_ADDR:
1102		if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1103		    ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1104				 iio->ident.addr.ctype3_hdr.addrlen)
1105			goto send_mal_query;
1106		switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1107		case ICMP_AFI_IP:
1108			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1109				goto send_mal_query;
1110			dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1111			break;
1112#if IS_ENABLED(CONFIG_IPV6)
1113		case ICMP_AFI_IP6:
1114			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1115				goto send_mal_query;
1116			dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1117			dev_hold(dev);
1118			break;
1119#endif
1120		default:
1121			goto send_mal_query;
1122		}
1123		break;
1124	default:
1125		goto send_mal_query;
1126	}
1127	if (!dev) {
1128		icmphdr->code = ICMP_EXT_CODE_NO_IF;
1129		return true;
1130	}
1131	/* Fill bits in reply message */
1132	if (dev->flags & IFF_UP)
1133		status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1134
1135	in_dev = __in_dev_get_rcu(dev);
1136	if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1137		status |= ICMP_EXT_ECHOREPLY_IPV4;
1138
1139	in6_dev = __in6_dev_get(dev);
1140	if (in6_dev && !list_empty(&in6_dev->addr_list))
1141		status |= ICMP_EXT_ECHOREPLY_IPV6;
1142
1143	dev_put(dev);
1144	icmphdr->un.echo.sequence |= htons(status);
1145	return true;
1146send_mal_query:
1147	icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1148	return true;
1149}
1150EXPORT_SYMBOL_GPL(icmp_build_probe);
1151
1152/*
1153 *	Handle ICMP Timestamp requests.
1154 *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1155 *		  SHOULD be in the kernel for minimum random latency.
1156 *		  MUST be accurate to a few minutes.
1157 *		  MUST be updated at least at 15Hz.
1158 */
1159static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1160{
1161	struct icmp_bxm icmp_param;
1162	/*
1163	 *	Too short.
1164	 */
1165	if (skb->len < 4)
1166		goto out_err;
1167
1168	/*
1169	 *	Fill in the current time as ms since midnight UT:
1170	 */
1171	icmp_param.data.times[1] = inet_current_timestamp();
1172	icmp_param.data.times[2] = icmp_param.data.times[1];
1173
1174	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1175
1176	icmp_param.data.icmph	   = *icmp_hdr(skb);
1177	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1178	icmp_param.data.icmph.code = 0;
1179	icmp_param.skb		   = skb;
1180	icmp_param.offset	   = 0;
1181	icmp_param.data_len	   = 0;
1182	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
1183	icmp_reply(&icmp_param, skb);
1184	return SKB_NOT_DROPPED_YET;
1185
1186out_err:
1187	__ICMP_INC_STATS(dev_net_rcu(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1188	return SKB_DROP_REASON_PKT_TOO_SMALL;
1189}
1190
1191static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1192{
1193	/* pretend it was a success */
1194	return SKB_NOT_DROPPED_YET;
1195}
1196
1197/*
1198 *	Deal with incoming ICMP packets.
1199 */
1200int icmp_rcv(struct sk_buff *skb)
1201{
1202	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1203	struct rtable *rt = skb_rtable(skb);
1204	struct net *net = dev_net_rcu(rt->dst.dev);
1205	struct icmphdr *icmph;
1206
1207	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1208		struct sec_path *sp = skb_sec_path(skb);
1209		int nh;
1210
1211		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1212				 XFRM_STATE_ICMP)) {
1213			reason = SKB_DROP_REASON_XFRM_POLICY;
1214			goto drop;
1215		}
1216
1217		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1218			goto drop;
1219
1220		nh = skb_network_offset(skb);
1221		skb_set_network_header(skb, sizeof(*icmph));
1222
1223		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1224						skb)) {
1225			reason = SKB_DROP_REASON_XFRM_POLICY;
1226			goto drop;
1227		}
1228
1229		skb_set_network_header(skb, nh);
1230	}
1231
1232	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1233
1234	if (skb_checksum_simple_validate(skb))
1235		goto csum_error;
1236
1237	if (!pskb_pull(skb, sizeof(*icmph)))
1238		goto error;
1239
1240	icmph = icmp_hdr(skb);
1241
1242	ICMPMSGIN_INC_STATS(net, icmph->type);
1243
1244	/* Check for ICMP Extended Echo (PROBE) messages */
1245	if (icmph->type == ICMP_EXT_ECHO) {
1246		/* We can't use icmp_pointers[].handler() because it is an array of
1247		 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1248		 */
1249		reason = icmp_echo(skb);
1250		goto reason_check;
1251	}
1252
1253	if (icmph->type == ICMP_EXT_ECHOREPLY) {
1254		reason = ping_rcv(skb);
1255		goto reason_check;
1256	}
1257
1258	/*
1259	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1260	 *
1261	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1262	 *		  discarded.
1263	 */
1264	if (icmph->type > NR_ICMP_TYPES) {
1265		reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1266		goto error;
1267	}
1268
1269	/*
1270	 *	Parse the ICMP message
1271	 */
1272
1273	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1274		/*
1275		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1276		 *	  silently ignored (we let user decide with a sysctl).
1277		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1278		 *	  discarded if to broadcast/multicast.
1279		 */
1280		if ((icmph->type == ICMP_ECHO ||
1281		     icmph->type == ICMP_TIMESTAMP) &&
1282		    READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1283			reason = SKB_DROP_REASON_INVALID_PROTO;
1284			goto error;
1285		}
1286		if (icmph->type != ICMP_ECHO &&
1287		    icmph->type != ICMP_TIMESTAMP &&
1288		    icmph->type != ICMP_ADDRESS &&
1289		    icmph->type != ICMP_ADDRESSREPLY) {
1290			reason = SKB_DROP_REASON_INVALID_PROTO;
1291			goto error;
1292		}
1293	}
1294
1295	reason = icmp_pointers[icmph->type].handler(skb);
1296reason_check:
1297	if (!reason)  {
1298		consume_skb(skb);
1299		return NET_RX_SUCCESS;
1300	}
1301
1302drop:
1303	kfree_skb_reason(skb, reason);
1304	return NET_RX_DROP;
1305csum_error:
1306	reason = SKB_DROP_REASON_ICMP_CSUM;
1307	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1308error:
1309	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1310	goto drop;
1311}
1312
1313static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1314{
1315	struct icmp_extobj_hdr *objh, _objh;
1316	struct icmp_ext_hdr *exth, _exth;
1317	u16 olen;
1318
1319	exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1320	if (!exth)
1321		return false;
1322	if (exth->version != 2)
1323		return true;
1324
1325	if (exth->checksum &&
1326	    csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1327		return false;
1328
1329	off += sizeof(_exth);
1330	while (off < skb->len) {
1331		objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1332		if (!objh)
1333			return false;
1334
1335		olen = ntohs(objh->length);
1336		if (olen < sizeof(_objh))
1337			return false;
1338
1339		off += olen;
1340		if (off > skb->len)
1341			return false;
1342	}
1343
1344	return true;
1345}
1346
1347void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1348			   struct sock_ee_data_rfc4884 *out,
1349			   int thlen, int off)
1350{
1351	int hlen;
1352
1353	/* original datagram headers: end of icmph to payload (skb->data) */
1354	hlen = -skb_transport_offset(skb) - thlen;
1355
1356	/* per rfc 4884: minimal datagram length of 128 bytes */
1357	if (off < 128 || off < hlen)
1358		return;
1359
1360	/* kernel has stripped headers: return payload offset in bytes */
1361	off -= hlen;
1362	if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1363		return;
1364
1365	out->len = off;
1366
1367	if (!ip_icmp_error_rfc4884_validate(skb, off))
1368		out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1369}
1370EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1371
1372int icmp_err(struct sk_buff *skb, u32 info)
1373{
1374	struct iphdr *iph = (struct iphdr *)skb->data;
1375	int offset = iph->ihl<<2;
1376	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1377	struct net *net = dev_net_rcu(skb->dev);
1378	int type = icmp_hdr(skb)->type;
1379	int code = icmp_hdr(skb)->code;
 
1380
1381	/*
1382	 * Use ping_err to handle all icmp errors except those
1383	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1384	 */
1385	if (icmph->type != ICMP_ECHOREPLY) {
1386		ping_err(skb, offset, info);
1387		return 0;
1388	}
1389
1390	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1391		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1392	else if (type == ICMP_REDIRECT)
1393		ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1394
1395	return 0;
1396}
1397
1398/*
1399 *	This table is the definition of how we handle ICMP.
1400 */
1401static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1402	[ICMP_ECHOREPLY] = {
1403		.handler = ping_rcv,
1404	},
1405	[1] = {
1406		.handler = icmp_discard,
1407		.error = 1,
1408	},
1409	[2] = {
1410		.handler = icmp_discard,
1411		.error = 1,
1412	},
1413	[ICMP_DEST_UNREACH] = {
1414		.handler = icmp_unreach,
1415		.error = 1,
1416	},
1417	[ICMP_SOURCE_QUENCH] = {
1418		.handler = icmp_unreach,
1419		.error = 1,
1420	},
1421	[ICMP_REDIRECT] = {
1422		.handler = icmp_redirect,
1423		.error = 1,
1424	},
1425	[6] = {
1426		.handler = icmp_discard,
1427		.error = 1,
1428	},
1429	[7] = {
1430		.handler = icmp_discard,
1431		.error = 1,
1432	},
1433	[ICMP_ECHO] = {
1434		.handler = icmp_echo,
1435	},
1436	[9] = {
1437		.handler = icmp_discard,
1438		.error = 1,
1439	},
1440	[10] = {
1441		.handler = icmp_discard,
1442		.error = 1,
1443	},
1444	[ICMP_TIME_EXCEEDED] = {
1445		.handler = icmp_unreach,
1446		.error = 1,
1447	},
1448	[ICMP_PARAMETERPROB] = {
1449		.handler = icmp_unreach,
1450		.error = 1,
1451	},
1452	[ICMP_TIMESTAMP] = {
1453		.handler = icmp_timestamp,
1454	},
1455	[ICMP_TIMESTAMPREPLY] = {
1456		.handler = icmp_discard,
1457	},
1458	[ICMP_INFO_REQUEST] = {
1459		.handler = icmp_discard,
1460	},
1461	[ICMP_INFO_REPLY] = {
1462		.handler = icmp_discard,
1463	},
1464	[ICMP_ADDRESS] = {
1465		.handler = icmp_discard,
1466	},
1467	[ICMP_ADDRESSREPLY] = {
1468		.handler = icmp_discard,
1469	},
1470};
1471
1472static int __net_init icmp_sk_init(struct net *net)
1473{
1474	/* Control parameters for ECHO replies. */
1475	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1476	net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1477	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1478
1479	/* Control parameter - ignore bogus broadcast responses? */
1480	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1481
1482	/*
1483	 * 	Configurable global rate limit.
1484	 *
1485	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1486	 *	bucket ratemask defines which icmp types are ratelimited by
1487	 *	setting	it's bit position.
1488	 *
1489	 *	default:
1490	 *	dest unreachable (3), source quench (4),
1491	 *	time exceeded (11), parameter problem (12)
1492	 */
1493
1494	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1495	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1496	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1497	net->ipv4.sysctl_icmp_msgs_per_sec = 1000;
1498	net->ipv4.sysctl_icmp_msgs_burst = 50;
1499
1500	return 0;
1501}
1502
1503static struct pernet_operations __net_initdata icmp_sk_ops = {
1504       .init = icmp_sk_init,
1505};
1506
1507int __init icmp_init(void)
1508{
1509	int err, i;
1510
1511	for_each_possible_cpu(i) {
1512		struct sock *sk;
1513
1514		err = inet_ctl_sock_create(&sk, PF_INET,
1515					   SOCK_RAW, IPPROTO_ICMP, &init_net);
1516		if (err < 0)
1517			return err;
1518
1519		per_cpu(ipv4_icmp_sk, i) = sk;
1520
1521		/* Enough space for 2 64K ICMP packets, including
1522		 * sk_buff/skb_shared_info struct overhead.
1523		 */
1524		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1525
1526		/*
1527		 * Speedup sock_wfree()
1528		 */
1529		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1530		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1531	}
1532	return register_pernet_subsys(&icmp_sk_ops);
1533}
v6.8
   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	enum skb_drop_reason (*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
 195static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
 196
 197/* Called with BH disabled */
 198static inline struct sock *icmp_xmit_lock(struct net *net)
 199{
 200	struct sock *sk;
 201
 202	sk = this_cpu_read(ipv4_icmp_sk);
 203
 204	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 205		/* This can happen if the output path signals a
 206		 * dst_link_failure() for an outgoing ICMP packet.
 207		 */
 208		return NULL;
 209	}
 210	sock_net_set(sk, net);
 211	return sk;
 212}
 213
 214static inline void icmp_xmit_unlock(struct sock *sk)
 215{
 216	sock_net_set(sk, &init_net);
 217	spin_unlock(&sk->sk_lock.slock);
 218}
 219
 220int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
 221int sysctl_icmp_msgs_burst __read_mostly = 50;
 222
 223static struct {
 224	spinlock_t	lock;
 225	u32		credit;
 226	u32		stamp;
 227} icmp_global = {
 228	.lock		= __SPIN_LOCK_UNLOCKED(icmp_global.lock),
 229};
 230
 231/**
 232 * icmp_global_allow - Are we allowed to send one more ICMP message ?
 
 233 *
 234 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
 235 * Returns false if we reached the limit and can not send another packet.
 236 * Note: called with BH disabled
 237 */
 238bool icmp_global_allow(void)
 239{
 240	u32 credit, delta, incr = 0, now = (u32)jiffies;
 241	bool rc = false;
 
 
 
 
 
 
 
 242
 243	/* Check if token bucket is empty and cannot be refilled
 244	 * without taking the spinlock. The READ_ONCE() are paired
 245	 * with the following WRITE_ONCE() in this same function.
 246	 */
 247	if (!READ_ONCE(icmp_global.credit)) {
 248		delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
 249		if (delta < HZ / 50)
 250			return false;
 251	}
 252
 253	spin_lock(&icmp_global.lock);
 254	delta = min_t(u32, now - icmp_global.stamp, HZ);
 255	if (delta >= HZ / 50) {
 256		incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
 257		if (incr)
 258			WRITE_ONCE(icmp_global.stamp, now);
 259	}
 260	credit = min_t(u32, icmp_global.credit + incr,
 261		       READ_ONCE(sysctl_icmp_msgs_burst));
 262	if (credit) {
 263		/* We want to use a credit of one in average, but need to randomize
 264		 * it for security reasons.
 265		 */
 266		credit = max_t(int, credit - get_random_u32_below(3), 0);
 267		rc = true;
 268	}
 269	WRITE_ONCE(icmp_global.credit, credit);
 270	spin_unlock(&icmp_global.lock);
 271	return rc;
 272}
 273EXPORT_SYMBOL(icmp_global_allow);
 274
 
 
 
 
 
 
 
 
 
 
 275static bool icmpv4_mask_allow(struct net *net, int type, int code)
 276{
 277	if (type > NR_ICMP_TYPES)
 278		return true;
 279
 280	/* Don't limit PMTU discovery. */
 281	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
 282		return true;
 283
 284	/* Limit if icmp type is enabled in ratemask. */
 285	if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
 286		return true;
 287
 288	return false;
 289}
 290
 291static bool icmpv4_global_allow(struct net *net, int type, int code)
 
 292{
 293	if (icmpv4_mask_allow(net, type, code))
 294		return true;
 295
 296	if (icmp_global_allow())
 
 297		return true;
 298
 299	__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
 300	return false;
 301}
 302
 303/*
 304 *	Send an ICMP frame.
 305 */
 306
 307static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
 308			       struct flowi4 *fl4, int type, int code)
 
 309{
 310	struct dst_entry *dst = &rt->dst;
 311	struct inet_peer *peer;
 312	bool rc = true;
 313	int vif;
 314
 315	if (icmpv4_mask_allow(net, type, code))
 316		goto out;
 317
 318	/* No rate limit on loopback */
 319	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
 320		goto out;
 321
 322	vif = l3mdev_master_ifindex(dst->dev);
 323	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
 
 324	rc = inet_peer_xrlim_allow(peer,
 325				   READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
 326	if (peer)
 327		inet_putpeer(peer);
 328out:
 329	if (!rc)
 330		__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
 
 
 331	return rc;
 332}
 333
 334/*
 335 *	Maintain the counters used in the SNMP statistics for outgoing ICMP
 336 */
 337void icmp_out_count(struct net *net, unsigned char type)
 338{
 339	ICMPMSGOUT_INC_STATS(net, type);
 340	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
 341}
 342
 343/*
 344 *	Checksum each fragment, and on the first include the headers and final
 345 *	checksum.
 346 */
 347static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 348			  struct sk_buff *skb)
 349{
 350	struct icmp_bxm *icmp_param = from;
 351	__wsum csum;
 352
 353	csum = skb_copy_and_csum_bits(icmp_param->skb,
 354				      icmp_param->offset + offset,
 355				      to, len);
 356
 357	skb->csum = csum_block_add(skb->csum, csum, odd);
 358	if (icmp_pointers[icmp_param->data.icmph.type].error)
 359		nf_ct_attach(skb, icmp_param->skb);
 360	return 0;
 361}
 362
 363static void icmp_push_reply(struct sock *sk,
 364			    struct icmp_bxm *icmp_param,
 365			    struct flowi4 *fl4,
 366			    struct ipcm_cookie *ipc, struct rtable **rt)
 367{
 368	struct sk_buff *skb;
 369
 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;
 379		struct sk_buff *skb1;
 380
 381		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
 382						 (char *)icmph,
 383						 icmp_param->head_len);
 384		skb_queue_walk(&sk->sk_write_queue, skb1) {
 385			csum = csum_add(csum, skb1->csum);
 386		}
 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	ipc.sockc.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_common(&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(sk, 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
 459/*
 460 * The device used for looking up which routing table to use for sending an ICMP
 461 * error is preferably the source whenever it is set, which should ensure the
 462 * icmp error can be sent to the source host, else lookup using the routing
 463 * table of the destination device, else use the main routing table (index 0).
 464 */
 465static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
 466{
 467	struct net_device *route_lookup_dev = NULL;
 468
 469	if (skb->dev)
 470		route_lookup_dev = skb->dev;
 471	else if (skb_dst(skb))
 472		route_lookup_dev = skb_dst(skb)->dev;
 473	return route_lookup_dev;
 474}
 475
 476static struct rtable *icmp_route_lookup(struct net *net,
 477					struct flowi4 *fl4,
 478					struct sk_buff *skb_in,
 479					const struct iphdr *iph,
 480					__be32 saddr, u8 tos, u32 mark,
 481					int type, int code,
 482					struct icmp_bxm *param)
 483{
 484	struct net_device *route_lookup_dev;
 
 485	struct rtable *rt, *rt2;
 486	struct flowi4 fl4_dec;
 487	int err;
 488
 489	memset(fl4, 0, sizeof(*fl4));
 490	fl4->daddr = (param->replyopts.opt.opt.srr ?
 491		      param->replyopts.opt.opt.faddr : iph->saddr);
 492	fl4->saddr = saddr;
 493	fl4->flowi4_mark = mark;
 494	fl4->flowi4_uid = sock_net_uid(net, NULL);
 495	fl4->flowi4_tos = RT_TOS(tos);
 496	fl4->flowi4_proto = IPPROTO_ICMP;
 497	fl4->fl4_icmp_type = type;
 498	fl4->fl4_icmp_code = code;
 499	route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
 500	fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
 501
 502	security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
 503	rt = ip_route_output_key_hash(net, fl4, skb_in);
 504	if (IS_ERR(rt))
 505		return rt;
 506
 507	/* No need to clone since we're just using its address. */
 508	rt2 = rt;
 509
 510	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
 511					   flowi4_to_flowi(fl4), NULL, 0);
 512	if (!IS_ERR(rt)) {
 
 513		if (rt != rt2)
 514			return rt;
 515	} else if (PTR_ERR(rt) == -EPERM) {
 
 
 
 516		rt = NULL;
 517	} else
 518		return rt;
 519
 520	err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
 521	if (err)
 522		goto relookup_failed;
 523
 524	if (inet_addr_type_dev_table(net, route_lookup_dev,
 525				     fl4_dec.saddr) == RTN_LOCAL) {
 526		rt2 = __ip_route_output_key(net, &fl4_dec);
 527		if (IS_ERR(rt2))
 528			err = PTR_ERR(rt2);
 529	} else {
 530		struct flowi4 fl4_2 = {};
 531		unsigned long orefdst;
 532
 533		fl4_2.daddr = fl4_dec.saddr;
 534		rt2 = ip_route_output_key(net, &fl4_2);
 535		if (IS_ERR(rt2)) {
 536			err = PTR_ERR(rt2);
 537			goto relookup_failed;
 538		}
 539		/* Ugh! */
 540		orefdst = skb_in->_skb_refdst; /* save old refdst */
 541		skb_dst_set(skb_in, NULL);
 542		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
 543				     RT_TOS(tos), rt2->dst.dev);
 544
 545		dst_release(&rt2->dst);
 546		rt2 = skb_rtable(skb_in);
 547		skb_in->_skb_refdst = orefdst; /* restore old refdst */
 548	}
 549
 550	if (err)
 551		goto relookup_failed;
 552
 553	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
 554					    flowi4_to_flowi(&fl4_dec), NULL,
 555					    XFRM_LOOKUP_ICMP);
 556	if (!IS_ERR(rt2)) {
 557		dst_release(&rt->dst);
 558		memcpy(fl4, &fl4_dec, sizeof(*fl4));
 559		rt = rt2;
 560	} else if (PTR_ERR(rt2) == -EPERM) {
 561		if (rt)
 562			dst_release(&rt->dst);
 563		return rt2;
 564	} else {
 565		err = PTR_ERR(rt2);
 566		goto relookup_failed;
 567	}
 568	return rt;
 569
 570relookup_failed:
 571	if (rt)
 572		return rt;
 573	return ERR_PTR(err);
 574}
 575
 576/*
 577 *	Send an ICMP message in response to a situation
 578 *
 579 *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
 580 *		  MAY send more (we do).
 581 *			MUST NOT change this header information.
 582 *			MUST NOT reply to a multicast/broadcast IP address.
 583 *			MUST NOT reply to a multicast/broadcast MAC address.
 584 *			MUST reply to only the first fragment.
 585 */
 586
 587void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
 588		 const struct ip_options *opt)
 589{
 590	struct iphdr *iph;
 591	int room;
 592	struct icmp_bxm icmp_param;
 593	struct rtable *rt = skb_rtable(skb_in);
 
 594	struct ipcm_cookie ipc;
 595	struct flowi4 fl4;
 596	__be32 saddr;
 597	u8  tos;
 598	u32 mark;
 599	struct net *net;
 600	struct sock *sk;
 601
 602	if (!rt)
 603		goto out;
 
 
 604
 605	if (rt->dst.dev)
 606		net = dev_net(rt->dst.dev);
 607	else if (skb_in->dev)
 608		net = dev_net(skb_in->dev);
 609	else
 610		goto out;
 611
 612	/*
 613	 *	Find the original header. It is expected to be valid, of course.
 614	 *	Check this, icmp_send is called from the most obscure devices
 615	 *	sometimes.
 616	 */
 617	iph = ip_hdr(skb_in);
 618
 619	if ((u8 *)iph < skb_in->head ||
 620	    (skb_network_header(skb_in) + sizeof(*iph)) >
 621	    skb_tail_pointer(skb_in))
 622		goto out;
 623
 624	/*
 625	 *	No replies to physical multicast/broadcast
 626	 */
 627	if (skb_in->pkt_type != PACKET_HOST)
 628		goto out;
 629
 630	/*
 631	 *	Now check at the protocol level
 632	 */
 633	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 634		goto out;
 635
 636	/*
 637	 *	Only reply to fragment 0. We byte re-order the constant
 638	 *	mask for efficiency.
 639	 */
 640	if (iph->frag_off & htons(IP_OFFSET))
 641		goto out;
 642
 643	/*
 644	 *	If we send an ICMP error to an ICMP error a mess would result..
 645	 */
 646	if (icmp_pointers[type].error) {
 647		/*
 648		 *	We are an error, check if we are replying to an
 649		 *	ICMP error
 650		 */
 651		if (iph->protocol == IPPROTO_ICMP) {
 652			u8 _inner_type, *itp;
 653
 654			itp = skb_header_pointer(skb_in,
 655						 skb_network_header(skb_in) +
 656						 (iph->ihl << 2) +
 657						 offsetof(struct icmphdr,
 658							  type) -
 659						 skb_in->data,
 660						 sizeof(_inner_type),
 661						 &_inner_type);
 662			if (!itp)
 663				goto out;
 664
 665			/*
 666			 *	Assume any unknown ICMP type is an error. This
 667			 *	isn't specified by the RFC, but think about it..
 668			 */
 669			if (*itp > NR_ICMP_TYPES ||
 670			    icmp_pointers[*itp].error)
 671				goto out;
 672		}
 673	}
 674
 675	/* Needed by both icmp_global_allow and icmp_xmit_lock */
 676	local_bh_disable();
 677
 678	/* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
 679	 * incoming dev is loopback.  If outgoing dev change to not be
 680	 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
 681	 */
 682	if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
 683	      !icmpv4_global_allow(net, type, code))
 684		goto out_bh_enable;
 685
 686	sk = icmp_xmit_lock(net);
 687	if (!sk)
 688		goto out_bh_enable;
 689
 690	/*
 691	 *	Construct source address and options.
 692	 */
 693
 694	saddr = iph->daddr;
 695	if (!(rt->rt_flags & RTCF_LOCAL)) {
 696		struct net_device *dev = NULL;
 697
 698		rcu_read_lock();
 699		if (rt_is_input_route(rt) &&
 700		    READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
 701			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
 702
 703		if (dev)
 704			saddr = inet_select_addr(dev, iph->saddr,
 705						 RT_SCOPE_LINK);
 706		else
 707			saddr = 0;
 708		rcu_read_unlock();
 709	}
 710
 711	tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
 712					   IPTOS_PREC_INTERNETCONTROL) :
 713					   iph->tos;
 714	mark = IP4_REPLY_MARK(net, skb_in->mark);
 715
 716	if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
 717		goto out_unlock;
 718
 719
 720	/*
 721	 *	Prepare data for ICMP header.
 722	 */
 723
 724	icmp_param.data.icmph.type	 = type;
 725	icmp_param.data.icmph.code	 = code;
 726	icmp_param.data.icmph.un.gateway = info;
 727	icmp_param.data.icmph.checksum	 = 0;
 728	icmp_param.skb	  = skb_in;
 729	icmp_param.offset = skb_network_offset(skb_in);
 730	inet_sk(sk)->tos = tos;
 731	ipcm_init(&ipc);
 732	ipc.addr = iph->saddr;
 733	ipc.opt = &icmp_param.replyopts.opt;
 734	ipc.sockc.mark = mark;
 735
 736	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
 737			       type, code, &icmp_param);
 
 738	if (IS_ERR(rt))
 739		goto out_unlock;
 740
 741	/* peer icmp_ratelimit */
 742	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
 743		goto ende;
 744
 745	/* RFC says return as much as we can without exceeding 576 bytes. */
 746
 747	room = dst_mtu(&rt->dst);
 748	if (room > 576)
 749		room = 576;
 750	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
 751	room -= sizeof(struct icmphdr);
 752	/* Guard against tiny mtu. We need to include at least one
 753	 * IP network header for this message to make any sense.
 754	 */
 755	if (room <= (int)sizeof(struct iphdr))
 756		goto ende;
 757
 758	icmp_param.data_len = skb_in->len - icmp_param.offset;
 759	if (icmp_param.data_len > room)
 760		icmp_param.data_len = room;
 761	icmp_param.head_len = sizeof(struct icmphdr);
 762
 763	/* if we don't have a source address at this point, fall back to the
 764	 * dummy address instead of sending out a packet with a source address
 765	 * of 0.0.0.0
 766	 */
 767	if (!fl4.saddr)
 768		fl4.saddr = htonl(INADDR_DUMMY);
 769
 
 
 770	icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
 771ende:
 772	ip_rt_put(rt);
 773out_unlock:
 774	icmp_xmit_unlock(sk);
 775out_bh_enable:
 776	local_bh_enable();
 777out:;
 
 778}
 779EXPORT_SYMBOL(__icmp_send);
 780
 781#if IS_ENABLED(CONFIG_NF_NAT)
 782#include <net/netfilter/nf_conntrack.h>
 783void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
 784{
 785	struct sk_buff *cloned_skb = NULL;
 786	struct ip_options opts = { 0 };
 787	enum ip_conntrack_info ctinfo;
 788	struct nf_conn *ct;
 789	__be32 orig_ip;
 790
 791	ct = nf_ct_get(skb_in, &ctinfo);
 792	if (!ct || !(ct->status & IPS_SRC_NAT)) {
 793		__icmp_send(skb_in, type, code, info, &opts);
 794		return;
 795	}
 796
 797	if (skb_shared(skb_in))
 798		skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
 799
 800	if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
 801	    (skb_network_header(skb_in) + sizeof(struct iphdr)) >
 802	    skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
 803	    skb_network_offset(skb_in) + sizeof(struct iphdr))))
 804		goto out;
 805
 806	orig_ip = ip_hdr(skb_in)->saddr;
 807	ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
 808	__icmp_send(skb_in, type, code, info, &opts);
 809	ip_hdr(skb_in)->saddr = orig_ip;
 810out:
 811	consume_skb(cloned_skb);
 812}
 813EXPORT_SYMBOL(icmp_ndo_send);
 814#endif
 815
 816static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
 817{
 818	const struct iphdr *iph = (const struct iphdr *)skb->data;
 819	const struct net_protocol *ipprot;
 820	int protocol = iph->protocol;
 821
 822	/* Checkin full IP header plus 8 bytes of protocol to
 823	 * avoid additional coding at protocol handlers.
 824	 */
 825	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
 826		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 827		return;
 828	}
 829
 830	raw_icmp_error(skb, protocol, info);
 831
 832	ipprot = rcu_dereference(inet_protos[protocol]);
 833	if (ipprot && ipprot->err_handler)
 834		ipprot->err_handler(skb, info);
 835}
 836
 837static bool icmp_tag_validation(int proto)
 838{
 839	bool ok;
 840
 841	rcu_read_lock();
 842	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
 843	rcu_read_unlock();
 844	return ok;
 845}
 846
 847/*
 848 *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
 849 *	ICMP_PARAMETERPROB.
 850 */
 851
 852static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
 853{
 854	enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
 855	const struct iphdr *iph;
 856	struct icmphdr *icmph;
 857	struct net *net;
 858	u32 info = 0;
 859
 860	net = dev_net(skb_dst(skb)->dev);
 861
 862	/*
 863	 *	Incomplete header ?
 864	 * 	Only checks for the IP header, there should be an
 865	 *	additional check for longer headers in upper levels.
 866	 */
 867
 868	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 869		goto out_err;
 870
 871	icmph = icmp_hdr(skb);
 872	iph   = (const struct iphdr *)skb->data;
 873
 874	if (iph->ihl < 5)  { /* Mangled header, drop. */
 875		reason = SKB_DROP_REASON_IP_INHDR;
 876		goto out_err;
 877	}
 878
 879	switch (icmph->type) {
 880	case ICMP_DEST_UNREACH:
 881		switch (icmph->code & 15) {
 882		case ICMP_NET_UNREACH:
 883		case ICMP_HOST_UNREACH:
 884		case ICMP_PROT_UNREACH:
 885		case ICMP_PORT_UNREACH:
 886			break;
 887		case ICMP_FRAG_NEEDED:
 888			/* for documentation of the ip_no_pmtu_disc
 889			 * values please see
 890			 * Documentation/networking/ip-sysctl.rst
 891			 */
 892			switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
 893			default:
 894				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
 895						    &iph->daddr);
 896				break;
 897			case 2:
 898				goto out;
 899			case 3:
 900				if (!icmp_tag_validation(iph->protocol))
 901					goto out;
 902				fallthrough;
 903			case 0:
 904				info = ntohs(icmph->un.frag.mtu);
 905			}
 906			break;
 907		case ICMP_SR_FAILED:
 908			net_dbg_ratelimited("%pI4: Source Route Failed\n",
 909					    &iph->daddr);
 910			break;
 911		default:
 912			break;
 913		}
 914		if (icmph->code > NR_ICMP_UNREACH)
 915			goto out;
 916		break;
 917	case ICMP_PARAMETERPROB:
 918		info = ntohl(icmph->un.gateway) >> 24;
 919		break;
 920	case ICMP_TIME_EXCEEDED:
 921		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
 922		if (icmph->code == ICMP_EXC_FRAGTIME)
 923			goto out;
 924		break;
 925	}
 926
 927	/*
 928	 *	Throw it at our lower layers
 929	 *
 930	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
 931	 *		  header.
 932	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
 933	 *		  transport layer.
 934	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
 935	 *		  transport layer.
 936	 */
 937
 938	/*
 939	 *	Check the other end isn't violating RFC 1122. Some routers send
 940	 *	bogus responses to broadcast frames. If you see this message
 941	 *	first check your netmask matches at both ends, if it does then
 942	 *	get the other vendor to fix their kit.
 943	 */
 944
 945	if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
 946	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
 947		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
 948				     &ip_hdr(skb)->saddr,
 949				     icmph->type, icmph->code,
 950				     &iph->daddr, skb->dev->name);
 951		goto out;
 952	}
 953
 954	icmp_socket_deliver(skb, info);
 955
 956out:
 957	return reason;
 958out_err:
 959	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 960	return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
 961}
 962
 963
 964/*
 965 *	Handle ICMP_REDIRECT.
 966 */
 967
 968static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
 969{
 970	if (skb->len < sizeof(struct iphdr)) {
 971		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
 972		return SKB_DROP_REASON_PKT_TOO_SMALL;
 973	}
 974
 975	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
 976		/* there aught to be a stat */
 977		return SKB_DROP_REASON_NOMEM;
 978	}
 979
 980	icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
 981	return SKB_NOT_DROPPED_YET;
 982}
 983
 984/*
 985 *	Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
 986 *
 987 *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
 988 *		  requests.
 989 *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
 990 *		  included in the reply.
 991 *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
 992 *		  echo requests, MUST have default=NOT.
 993 *	RFC 8335: 8 MUST have a config option to enable/disable ICMP
 994 *		  Extended Echo Functionality, MUST be disabled by default
 995 *	See also WRT handling of options once they are done and working.
 996 */
 997
 998static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
 999{
1000	struct icmp_bxm icmp_param;
1001	struct net *net;
1002
1003	net = dev_net(skb_dst(skb)->dev);
1004	/* should there be an ICMP stat for ignored echos? */
1005	if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1006		return SKB_NOT_DROPPED_YET;
1007
1008	icmp_param.data.icmph	   = *icmp_hdr(skb);
1009	icmp_param.skb		   = skb;
1010	icmp_param.offset	   = 0;
1011	icmp_param.data_len	   = skb->len;
1012	icmp_param.head_len	   = sizeof(struct icmphdr);
1013
1014	if (icmp_param.data.icmph.type == ICMP_ECHO)
1015		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1016	else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1017		return SKB_NOT_DROPPED_YET;
1018
1019	icmp_reply(&icmp_param, skb);
1020	return SKB_NOT_DROPPED_YET;
1021}
1022
1023/*	Helper for icmp_echo and icmpv6_echo_reply.
1024 *	Searches for net_device that matches PROBE interface identifier
1025 *		and builds PROBE reply message in icmphdr.
1026 *
1027 *	Returns false if PROBE responses are disabled via sysctl
1028 */
1029
1030bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1031{
 
1032	struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1033	struct icmp_ext_echo_iio *iio, _iio;
1034	struct net *net = dev_net(skb->dev);
 
1035	struct net_device *dev;
1036	char buff[IFNAMSIZ];
1037	u16 ident_len;
1038	u8 status;
1039
1040	if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1041		return false;
1042
1043	/* We currently only support probing interfaces on the proxy node
1044	 * Check to ensure L-bit is set
1045	 */
1046	if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1047		return false;
1048	/* Clear status bits in reply message */
1049	icmphdr->un.echo.sequence &= htons(0xFF00);
1050	if (icmphdr->type == ICMP_EXT_ECHO)
1051		icmphdr->type = ICMP_EXT_ECHOREPLY;
1052	else
1053		icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1054	ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1055	/* Size of iio is class_type dependent.
1056	 * Only check header here and assign length based on ctype in the switch statement
1057	 */
1058	iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1059	if (!ext_hdr || !iio)
1060		goto send_mal_query;
1061	if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1062	    ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1063		goto send_mal_query;
1064	ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1065	iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1066				 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1067	if (!iio)
1068		goto send_mal_query;
1069
1070	status = 0;
1071	dev = NULL;
1072	switch (iio->extobj_hdr.class_type) {
1073	case ICMP_EXT_ECHO_CTYPE_NAME:
1074		if (ident_len >= IFNAMSIZ)
1075			goto send_mal_query;
1076		memset(buff, 0, sizeof(buff));
1077		memcpy(buff, &iio->ident.name, ident_len);
1078		dev = dev_get_by_name(net, buff);
1079		break;
1080	case ICMP_EXT_ECHO_CTYPE_INDEX:
1081		if (ident_len != sizeof(iio->ident.ifindex))
1082			goto send_mal_query;
1083		dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1084		break;
1085	case ICMP_EXT_ECHO_CTYPE_ADDR:
1086		if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1087		    ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1088				 iio->ident.addr.ctype3_hdr.addrlen)
1089			goto send_mal_query;
1090		switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1091		case ICMP_AFI_IP:
1092			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1093				goto send_mal_query;
1094			dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1095			break;
1096#if IS_ENABLED(CONFIG_IPV6)
1097		case ICMP_AFI_IP6:
1098			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1099				goto send_mal_query;
1100			dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1101			dev_hold(dev);
1102			break;
1103#endif
1104		default:
1105			goto send_mal_query;
1106		}
1107		break;
1108	default:
1109		goto send_mal_query;
1110	}
1111	if (!dev) {
1112		icmphdr->code = ICMP_EXT_CODE_NO_IF;
1113		return true;
1114	}
1115	/* Fill bits in reply message */
1116	if (dev->flags & IFF_UP)
1117		status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1118	if (__in_dev_get_rcu(dev) && __in_dev_get_rcu(dev)->ifa_list)
 
 
1119		status |= ICMP_EXT_ECHOREPLY_IPV4;
1120	if (!list_empty(&rcu_dereference(dev->ip6_ptr)->addr_list))
 
 
1121		status |= ICMP_EXT_ECHOREPLY_IPV6;
 
1122	dev_put(dev);
1123	icmphdr->un.echo.sequence |= htons(status);
1124	return true;
1125send_mal_query:
1126	icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1127	return true;
1128}
1129EXPORT_SYMBOL_GPL(icmp_build_probe);
1130
1131/*
1132 *	Handle ICMP Timestamp requests.
1133 *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1134 *		  SHOULD be in the kernel for minimum random latency.
1135 *		  MUST be accurate to a few minutes.
1136 *		  MUST be updated at least at 15Hz.
1137 */
1138static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1139{
1140	struct icmp_bxm icmp_param;
1141	/*
1142	 *	Too short.
1143	 */
1144	if (skb->len < 4)
1145		goto out_err;
1146
1147	/*
1148	 *	Fill in the current time as ms since midnight UT:
1149	 */
1150	icmp_param.data.times[1] = inet_current_timestamp();
1151	icmp_param.data.times[2] = icmp_param.data.times[1];
1152
1153	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1154
1155	icmp_param.data.icmph	   = *icmp_hdr(skb);
1156	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1157	icmp_param.data.icmph.code = 0;
1158	icmp_param.skb		   = skb;
1159	icmp_param.offset	   = 0;
1160	icmp_param.data_len	   = 0;
1161	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
1162	icmp_reply(&icmp_param, skb);
1163	return SKB_NOT_DROPPED_YET;
1164
1165out_err:
1166	__ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1167	return SKB_DROP_REASON_PKT_TOO_SMALL;
1168}
1169
1170static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1171{
1172	/* pretend it was a success */
1173	return SKB_NOT_DROPPED_YET;
1174}
1175
1176/*
1177 *	Deal with incoming ICMP packets.
1178 */
1179int icmp_rcv(struct sk_buff *skb)
1180{
1181	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1182	struct rtable *rt = skb_rtable(skb);
1183	struct net *net = dev_net(rt->dst.dev);
1184	struct icmphdr *icmph;
1185
1186	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1187		struct sec_path *sp = skb_sec_path(skb);
1188		int nh;
1189
1190		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1191				 XFRM_STATE_ICMP)) {
1192			reason = SKB_DROP_REASON_XFRM_POLICY;
1193			goto drop;
1194		}
1195
1196		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1197			goto drop;
1198
1199		nh = skb_network_offset(skb);
1200		skb_set_network_header(skb, sizeof(*icmph));
1201
1202		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1203						skb)) {
1204			reason = SKB_DROP_REASON_XFRM_POLICY;
1205			goto drop;
1206		}
1207
1208		skb_set_network_header(skb, nh);
1209	}
1210
1211	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1212
1213	if (skb_checksum_simple_validate(skb))
1214		goto csum_error;
1215
1216	if (!pskb_pull(skb, sizeof(*icmph)))
1217		goto error;
1218
1219	icmph = icmp_hdr(skb);
1220
1221	ICMPMSGIN_INC_STATS(net, icmph->type);
1222
1223	/* Check for ICMP Extended Echo (PROBE) messages */
1224	if (icmph->type == ICMP_EXT_ECHO) {
1225		/* We can't use icmp_pointers[].handler() because it is an array of
1226		 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1227		 */
1228		reason = icmp_echo(skb);
1229		goto reason_check;
1230	}
1231
1232	if (icmph->type == ICMP_EXT_ECHOREPLY) {
1233		reason = ping_rcv(skb);
1234		goto reason_check;
1235	}
1236
1237	/*
1238	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1239	 *
1240	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1241	 *		  discarded.
1242	 */
1243	if (icmph->type > NR_ICMP_TYPES) {
1244		reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1245		goto error;
1246	}
1247
1248	/*
1249	 *	Parse the ICMP message
1250	 */
1251
1252	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1253		/*
1254		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1255		 *	  silently ignored (we let user decide with a sysctl).
1256		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1257		 *	  discarded if to broadcast/multicast.
1258		 */
1259		if ((icmph->type == ICMP_ECHO ||
1260		     icmph->type == ICMP_TIMESTAMP) &&
1261		    READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1262			reason = SKB_DROP_REASON_INVALID_PROTO;
1263			goto error;
1264		}
1265		if (icmph->type != ICMP_ECHO &&
1266		    icmph->type != ICMP_TIMESTAMP &&
1267		    icmph->type != ICMP_ADDRESS &&
1268		    icmph->type != ICMP_ADDRESSREPLY) {
1269			reason = SKB_DROP_REASON_INVALID_PROTO;
1270			goto error;
1271		}
1272	}
1273
1274	reason = icmp_pointers[icmph->type].handler(skb);
1275reason_check:
1276	if (!reason)  {
1277		consume_skb(skb);
1278		return NET_RX_SUCCESS;
1279	}
1280
1281drop:
1282	kfree_skb_reason(skb, reason);
1283	return NET_RX_DROP;
1284csum_error:
1285	reason = SKB_DROP_REASON_ICMP_CSUM;
1286	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1287error:
1288	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1289	goto drop;
1290}
1291
1292static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1293{
1294	struct icmp_extobj_hdr *objh, _objh;
1295	struct icmp_ext_hdr *exth, _exth;
1296	u16 olen;
1297
1298	exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1299	if (!exth)
1300		return false;
1301	if (exth->version != 2)
1302		return true;
1303
1304	if (exth->checksum &&
1305	    csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1306		return false;
1307
1308	off += sizeof(_exth);
1309	while (off < skb->len) {
1310		objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1311		if (!objh)
1312			return false;
1313
1314		olen = ntohs(objh->length);
1315		if (olen < sizeof(_objh))
1316			return false;
1317
1318		off += olen;
1319		if (off > skb->len)
1320			return false;
1321	}
1322
1323	return true;
1324}
1325
1326void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1327			   struct sock_ee_data_rfc4884 *out,
1328			   int thlen, int off)
1329{
1330	int hlen;
1331
1332	/* original datagram headers: end of icmph to payload (skb->data) */
1333	hlen = -skb_transport_offset(skb) - thlen;
1334
1335	/* per rfc 4884: minimal datagram length of 128 bytes */
1336	if (off < 128 || off < hlen)
1337		return;
1338
1339	/* kernel has stripped headers: return payload offset in bytes */
1340	off -= hlen;
1341	if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1342		return;
1343
1344	out->len = off;
1345
1346	if (!ip_icmp_error_rfc4884_validate(skb, off))
1347		out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1348}
1349EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1350
1351int icmp_err(struct sk_buff *skb, u32 info)
1352{
1353	struct iphdr *iph = (struct iphdr *)skb->data;
1354	int offset = iph->ihl<<2;
1355	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
 
1356	int type = icmp_hdr(skb)->type;
1357	int code = icmp_hdr(skb)->code;
1358	struct net *net = dev_net(skb->dev);
1359
1360	/*
1361	 * Use ping_err to handle all icmp errors except those
1362	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1363	 */
1364	if (icmph->type != ICMP_ECHOREPLY) {
1365		ping_err(skb, offset, info);
1366		return 0;
1367	}
1368
1369	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1370		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1371	else if (type == ICMP_REDIRECT)
1372		ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1373
1374	return 0;
1375}
1376
1377/*
1378 *	This table is the definition of how we handle ICMP.
1379 */
1380static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1381	[ICMP_ECHOREPLY] = {
1382		.handler = ping_rcv,
1383	},
1384	[1] = {
1385		.handler = icmp_discard,
1386		.error = 1,
1387	},
1388	[2] = {
1389		.handler = icmp_discard,
1390		.error = 1,
1391	},
1392	[ICMP_DEST_UNREACH] = {
1393		.handler = icmp_unreach,
1394		.error = 1,
1395	},
1396	[ICMP_SOURCE_QUENCH] = {
1397		.handler = icmp_unreach,
1398		.error = 1,
1399	},
1400	[ICMP_REDIRECT] = {
1401		.handler = icmp_redirect,
1402		.error = 1,
1403	},
1404	[6] = {
1405		.handler = icmp_discard,
1406		.error = 1,
1407	},
1408	[7] = {
1409		.handler = icmp_discard,
1410		.error = 1,
1411	},
1412	[ICMP_ECHO] = {
1413		.handler = icmp_echo,
1414	},
1415	[9] = {
1416		.handler = icmp_discard,
1417		.error = 1,
1418	},
1419	[10] = {
1420		.handler = icmp_discard,
1421		.error = 1,
1422	},
1423	[ICMP_TIME_EXCEEDED] = {
1424		.handler = icmp_unreach,
1425		.error = 1,
1426	},
1427	[ICMP_PARAMETERPROB] = {
1428		.handler = icmp_unreach,
1429		.error = 1,
1430	},
1431	[ICMP_TIMESTAMP] = {
1432		.handler = icmp_timestamp,
1433	},
1434	[ICMP_TIMESTAMPREPLY] = {
1435		.handler = icmp_discard,
1436	},
1437	[ICMP_INFO_REQUEST] = {
1438		.handler = icmp_discard,
1439	},
1440	[ICMP_INFO_REPLY] = {
1441		.handler = icmp_discard,
1442	},
1443	[ICMP_ADDRESS] = {
1444		.handler = icmp_discard,
1445	},
1446	[ICMP_ADDRESSREPLY] = {
1447		.handler = icmp_discard,
1448	},
1449};
1450
1451static int __net_init icmp_sk_init(struct net *net)
1452{
1453	/* Control parameters for ECHO replies. */
1454	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1455	net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1456	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1457
1458	/* Control parameter - ignore bogus broadcast responses? */
1459	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1460
1461	/*
1462	 * 	Configurable global rate limit.
1463	 *
1464	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1465	 *	bucket ratemask defines which icmp types are ratelimited by
1466	 *	setting	it's bit position.
1467	 *
1468	 *	default:
1469	 *	dest unreachable (3), source quench (4),
1470	 *	time exceeded (11), parameter problem (12)
1471	 */
1472
1473	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1474	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1475	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
 
 
1476
1477	return 0;
1478}
1479
1480static struct pernet_operations __net_initdata icmp_sk_ops = {
1481       .init = icmp_sk_init,
1482};
1483
1484int __init icmp_init(void)
1485{
1486	int err, i;
1487
1488	for_each_possible_cpu(i) {
1489		struct sock *sk;
1490
1491		err = inet_ctl_sock_create(&sk, PF_INET,
1492					   SOCK_RAW, IPPROTO_ICMP, &init_net);
1493		if (err < 0)
1494			return err;
1495
1496		per_cpu(ipv4_icmp_sk, i) = sk;
1497
1498		/* Enough space for 2 64K ICMP packets, including
1499		 * sk_buff/skb_shared_info struct overhead.
1500		 */
1501		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1502
1503		/*
1504		 * Speedup sock_wfree()
1505		 */
1506		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1507		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1508	}
1509	return register_pernet_subsys(&icmp_sk_ops);
1510}