xref: /linux/net/ipv4/icmp.c (revision b056b4cd)
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 #define CREATE_TRACE_POINTS
97 #include <trace/events/icmp.h>
98 
99 /*
100  *	Build xmit assembly blocks
101  */
102 
103 struct icmp_bxm {
104 	struct sk_buff *skb;
105 	int offset;
106 	int data_len;
107 
108 	struct {
109 		struct icmphdr icmph;
110 		__be32	       times[3];
111 	} data;
112 	int head_len;
113 	struct ip_options_data replyopts;
114 };
115 
116 /* An array of errno for error messages from dest unreach. */
117 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
118 
119 const struct icmp_err icmp_err_convert[] = {
120 	{
121 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
122 		.fatal = 0,
123 	},
124 	{
125 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
126 		.fatal = 0,
127 	},
128 	{
129 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
130 		.fatal = 1,
131 	},
132 	{
133 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
134 		.fatal = 1,
135 	},
136 	{
137 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
138 		.fatal = 0,
139 	},
140 	{
141 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
142 		.fatal = 0,
143 	},
144 	{
145 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
146 		.fatal = 1,
147 	},
148 	{
149 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
150 		.fatal = 1,
151 	},
152 	{
153 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
154 		.fatal = 1,
155 	},
156 	{
157 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
158 		.fatal = 1,
159 	},
160 	{
161 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
162 		.fatal = 1,
163 	},
164 	{
165 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
166 		.fatal = 0,
167 	},
168 	{
169 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
170 		.fatal = 0,
171 	},
172 	{
173 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
174 		.fatal = 1,
175 	},
176 	{
177 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
178 		.fatal = 1,
179 	},
180 	{
181 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
182 		.fatal = 1,
183 	},
184 };
185 EXPORT_SYMBOL(icmp_err_convert);
186 
187 /*
188  *	ICMP control array. This specifies what to do with each ICMP.
189  */
190 
191 struct icmp_control {
192 	enum skb_drop_reason (*handler)(struct sk_buff *skb);
193 	short   error;		/* This ICMP is classed as an error message */
194 };
195 
196 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
197 
198 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
199 
200 /* Called with BH disabled */
201 static inline struct sock *icmp_xmit_lock(struct net *net)
202 {
203 	struct sock *sk;
204 
205 	sk = this_cpu_read(ipv4_icmp_sk);
206 
207 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
208 		/* This can happen if the output path signals a
209 		 * dst_link_failure() for an outgoing ICMP packet.
210 		 */
211 		return NULL;
212 	}
213 	sock_net_set(sk, net);
214 	return sk;
215 }
216 
217 static inline void icmp_xmit_unlock(struct sock *sk)
218 {
219 	sock_net_set(sk, &init_net);
220 	spin_unlock(&sk->sk_lock.slock);
221 }
222 
223 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
224 int sysctl_icmp_msgs_burst __read_mostly = 50;
225 
226 /**
227  * icmp_global_allow - Are we allowed to send one more ICMP message ?
228  * @net: network namespace
229  *
230  * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
231  * Returns false if we reached the limit and can not send another packet.
232  * Works in tandem with icmp_global_consume().
233  */
234 bool icmp_global_allow(struct net *net)
235 {
236 	u32 delta, now, oldstamp;
237 	int incr, new, old;
238 
239 	/* Note: many cpus could find this condition true.
240 	 * Then later icmp_global_consume() could consume more credits,
241 	 * this is an acceptable race.
242 	 */
243 	if (atomic_read(&net->ipv4.icmp_global_credit) > 0)
244 		return true;
245 
246 	now = jiffies;
247 	oldstamp = READ_ONCE(net->ipv4.icmp_global_stamp);
248 	delta = min_t(u32, now - oldstamp, HZ);
249 	if (delta < HZ / 50)
250 		return false;
251 
252 	incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
253 	if (!incr)
254 		return false;
255 
256 	if (cmpxchg(&net->ipv4.icmp_global_stamp, oldstamp, now) == oldstamp) {
257 		old = atomic_read(&net->ipv4.icmp_global_credit);
258 		do {
259 			new = min(old + incr, READ_ONCE(sysctl_icmp_msgs_burst));
260 		} while (!atomic_try_cmpxchg(&net->ipv4.icmp_global_credit, &old, new));
261 	}
262 	return true;
263 }
264 EXPORT_SYMBOL(icmp_global_allow);
265 
266 void icmp_global_consume(struct net *net)
267 {
268 	int credits = get_random_u32_below(3);
269 
270 	/* Note: this might make icmp_global.credit negative. */
271 	if (credits)
272 		atomic_sub(credits, &net->ipv4.icmp_global_credit);
273 }
274 EXPORT_SYMBOL(icmp_global_consume);
275 
276 static bool icmpv4_mask_allow(struct net *net, int type, int code)
277 {
278 	if (type > NR_ICMP_TYPES)
279 		return true;
280 
281 	/* Don't limit PMTU discovery. */
282 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
283 		return true;
284 
285 	/* Limit if icmp type is enabled in ratemask. */
286 	if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
287 		return true;
288 
289 	return false;
290 }
291 
292 static bool icmpv4_global_allow(struct net *net, int type, int code,
293 				bool *apply_ratelimit)
294 {
295 	if (icmpv4_mask_allow(net, type, code))
296 		return true;
297 
298 	if (icmp_global_allow(net)) {
299 		*apply_ratelimit = true;
300 		return true;
301 	}
302 	__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
303 	return false;
304 }
305 
306 /*
307  *	Send an ICMP frame.
308  */
309 
310 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
311 			       struct flowi4 *fl4, int type, int code,
312 			       bool apply_ratelimit)
313 {
314 	struct dst_entry *dst = &rt->dst;
315 	struct inet_peer *peer;
316 	bool rc = true;
317 	int vif;
318 
319 	if (!apply_ratelimit)
320 		return true;
321 
322 	/* No rate limit on loopback */
323 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
324 		goto out;
325 
326 	vif = l3mdev_master_ifindex(dst->dev);
327 	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
328 	rc = inet_peer_xrlim_allow(peer,
329 				   READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
330 	if (peer)
331 		inet_putpeer(peer);
332 out:
333 	if (!rc)
334 		__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
335 	else
336 		icmp_global_consume(net);
337 	return rc;
338 }
339 
340 /*
341  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
342  */
343 void icmp_out_count(struct net *net, unsigned char type)
344 {
345 	ICMPMSGOUT_INC_STATS(net, type);
346 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
347 }
348 
349 /*
350  *	Checksum each fragment, and on the first include the headers and final
351  *	checksum.
352  */
353 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
354 			  struct sk_buff *skb)
355 {
356 	struct icmp_bxm *icmp_param = from;
357 	__wsum csum;
358 
359 	csum = skb_copy_and_csum_bits(icmp_param->skb,
360 				      icmp_param->offset + offset,
361 				      to, len);
362 
363 	skb->csum = csum_block_add(skb->csum, csum, odd);
364 	if (icmp_pointers[icmp_param->data.icmph.type].error)
365 		nf_ct_attach(skb, icmp_param->skb);
366 	return 0;
367 }
368 
369 static void icmp_push_reply(struct sock *sk,
370 			    struct icmp_bxm *icmp_param,
371 			    struct flowi4 *fl4,
372 			    struct ipcm_cookie *ipc, struct rtable **rt)
373 {
374 	struct sk_buff *skb;
375 
376 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
377 			   icmp_param->data_len+icmp_param->head_len,
378 			   icmp_param->head_len,
379 			   ipc, rt, MSG_DONTWAIT) < 0) {
380 		__ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
381 		ip_flush_pending_frames(sk);
382 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
383 		struct icmphdr *icmph = icmp_hdr(skb);
384 		__wsum csum;
385 		struct sk_buff *skb1;
386 
387 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
388 						 (char *)icmph,
389 						 icmp_param->head_len);
390 		skb_queue_walk(&sk->sk_write_queue, skb1) {
391 			csum = csum_add(csum, skb1->csum);
392 		}
393 		icmph->checksum = csum_fold(csum);
394 		skb->ip_summed = CHECKSUM_NONE;
395 		ip_push_pending_frames(sk, fl4);
396 	}
397 }
398 
399 /*
400  *	Driving logic for building and sending ICMP messages.
401  */
402 
403 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
404 {
405 	struct ipcm_cookie ipc;
406 	struct rtable *rt = skb_rtable(skb);
407 	struct net *net = dev_net(rt->dst.dev);
408 	bool apply_ratelimit = false;
409 	struct flowi4 fl4;
410 	struct sock *sk;
411 	struct inet_sock *inet;
412 	__be32 daddr, saddr;
413 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
414 	int type = icmp_param->data.icmph.type;
415 	int code = icmp_param->data.icmph.code;
416 
417 	if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
418 		return;
419 
420 	/* Needed by both icmpv4_global_allow and icmp_xmit_lock */
421 	local_bh_disable();
422 
423 	/* is global icmp_msgs_per_sec exhausted ? */
424 	if (!icmpv4_global_allow(net, type, code, &apply_ratelimit))
425 		goto out_bh_enable;
426 
427 	sk = icmp_xmit_lock(net);
428 	if (!sk)
429 		goto out_bh_enable;
430 	inet = inet_sk(sk);
431 
432 	icmp_param->data.icmph.checksum = 0;
433 
434 	ipcm_init(&ipc);
435 	inet->tos = ip_hdr(skb)->tos;
436 	ipc.sockc.mark = mark;
437 	daddr = ipc.addr = ip_hdr(skb)->saddr;
438 	saddr = fib_compute_spec_dst(skb);
439 
440 	if (icmp_param->replyopts.opt.opt.optlen) {
441 		ipc.opt = &icmp_param->replyopts.opt;
442 		if (ipc.opt->opt.srr)
443 			daddr = icmp_param->replyopts.opt.opt.faddr;
444 	}
445 	memset(&fl4, 0, sizeof(fl4));
446 	fl4.daddr = daddr;
447 	fl4.saddr = saddr;
448 	fl4.flowi4_mark = mark;
449 	fl4.flowi4_uid = sock_net_uid(net, NULL);
450 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
451 	fl4.flowi4_proto = IPPROTO_ICMP;
452 	fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
453 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
454 	rt = ip_route_output_key(net, &fl4);
455 	if (IS_ERR(rt))
456 		goto out_unlock;
457 	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
458 		icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
459 	ip_rt_put(rt);
460 out_unlock:
461 	icmp_xmit_unlock(sk);
462 out_bh_enable:
463 	local_bh_enable();
464 }
465 
466 /*
467  * The device used for looking up which routing table to use for sending an ICMP
468  * error is preferably the source whenever it is set, which should ensure the
469  * icmp error can be sent to the source host, else lookup using the routing
470  * table of the destination device, else use the main routing table (index 0).
471  */
472 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
473 {
474 	struct net_device *route_lookup_dev = NULL;
475 
476 	if (skb->dev)
477 		route_lookup_dev = skb->dev;
478 	else if (skb_dst(skb))
479 		route_lookup_dev = skb_dst(skb)->dev;
480 	return route_lookup_dev;
481 }
482 
483 static struct rtable *icmp_route_lookup(struct net *net,
484 					struct flowi4 *fl4,
485 					struct sk_buff *skb_in,
486 					const struct iphdr *iph,
487 					__be32 saddr, u8 tos, u32 mark,
488 					int type, int code,
489 					struct icmp_bxm *param)
490 {
491 	struct net_device *route_lookup_dev;
492 	struct dst_entry *dst, *dst2;
493 	struct rtable *rt, *rt2;
494 	struct flowi4 fl4_dec;
495 	int err;
496 
497 	memset(fl4, 0, sizeof(*fl4));
498 	fl4->daddr = (param->replyopts.opt.opt.srr ?
499 		      param->replyopts.opt.opt.faddr : iph->saddr);
500 	fl4->saddr = saddr;
501 	fl4->flowi4_mark = mark;
502 	fl4->flowi4_uid = sock_net_uid(net, NULL);
503 	fl4->flowi4_tos = RT_TOS(tos);
504 	fl4->flowi4_proto = IPPROTO_ICMP;
505 	fl4->fl4_icmp_type = type;
506 	fl4->fl4_icmp_code = code;
507 	route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
508 	fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
509 
510 	security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
511 	rt = ip_route_output_key_hash(net, fl4, skb_in);
512 	if (IS_ERR(rt))
513 		return rt;
514 
515 	/* No need to clone since we're just using its address. */
516 	rt2 = rt;
517 
518 	dst = xfrm_lookup(net, &rt->dst,
519 			  flowi4_to_flowi(fl4), NULL, 0);
520 	rt = dst_rtable(dst);
521 	if (!IS_ERR(dst)) {
522 		if (rt != rt2)
523 			return rt;
524 	} else if (PTR_ERR(dst) == -EPERM) {
525 		rt = NULL;
526 	} else {
527 		return rt;
528 	}
529 	err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
530 	if (err)
531 		goto relookup_failed;
532 
533 	if (inet_addr_type_dev_table(net, route_lookup_dev,
534 				     fl4_dec.saddr) == RTN_LOCAL) {
535 		rt2 = __ip_route_output_key(net, &fl4_dec);
536 		if (IS_ERR(rt2))
537 			err = PTR_ERR(rt2);
538 	} else {
539 		struct flowi4 fl4_2 = {};
540 		unsigned long orefdst;
541 
542 		fl4_2.daddr = fl4_dec.saddr;
543 		rt2 = ip_route_output_key(net, &fl4_2);
544 		if (IS_ERR(rt2)) {
545 			err = PTR_ERR(rt2);
546 			goto relookup_failed;
547 		}
548 		/* Ugh! */
549 		orefdst = skb_in->_skb_refdst; /* save old refdst */
550 		skb_dst_set(skb_in, NULL);
551 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
552 				     tos, rt2->dst.dev);
553 
554 		dst_release(&rt2->dst);
555 		rt2 = skb_rtable(skb_in);
556 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
557 	}
558 
559 	if (err)
560 		goto relookup_failed;
561 
562 	dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL,
563 			   XFRM_LOOKUP_ICMP);
564 	rt2 = dst_rtable(dst2);
565 	if (!IS_ERR(dst2)) {
566 		dst_release(&rt->dst);
567 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
568 		rt = rt2;
569 	} else if (PTR_ERR(dst2) == -EPERM) {
570 		if (rt)
571 			dst_release(&rt->dst);
572 		return rt2;
573 	} else {
574 		err = PTR_ERR(dst2);
575 		goto relookup_failed;
576 	}
577 	return rt;
578 
579 relookup_failed:
580 	if (rt)
581 		return rt;
582 	return ERR_PTR(err);
583 }
584 
585 /*
586  *	Send an ICMP message in response to a situation
587  *
588  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
589  *		  MAY send more (we do).
590  *			MUST NOT change this header information.
591  *			MUST NOT reply to a multicast/broadcast IP address.
592  *			MUST NOT reply to a multicast/broadcast MAC address.
593  *			MUST reply to only the first fragment.
594  */
595 
596 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
597 		 const struct ip_options *opt)
598 {
599 	struct iphdr *iph;
600 	int room;
601 	struct icmp_bxm icmp_param;
602 	struct rtable *rt = skb_rtable(skb_in);
603 	bool apply_ratelimit = false;
604 	struct ipcm_cookie ipc;
605 	struct flowi4 fl4;
606 	__be32 saddr;
607 	u8  tos;
608 	u32 mark;
609 	struct net *net;
610 	struct sock *sk;
611 
612 	if (!rt)
613 		goto out;
614 
615 	if (rt->dst.dev)
616 		net = dev_net(rt->dst.dev);
617 	else if (skb_in->dev)
618 		net = dev_net(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, tos, mark,
747 			       type, code, &icmp_param);
748 	if (IS_ERR(rt))
749 		goto out_unlock;
750 
751 	/* peer icmp_ratelimit */
752 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
753 		goto ende;
754 
755 	/* RFC says return as much as we can without exceeding 576 bytes. */
756 
757 	room = dst_mtu(&rt->dst);
758 	if (room > 576)
759 		room = 576;
760 	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
761 	room -= sizeof(struct icmphdr);
762 	/* Guard against tiny mtu. We need to include at least one
763 	 * IP network header for this message to make any sense.
764 	 */
765 	if (room <= (int)sizeof(struct iphdr))
766 		goto ende;
767 
768 	icmp_param.data_len = skb_in->len - icmp_param.offset;
769 	if (icmp_param.data_len > room)
770 		icmp_param.data_len = room;
771 	icmp_param.head_len = sizeof(struct icmphdr);
772 
773 	/* if we don't have a source address at this point, fall back to the
774 	 * dummy address instead of sending out a packet with a source address
775 	 * of 0.0.0.0
776 	 */
777 	if (!fl4.saddr)
778 		fl4.saddr = htonl(INADDR_DUMMY);
779 
780 	trace_icmp_send(skb_in, type, code);
781 
782 	icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
783 ende:
784 	ip_rt_put(rt);
785 out_unlock:
786 	icmp_xmit_unlock(sk);
787 out_bh_enable:
788 	local_bh_enable();
789 out:;
790 }
791 EXPORT_SYMBOL(__icmp_send);
792 
793 #if IS_ENABLED(CONFIG_NF_NAT)
794 #include <net/netfilter/nf_conntrack.h>
795 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
796 {
797 	struct sk_buff *cloned_skb = NULL;
798 	struct ip_options opts = { 0 };
799 	enum ip_conntrack_info ctinfo;
800 	struct nf_conn *ct;
801 	__be32 orig_ip;
802 
803 	ct = nf_ct_get(skb_in, &ctinfo);
804 	if (!ct || !(ct->status & IPS_SRC_NAT)) {
805 		__icmp_send(skb_in, type, code, info, &opts);
806 		return;
807 	}
808 
809 	if (skb_shared(skb_in))
810 		skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
811 
812 	if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
813 	    (skb_network_header(skb_in) + sizeof(struct iphdr)) >
814 	    skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
815 	    skb_network_offset(skb_in) + sizeof(struct iphdr))))
816 		goto out;
817 
818 	orig_ip = ip_hdr(skb_in)->saddr;
819 	ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
820 	__icmp_send(skb_in, type, code, info, &opts);
821 	ip_hdr(skb_in)->saddr = orig_ip;
822 out:
823 	consume_skb(cloned_skb);
824 }
825 EXPORT_SYMBOL(icmp_ndo_send);
826 #endif
827 
828 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
829 {
830 	const struct iphdr *iph = (const struct iphdr *)skb->data;
831 	const struct net_protocol *ipprot;
832 	int protocol = iph->protocol;
833 
834 	/* Checkin full IP header plus 8 bytes of protocol to
835 	 * avoid additional coding at protocol handlers.
836 	 */
837 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
838 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
839 		return;
840 	}
841 
842 	raw_icmp_error(skb, protocol, info);
843 
844 	ipprot = rcu_dereference(inet_protos[protocol]);
845 	if (ipprot && ipprot->err_handler)
846 		ipprot->err_handler(skb, info);
847 }
848 
849 static bool icmp_tag_validation(int proto)
850 {
851 	bool ok;
852 
853 	rcu_read_lock();
854 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
855 	rcu_read_unlock();
856 	return ok;
857 }
858 
859 /*
860  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
861  *	ICMP_PARAMETERPROB.
862  */
863 
864 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
865 {
866 	enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
867 	const struct iphdr *iph;
868 	struct icmphdr *icmph;
869 	struct net *net;
870 	u32 info = 0;
871 
872 	net = dev_net(skb_dst(skb)->dev);
873 
874 	/*
875 	 *	Incomplete header ?
876 	 * 	Only checks for the IP header, there should be an
877 	 *	additional check for longer headers in upper levels.
878 	 */
879 
880 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
881 		goto out_err;
882 
883 	icmph = icmp_hdr(skb);
884 	iph   = (const struct iphdr *)skb->data;
885 
886 	if (iph->ihl < 5)  { /* Mangled header, drop. */
887 		reason = SKB_DROP_REASON_IP_INHDR;
888 		goto out_err;
889 	}
890 
891 	switch (icmph->type) {
892 	case ICMP_DEST_UNREACH:
893 		switch (icmph->code & 15) {
894 		case ICMP_NET_UNREACH:
895 		case ICMP_HOST_UNREACH:
896 		case ICMP_PROT_UNREACH:
897 		case ICMP_PORT_UNREACH:
898 			break;
899 		case ICMP_FRAG_NEEDED:
900 			/* for documentation of the ip_no_pmtu_disc
901 			 * values please see
902 			 * Documentation/networking/ip-sysctl.rst
903 			 */
904 			switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
905 			default:
906 				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
907 						    &iph->daddr);
908 				break;
909 			case 2:
910 				goto out;
911 			case 3:
912 				if (!icmp_tag_validation(iph->protocol))
913 					goto out;
914 				fallthrough;
915 			case 0:
916 				info = ntohs(icmph->un.frag.mtu);
917 			}
918 			break;
919 		case ICMP_SR_FAILED:
920 			net_dbg_ratelimited("%pI4: Source Route Failed\n",
921 					    &iph->daddr);
922 			break;
923 		default:
924 			break;
925 		}
926 		if (icmph->code > NR_ICMP_UNREACH)
927 			goto out;
928 		break;
929 	case ICMP_PARAMETERPROB:
930 		info = ntohl(icmph->un.gateway) >> 24;
931 		break;
932 	case ICMP_TIME_EXCEEDED:
933 		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
934 		if (icmph->code == ICMP_EXC_FRAGTIME)
935 			goto out;
936 		break;
937 	}
938 
939 	/*
940 	 *	Throw it at our lower layers
941 	 *
942 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
943 	 *		  header.
944 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
945 	 *		  transport layer.
946 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
947 	 *		  transport layer.
948 	 */
949 
950 	/*
951 	 *	Check the other end isn't violating RFC 1122. Some routers send
952 	 *	bogus responses to broadcast frames. If you see this message
953 	 *	first check your netmask matches at both ends, if it does then
954 	 *	get the other vendor to fix their kit.
955 	 */
956 
957 	if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
958 	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
959 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
960 				     &ip_hdr(skb)->saddr,
961 				     icmph->type, icmph->code,
962 				     &iph->daddr, skb->dev->name);
963 		goto out;
964 	}
965 
966 	icmp_socket_deliver(skb, info);
967 
968 out:
969 	return reason;
970 out_err:
971 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
972 	return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
973 }
974 
975 
976 /*
977  *	Handle ICMP_REDIRECT.
978  */
979 
980 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
981 {
982 	if (skb->len < sizeof(struct iphdr)) {
983 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
984 		return SKB_DROP_REASON_PKT_TOO_SMALL;
985 	}
986 
987 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
988 		/* there aught to be a stat */
989 		return SKB_DROP_REASON_NOMEM;
990 	}
991 
992 	icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
993 	return SKB_NOT_DROPPED_YET;
994 }
995 
996 /*
997  *	Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
998  *
999  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
1000  *		  requests.
1001  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
1002  *		  included in the reply.
1003  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
1004  *		  echo requests, MUST have default=NOT.
1005  *	RFC 8335: 8 MUST have a config option to enable/disable ICMP
1006  *		  Extended Echo Functionality, MUST be disabled by default
1007  *	See also WRT handling of options once they are done and working.
1008  */
1009 
1010 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1011 {
1012 	struct icmp_bxm icmp_param;
1013 	struct net *net;
1014 
1015 	net = dev_net(skb_dst(skb)->dev);
1016 	/* should there be an ICMP stat for ignored echos? */
1017 	if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1018 		return SKB_NOT_DROPPED_YET;
1019 
1020 	icmp_param.data.icmph	   = *icmp_hdr(skb);
1021 	icmp_param.skb		   = skb;
1022 	icmp_param.offset	   = 0;
1023 	icmp_param.data_len	   = skb->len;
1024 	icmp_param.head_len	   = sizeof(struct icmphdr);
1025 
1026 	if (icmp_param.data.icmph.type == ICMP_ECHO)
1027 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1028 	else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1029 		return SKB_NOT_DROPPED_YET;
1030 
1031 	icmp_reply(&icmp_param, skb);
1032 	return SKB_NOT_DROPPED_YET;
1033 }
1034 
1035 /*	Helper for icmp_echo and icmpv6_echo_reply.
1036  *	Searches for net_device that matches PROBE interface identifier
1037  *		and builds PROBE reply message in icmphdr.
1038  *
1039  *	Returns false if PROBE responses are disabled via sysctl
1040  */
1041 
1042 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1043 {
1044 	struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1045 	struct icmp_ext_echo_iio *iio, _iio;
1046 	struct net *net = dev_net(skb->dev);
1047 	struct inet6_dev *in6_dev;
1048 	struct in_device *in_dev;
1049 	struct net_device *dev;
1050 	char buff[IFNAMSIZ];
1051 	u16 ident_len;
1052 	u8 status;
1053 
1054 	if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1055 		return false;
1056 
1057 	/* We currently only support probing interfaces on the proxy node
1058 	 * Check to ensure L-bit is set
1059 	 */
1060 	if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1061 		return false;
1062 	/* Clear status bits in reply message */
1063 	icmphdr->un.echo.sequence &= htons(0xFF00);
1064 	if (icmphdr->type == ICMP_EXT_ECHO)
1065 		icmphdr->type = ICMP_EXT_ECHOREPLY;
1066 	else
1067 		icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1068 	ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1069 	/* Size of iio is class_type dependent.
1070 	 * Only check header here and assign length based on ctype in the switch statement
1071 	 */
1072 	iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1073 	if (!ext_hdr || !iio)
1074 		goto send_mal_query;
1075 	if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1076 	    ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1077 		goto send_mal_query;
1078 	ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1079 	iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1080 				 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1081 	if (!iio)
1082 		goto send_mal_query;
1083 
1084 	status = 0;
1085 	dev = NULL;
1086 	switch (iio->extobj_hdr.class_type) {
1087 	case ICMP_EXT_ECHO_CTYPE_NAME:
1088 		if (ident_len >= IFNAMSIZ)
1089 			goto send_mal_query;
1090 		memset(buff, 0, sizeof(buff));
1091 		memcpy(buff, &iio->ident.name, ident_len);
1092 		dev = dev_get_by_name(net, buff);
1093 		break;
1094 	case ICMP_EXT_ECHO_CTYPE_INDEX:
1095 		if (ident_len != sizeof(iio->ident.ifindex))
1096 			goto send_mal_query;
1097 		dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1098 		break;
1099 	case ICMP_EXT_ECHO_CTYPE_ADDR:
1100 		if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1101 		    ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1102 				 iio->ident.addr.ctype3_hdr.addrlen)
1103 			goto send_mal_query;
1104 		switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1105 		case ICMP_AFI_IP:
1106 			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1107 				goto send_mal_query;
1108 			dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1109 			break;
1110 #if IS_ENABLED(CONFIG_IPV6)
1111 		case ICMP_AFI_IP6:
1112 			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1113 				goto send_mal_query;
1114 			dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1115 			dev_hold(dev);
1116 			break;
1117 #endif
1118 		default:
1119 			goto send_mal_query;
1120 		}
1121 		break;
1122 	default:
1123 		goto send_mal_query;
1124 	}
1125 	if (!dev) {
1126 		icmphdr->code = ICMP_EXT_CODE_NO_IF;
1127 		return true;
1128 	}
1129 	/* Fill bits in reply message */
1130 	if (dev->flags & IFF_UP)
1131 		status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1132 
1133 	in_dev = __in_dev_get_rcu(dev);
1134 	if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1135 		status |= ICMP_EXT_ECHOREPLY_IPV4;
1136 
1137 	in6_dev = __in6_dev_get(dev);
1138 	if (in6_dev && !list_empty(&in6_dev->addr_list))
1139 		status |= ICMP_EXT_ECHOREPLY_IPV6;
1140 
1141 	dev_put(dev);
1142 	icmphdr->un.echo.sequence |= htons(status);
1143 	return true;
1144 send_mal_query:
1145 	icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1146 	return true;
1147 }
1148 EXPORT_SYMBOL_GPL(icmp_build_probe);
1149 
1150 /*
1151  *	Handle ICMP Timestamp requests.
1152  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1153  *		  SHOULD be in the kernel for minimum random latency.
1154  *		  MUST be accurate to a few minutes.
1155  *		  MUST be updated at least at 15Hz.
1156  */
1157 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1158 {
1159 	struct icmp_bxm icmp_param;
1160 	/*
1161 	 *	Too short.
1162 	 */
1163 	if (skb->len < 4)
1164 		goto out_err;
1165 
1166 	/*
1167 	 *	Fill in the current time as ms since midnight UT:
1168 	 */
1169 	icmp_param.data.times[1] = inet_current_timestamp();
1170 	icmp_param.data.times[2] = icmp_param.data.times[1];
1171 
1172 	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1173 
1174 	icmp_param.data.icmph	   = *icmp_hdr(skb);
1175 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1176 	icmp_param.data.icmph.code = 0;
1177 	icmp_param.skb		   = skb;
1178 	icmp_param.offset	   = 0;
1179 	icmp_param.data_len	   = 0;
1180 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
1181 	icmp_reply(&icmp_param, skb);
1182 	return SKB_NOT_DROPPED_YET;
1183 
1184 out_err:
1185 	__ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1186 	return SKB_DROP_REASON_PKT_TOO_SMALL;
1187 }
1188 
1189 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1190 {
1191 	/* pretend it was a success */
1192 	return SKB_NOT_DROPPED_YET;
1193 }
1194 
1195 /*
1196  *	Deal with incoming ICMP packets.
1197  */
1198 int icmp_rcv(struct sk_buff *skb)
1199 {
1200 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1201 	struct rtable *rt = skb_rtable(skb);
1202 	struct net *net = dev_net(rt->dst.dev);
1203 	struct icmphdr *icmph;
1204 
1205 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1206 		struct sec_path *sp = skb_sec_path(skb);
1207 		int nh;
1208 
1209 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1210 				 XFRM_STATE_ICMP)) {
1211 			reason = SKB_DROP_REASON_XFRM_POLICY;
1212 			goto drop;
1213 		}
1214 
1215 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1216 			goto drop;
1217 
1218 		nh = skb_network_offset(skb);
1219 		skb_set_network_header(skb, sizeof(*icmph));
1220 
1221 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1222 						skb)) {
1223 			reason = SKB_DROP_REASON_XFRM_POLICY;
1224 			goto drop;
1225 		}
1226 
1227 		skb_set_network_header(skb, nh);
1228 	}
1229 
1230 	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1231 
1232 	if (skb_checksum_simple_validate(skb))
1233 		goto csum_error;
1234 
1235 	if (!pskb_pull(skb, sizeof(*icmph)))
1236 		goto error;
1237 
1238 	icmph = icmp_hdr(skb);
1239 
1240 	ICMPMSGIN_INC_STATS(net, icmph->type);
1241 
1242 	/* Check for ICMP Extended Echo (PROBE) messages */
1243 	if (icmph->type == ICMP_EXT_ECHO) {
1244 		/* We can't use icmp_pointers[].handler() because it is an array of
1245 		 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1246 		 */
1247 		reason = icmp_echo(skb);
1248 		goto reason_check;
1249 	}
1250 
1251 	if (icmph->type == ICMP_EXT_ECHOREPLY) {
1252 		reason = ping_rcv(skb);
1253 		goto reason_check;
1254 	}
1255 
1256 	/*
1257 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1258 	 *
1259 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1260 	 *		  discarded.
1261 	 */
1262 	if (icmph->type > NR_ICMP_TYPES) {
1263 		reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1264 		goto error;
1265 	}
1266 
1267 	/*
1268 	 *	Parse the ICMP message
1269 	 */
1270 
1271 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1272 		/*
1273 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1274 		 *	  silently ignored (we let user decide with a sysctl).
1275 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1276 		 *	  discarded if to broadcast/multicast.
1277 		 */
1278 		if ((icmph->type == ICMP_ECHO ||
1279 		     icmph->type == ICMP_TIMESTAMP) &&
1280 		    READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1281 			reason = SKB_DROP_REASON_INVALID_PROTO;
1282 			goto error;
1283 		}
1284 		if (icmph->type != ICMP_ECHO &&
1285 		    icmph->type != ICMP_TIMESTAMP &&
1286 		    icmph->type != ICMP_ADDRESS &&
1287 		    icmph->type != ICMP_ADDRESSREPLY) {
1288 			reason = SKB_DROP_REASON_INVALID_PROTO;
1289 			goto error;
1290 		}
1291 	}
1292 
1293 	reason = icmp_pointers[icmph->type].handler(skb);
1294 reason_check:
1295 	if (!reason)  {
1296 		consume_skb(skb);
1297 		return NET_RX_SUCCESS;
1298 	}
1299 
1300 drop:
1301 	kfree_skb_reason(skb, reason);
1302 	return NET_RX_DROP;
1303 csum_error:
1304 	reason = SKB_DROP_REASON_ICMP_CSUM;
1305 	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1306 error:
1307 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1308 	goto drop;
1309 }
1310 
1311 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1312 {
1313 	struct icmp_extobj_hdr *objh, _objh;
1314 	struct icmp_ext_hdr *exth, _exth;
1315 	u16 olen;
1316 
1317 	exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1318 	if (!exth)
1319 		return false;
1320 	if (exth->version != 2)
1321 		return true;
1322 
1323 	if (exth->checksum &&
1324 	    csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1325 		return false;
1326 
1327 	off += sizeof(_exth);
1328 	while (off < skb->len) {
1329 		objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1330 		if (!objh)
1331 			return false;
1332 
1333 		olen = ntohs(objh->length);
1334 		if (olen < sizeof(_objh))
1335 			return false;
1336 
1337 		off += olen;
1338 		if (off > skb->len)
1339 			return false;
1340 	}
1341 
1342 	return true;
1343 }
1344 
1345 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1346 			   struct sock_ee_data_rfc4884 *out,
1347 			   int thlen, int off)
1348 {
1349 	int hlen;
1350 
1351 	/* original datagram headers: end of icmph to payload (skb->data) */
1352 	hlen = -skb_transport_offset(skb) - thlen;
1353 
1354 	/* per rfc 4884: minimal datagram length of 128 bytes */
1355 	if (off < 128 || off < hlen)
1356 		return;
1357 
1358 	/* kernel has stripped headers: return payload offset in bytes */
1359 	off -= hlen;
1360 	if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1361 		return;
1362 
1363 	out->len = off;
1364 
1365 	if (!ip_icmp_error_rfc4884_validate(skb, off))
1366 		out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1367 }
1368 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1369 
1370 int icmp_err(struct sk_buff *skb, u32 info)
1371 {
1372 	struct iphdr *iph = (struct iphdr *)skb->data;
1373 	int offset = iph->ihl<<2;
1374 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1375 	int type = icmp_hdr(skb)->type;
1376 	int code = icmp_hdr(skb)->code;
1377 	struct net *net = dev_net(skb->dev);
1378 
1379 	/*
1380 	 * Use ping_err to handle all icmp errors except those
1381 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1382 	 */
1383 	if (icmph->type != ICMP_ECHOREPLY) {
1384 		ping_err(skb, offset, info);
1385 		return 0;
1386 	}
1387 
1388 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1389 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1390 	else if (type == ICMP_REDIRECT)
1391 		ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1392 
1393 	return 0;
1394 }
1395 
1396 /*
1397  *	This table is the definition of how we handle ICMP.
1398  */
1399 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1400 	[ICMP_ECHOREPLY] = {
1401 		.handler = ping_rcv,
1402 	},
1403 	[1] = {
1404 		.handler = icmp_discard,
1405 		.error = 1,
1406 	},
1407 	[2] = {
1408 		.handler = icmp_discard,
1409 		.error = 1,
1410 	},
1411 	[ICMP_DEST_UNREACH] = {
1412 		.handler = icmp_unreach,
1413 		.error = 1,
1414 	},
1415 	[ICMP_SOURCE_QUENCH] = {
1416 		.handler = icmp_unreach,
1417 		.error = 1,
1418 	},
1419 	[ICMP_REDIRECT] = {
1420 		.handler = icmp_redirect,
1421 		.error = 1,
1422 	},
1423 	[6] = {
1424 		.handler = icmp_discard,
1425 		.error = 1,
1426 	},
1427 	[7] = {
1428 		.handler = icmp_discard,
1429 		.error = 1,
1430 	},
1431 	[ICMP_ECHO] = {
1432 		.handler = icmp_echo,
1433 	},
1434 	[9] = {
1435 		.handler = icmp_discard,
1436 		.error = 1,
1437 	},
1438 	[10] = {
1439 		.handler = icmp_discard,
1440 		.error = 1,
1441 	},
1442 	[ICMP_TIME_EXCEEDED] = {
1443 		.handler = icmp_unreach,
1444 		.error = 1,
1445 	},
1446 	[ICMP_PARAMETERPROB] = {
1447 		.handler = icmp_unreach,
1448 		.error = 1,
1449 	},
1450 	[ICMP_TIMESTAMP] = {
1451 		.handler = icmp_timestamp,
1452 	},
1453 	[ICMP_TIMESTAMPREPLY] = {
1454 		.handler = icmp_discard,
1455 	},
1456 	[ICMP_INFO_REQUEST] = {
1457 		.handler = icmp_discard,
1458 	},
1459 	[ICMP_INFO_REPLY] = {
1460 		.handler = icmp_discard,
1461 	},
1462 	[ICMP_ADDRESS] = {
1463 		.handler = icmp_discard,
1464 	},
1465 	[ICMP_ADDRESSREPLY] = {
1466 		.handler = icmp_discard,
1467 	},
1468 };
1469 
1470 static int __net_init icmp_sk_init(struct net *net)
1471 {
1472 	/* Control parameters for ECHO replies. */
1473 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1474 	net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1475 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1476 
1477 	/* Control parameter - ignore bogus broadcast responses? */
1478 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1479 
1480 	/*
1481 	 * 	Configurable global rate limit.
1482 	 *
1483 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1484 	 *	bucket ratemask defines which icmp types are ratelimited by
1485 	 *	setting	it's bit position.
1486 	 *
1487 	 *	default:
1488 	 *	dest unreachable (3), source quench (4),
1489 	 *	time exceeded (11), parameter problem (12)
1490 	 */
1491 
1492 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1493 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1494 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1495 
1496 	return 0;
1497 }
1498 
1499 static struct pernet_operations __net_initdata icmp_sk_ops = {
1500        .init = icmp_sk_init,
1501 };
1502 
1503 int __init icmp_init(void)
1504 {
1505 	int err, i;
1506 
1507 	for_each_possible_cpu(i) {
1508 		struct sock *sk;
1509 
1510 		err = inet_ctl_sock_create(&sk, PF_INET,
1511 					   SOCK_RAW, IPPROTO_ICMP, &init_net);
1512 		if (err < 0)
1513 			return err;
1514 
1515 		per_cpu(ipv4_icmp_sk, i) = sk;
1516 
1517 		/* Enough space for 2 64K ICMP packets, including
1518 		 * sk_buff/skb_shared_info struct overhead.
1519 		 */
1520 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1521 
1522 		/*
1523 		 * Speedup sock_wfree()
1524 		 */
1525 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1526 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1527 	}
1528 	return register_pernet_subsys(&icmp_sk_ops);
1529 }
1530