xref: /openbmc/linux/net/ipv4/icmp.c (revision 6fcd6fea)
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 
97 /*
98  *	Build xmit assembly blocks
99  */
100 
101 struct icmp_bxm {
102 	struct sk_buff *skb;
103 	int offset;
104 	int data_len;
105 
106 	struct {
107 		struct icmphdr icmph;
108 		__be32	       times[3];
109 	} data;
110 	int head_len;
111 	struct ip_options_data replyopts;
112 };
113 
114 /* An array of errno for error messages from dest unreach. */
115 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
116 
117 const struct icmp_err icmp_err_convert[] = {
118 	{
119 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
120 		.fatal = 0,
121 	},
122 	{
123 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
124 		.fatal = 0,
125 	},
126 	{
127 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
128 		.fatal = 1,
129 	},
130 	{
131 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
132 		.fatal = 1,
133 	},
134 	{
135 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
136 		.fatal = 0,
137 	},
138 	{
139 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
140 		.fatal = 0,
141 	},
142 	{
143 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
144 		.fatal = 1,
145 	},
146 	{
147 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
148 		.fatal = 1,
149 	},
150 	{
151 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
152 		.fatal = 1,
153 	},
154 	{
155 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
156 		.fatal = 1,
157 	},
158 	{
159 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
160 		.fatal = 1,
161 	},
162 	{
163 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
164 		.fatal = 0,
165 	},
166 	{
167 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
168 		.fatal = 0,
169 	},
170 	{
171 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
172 		.fatal = 1,
173 	},
174 	{
175 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
176 		.fatal = 1,
177 	},
178 	{
179 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
180 		.fatal = 1,
181 	},
182 };
183 EXPORT_SYMBOL(icmp_err_convert);
184 
185 /*
186  *	ICMP control array. This specifies what to do with each ICMP.
187  */
188 
189 struct icmp_control {
190 	enum skb_drop_reason (*handler)(struct sk_buff *skb);
191 	short   error;		/* This ICMP is classed as an error message */
192 };
193 
194 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
195 
196 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
197 
198 /* Called with BH disabled */
199 static inline struct sock *icmp_xmit_lock(struct net *net)
200 {
201 	struct sock *sk;
202 
203 	sk = this_cpu_read(ipv4_icmp_sk);
204 
205 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
206 		/* This can happen if the output path signals a
207 		 * dst_link_failure() for an outgoing ICMP packet.
208 		 */
209 		return NULL;
210 	}
211 	sock_net_set(sk, net);
212 	return sk;
213 }
214 
215 static inline void icmp_xmit_unlock(struct sock *sk)
216 {
217 	sock_net_set(sk, &init_net);
218 	spin_unlock(&sk->sk_lock.slock);
219 }
220 
221 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
222 int sysctl_icmp_msgs_burst __read_mostly = 50;
223 
224 static struct {
225 	atomic_t	credit;
226 	u32		stamp;
227 } icmp_global;
228 
229 /**
230  * icmp_global_allow - Are we allowed to send one more ICMP message ?
231  *
232  * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
233  * Returns false if we reached the limit and can not send another packet.
234  * Works in tandem with icmp_global_consume().
235  */
236 bool icmp_global_allow(void)
237 {
238 	u32 delta, now, oldstamp;
239 	int incr, new, old;
240 
241 	/* Note: many cpus could find this condition true.
242 	 * Then later icmp_global_consume() could consume more credits,
243 	 * this is an acceptable race.
244 	 */
245 	if (atomic_read(&icmp_global.credit) > 0)
246 		return true;
247 
248 	now = jiffies;
249 	oldstamp = READ_ONCE(icmp_global.stamp);
250 	delta = min_t(u32, now - oldstamp, HZ);
251 	if (delta < HZ / 50)
252 		return false;
253 
254 	incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
255 	if (!incr)
256 		return false;
257 
258 	if (cmpxchg(&icmp_global.stamp, oldstamp, now) == oldstamp) {
259 		old = atomic_read(&icmp_global.credit);
260 		do {
261 			new = min(old + incr, READ_ONCE(sysctl_icmp_msgs_burst));
262 		} while (!atomic_try_cmpxchg(&icmp_global.credit, &old, new));
263 	}
264 	return true;
265 }
266 EXPORT_SYMBOL(icmp_global_allow);
267 
268 void icmp_global_consume(void)
269 {
270 	int credits = get_random_u32_below(3);
271 
272 	/* Note: this might make icmp_global.credit negative. */
273 	if (credits)
274 		atomic_sub(credits, &icmp_global.credit);
275 }
276 EXPORT_SYMBOL(icmp_global_consume);
277 
278 static bool icmpv4_mask_allow(struct net *net, int type, int code)
279 {
280 	if (type > NR_ICMP_TYPES)
281 		return true;
282 
283 	/* Don't limit PMTU discovery. */
284 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
285 		return true;
286 
287 	/* Limit if icmp type is enabled in ratemask. */
288 	if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
289 		return true;
290 
291 	return false;
292 }
293 
294 static bool icmpv4_global_allow(struct net *net, int type, int code,
295 				bool *apply_ratelimit)
296 {
297 	if (icmpv4_mask_allow(net, type, code))
298 		return true;
299 
300 	if (icmp_global_allow()) {
301 		*apply_ratelimit = true;
302 		return true;
303 	}
304 	__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
305 	return false;
306 }
307 
308 /*
309  *	Send an ICMP frame.
310  */
311 
312 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
313 			       struct flowi4 *fl4, int type, int code,
314 			       bool apply_ratelimit)
315 {
316 	struct dst_entry *dst = &rt->dst;
317 	struct inet_peer *peer;
318 	bool rc = true;
319 	int vif;
320 
321 	if (!apply_ratelimit)
322 		return true;
323 
324 	/* No rate limit on loopback */
325 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
326 		goto out;
327 
328 	vif = l3mdev_master_ifindex(dst->dev);
329 	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
330 	rc = inet_peer_xrlim_allow(peer,
331 				   READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
332 	if (peer)
333 		inet_putpeer(peer);
334 out:
335 	if (!rc)
336 		__ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
337 	else
338 		icmp_global_consume();
339 	return rc;
340 }
341 
342 /*
343  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
344  */
345 void icmp_out_count(struct net *net, unsigned char type)
346 {
347 	ICMPMSGOUT_INC_STATS(net, type);
348 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
349 }
350 
351 /*
352  *	Checksum each fragment, and on the first include the headers and final
353  *	checksum.
354  */
355 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
356 			  struct sk_buff *skb)
357 {
358 	struct icmp_bxm *icmp_param = from;
359 	__wsum csum;
360 
361 	csum = skb_copy_and_csum_bits(icmp_param->skb,
362 				      icmp_param->offset + offset,
363 				      to, len);
364 
365 	skb->csum = csum_block_add(skb->csum, csum, odd);
366 	if (icmp_pointers[icmp_param->data.icmph.type].error)
367 		nf_ct_attach(skb, icmp_param->skb);
368 	return 0;
369 }
370 
371 static void icmp_push_reply(struct sock *sk,
372 			    struct icmp_bxm *icmp_param,
373 			    struct flowi4 *fl4,
374 			    struct ipcm_cookie *ipc, struct rtable **rt)
375 {
376 	struct sk_buff *skb;
377 
378 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
379 			   icmp_param->data_len+icmp_param->head_len,
380 			   icmp_param->head_len,
381 			   ipc, rt, MSG_DONTWAIT) < 0) {
382 		__ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
383 		ip_flush_pending_frames(sk);
384 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
385 		struct icmphdr *icmph = icmp_hdr(skb);
386 		__wsum csum;
387 		struct sk_buff *skb1;
388 
389 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
390 						 (char *)icmph,
391 						 icmp_param->head_len);
392 		skb_queue_walk(&sk->sk_write_queue, skb1) {
393 			csum = csum_add(csum, skb1->csum);
394 		}
395 		icmph->checksum = csum_fold(csum);
396 		skb->ip_summed = CHECKSUM_NONE;
397 		ip_push_pending_frames(sk, fl4);
398 	}
399 }
400 
401 /*
402  *	Driving logic for building and sending ICMP messages.
403  */
404 
405 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
406 {
407 	struct ipcm_cookie ipc;
408 	struct rtable *rt = skb_rtable(skb);
409 	struct net *net = dev_net(rt->dst.dev);
410 	bool apply_ratelimit = false;
411 	struct flowi4 fl4;
412 	struct sock *sk;
413 	struct inet_sock *inet;
414 	__be32 daddr, saddr;
415 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
416 	int type = icmp_param->data.icmph.type;
417 	int code = icmp_param->data.icmph.code;
418 
419 	if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
420 		return;
421 
422 	/* Needed by both icmpv4_global_allow and icmp_xmit_lock */
423 	local_bh_disable();
424 
425 	/* is global icmp_msgs_per_sec exhausted ? */
426 	if (!icmpv4_global_allow(net, type, code, &apply_ratelimit))
427 		goto out_bh_enable;
428 
429 	sk = icmp_xmit_lock(net);
430 	if (!sk)
431 		goto out_bh_enable;
432 	inet = inet_sk(sk);
433 
434 	icmp_param->data.icmph.checksum = 0;
435 
436 	ipcm_init(&ipc);
437 	inet->tos = ip_hdr(skb)->tos;
438 	ipc.sockc.mark = mark;
439 	daddr = ipc.addr = ip_hdr(skb)->saddr;
440 	saddr = fib_compute_spec_dst(skb);
441 
442 	if (icmp_param->replyopts.opt.opt.optlen) {
443 		ipc.opt = &icmp_param->replyopts.opt;
444 		if (ipc.opt->opt.srr)
445 			daddr = icmp_param->replyopts.opt.opt.faddr;
446 	}
447 	memset(&fl4, 0, sizeof(fl4));
448 	fl4.daddr = daddr;
449 	fl4.saddr = saddr;
450 	fl4.flowi4_mark = mark;
451 	fl4.flowi4_uid = sock_net_uid(net, NULL);
452 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
453 	fl4.flowi4_proto = IPPROTO_ICMP;
454 	fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
455 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
456 	rt = ip_route_output_key(net, &fl4);
457 	if (IS_ERR(rt))
458 		goto out_unlock;
459 	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
460 		icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
461 	ip_rt_put(rt);
462 out_unlock:
463 	icmp_xmit_unlock(sk);
464 out_bh_enable:
465 	local_bh_enable();
466 }
467 
468 /*
469  * The device used for looking up which routing table to use for sending an ICMP
470  * error is preferably the source whenever it is set, which should ensure the
471  * icmp error can be sent to the source host, else lookup using the routing
472  * table of the destination device, else use the main routing table (index 0).
473  */
474 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
475 {
476 	struct net_device *route_lookup_dev = NULL;
477 
478 	if (skb->dev)
479 		route_lookup_dev = skb->dev;
480 	else if (skb_dst(skb))
481 		route_lookup_dev = skb_dst(skb)->dev;
482 	return route_lookup_dev;
483 }
484 
485 static struct rtable *icmp_route_lookup(struct net *net,
486 					struct flowi4 *fl4,
487 					struct sk_buff *skb_in,
488 					const struct iphdr *iph,
489 					__be32 saddr, u8 tos, u32 mark,
490 					int type, int code,
491 					struct icmp_bxm *param)
492 {
493 	struct net_device *route_lookup_dev;
494 	struct rtable *rt, *rt2;
495 	struct flowi4 fl4_dec;
496 	int err;
497 
498 	memset(fl4, 0, sizeof(*fl4));
499 	fl4->daddr = (param->replyopts.opt.opt.srr ?
500 		      param->replyopts.opt.opt.faddr : iph->saddr);
501 	fl4->saddr = saddr;
502 	fl4->flowi4_mark = mark;
503 	fl4->flowi4_uid = sock_net_uid(net, NULL);
504 	fl4->flowi4_tos = RT_TOS(tos);
505 	fl4->flowi4_proto = IPPROTO_ICMP;
506 	fl4->fl4_icmp_type = type;
507 	fl4->fl4_icmp_code = code;
508 	route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
509 	fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
510 
511 	security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
512 	rt = ip_route_output_key_hash(net, fl4, skb_in);
513 	if (IS_ERR(rt))
514 		return rt;
515 
516 	/* No need to clone since we're just using its address. */
517 	rt2 = rt;
518 
519 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
520 					   flowi4_to_flowi(fl4), NULL, 0);
521 	if (!IS_ERR(rt)) {
522 		if (rt != rt2)
523 			return rt;
524 	} else if (PTR_ERR(rt) == -EPERM) {
525 		rt = NULL;
526 	} else
527 		return rt;
528 
529 	err = xfrm_decode_session_reverse(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 				     RT_TOS(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 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
563 					    flowi4_to_flowi(&fl4_dec), NULL,
564 					    XFRM_LOOKUP_ICMP);
565 	if (!IS_ERR(rt2)) {
566 		dst_release(&rt->dst);
567 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
568 		rt = rt2;
569 	} else if (PTR_ERR(rt2) == -EPERM) {
570 		if (rt)
571 			dst_release(&rt->dst);
572 		return rt2;
573 	} else {
574 		err = PTR_ERR(rt2);
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 	icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
781 ende:
782 	ip_rt_put(rt);
783 out_unlock:
784 	icmp_xmit_unlock(sk);
785 out_bh_enable:
786 	local_bh_enable();
787 out:;
788 }
789 EXPORT_SYMBOL(__icmp_send);
790 
791 #if IS_ENABLED(CONFIG_NF_NAT)
792 #include <net/netfilter/nf_conntrack.h>
793 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
794 {
795 	struct sk_buff *cloned_skb = NULL;
796 	struct ip_options opts = { 0 };
797 	enum ip_conntrack_info ctinfo;
798 	struct nf_conn *ct;
799 	__be32 orig_ip;
800 
801 	ct = nf_ct_get(skb_in, &ctinfo);
802 	if (!ct || !(ct->status & IPS_SRC_NAT)) {
803 		__icmp_send(skb_in, type, code, info, &opts);
804 		return;
805 	}
806 
807 	if (skb_shared(skb_in))
808 		skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
809 
810 	if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
811 	    (skb_network_header(skb_in) + sizeof(struct iphdr)) >
812 	    skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
813 	    skb_network_offset(skb_in) + sizeof(struct iphdr))))
814 		goto out;
815 
816 	orig_ip = ip_hdr(skb_in)->saddr;
817 	ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
818 	__icmp_send(skb_in, type, code, info, &opts);
819 	ip_hdr(skb_in)->saddr = orig_ip;
820 out:
821 	consume_skb(cloned_skb);
822 }
823 EXPORT_SYMBOL(icmp_ndo_send);
824 #endif
825 
826 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
827 {
828 	const struct iphdr *iph = (const struct iphdr *)skb->data;
829 	const struct net_protocol *ipprot;
830 	int protocol = iph->protocol;
831 
832 	/* Checkin full IP header plus 8 bytes of protocol to
833 	 * avoid additional coding at protocol handlers.
834 	 */
835 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
836 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
837 		return;
838 	}
839 
840 	raw_icmp_error(skb, protocol, info);
841 
842 	ipprot = rcu_dereference(inet_protos[protocol]);
843 	if (ipprot && ipprot->err_handler)
844 		ipprot->err_handler(skb, info);
845 }
846 
847 static bool icmp_tag_validation(int proto)
848 {
849 	bool ok;
850 
851 	rcu_read_lock();
852 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
853 	rcu_read_unlock();
854 	return ok;
855 }
856 
857 /*
858  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
859  *	ICMP_PARAMETERPROB.
860  */
861 
862 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
863 {
864 	enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
865 	const struct iphdr *iph;
866 	struct icmphdr *icmph;
867 	struct net *net;
868 	u32 info = 0;
869 
870 	net = dev_net(skb_dst(skb)->dev);
871 
872 	/*
873 	 *	Incomplete header ?
874 	 * 	Only checks for the IP header, there should be an
875 	 *	additional check for longer headers in upper levels.
876 	 */
877 
878 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
879 		goto out_err;
880 
881 	icmph = icmp_hdr(skb);
882 	iph   = (const struct iphdr *)skb->data;
883 
884 	if (iph->ihl < 5)  { /* Mangled header, drop. */
885 		reason = SKB_DROP_REASON_IP_INHDR;
886 		goto out_err;
887 	}
888 
889 	switch (icmph->type) {
890 	case ICMP_DEST_UNREACH:
891 		switch (icmph->code & 15) {
892 		case ICMP_NET_UNREACH:
893 		case ICMP_HOST_UNREACH:
894 		case ICMP_PROT_UNREACH:
895 		case ICMP_PORT_UNREACH:
896 			break;
897 		case ICMP_FRAG_NEEDED:
898 			/* for documentation of the ip_no_pmtu_disc
899 			 * values please see
900 			 * Documentation/networking/ip-sysctl.rst
901 			 */
902 			switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
903 			default:
904 				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
905 						    &iph->daddr);
906 				break;
907 			case 2:
908 				goto out;
909 			case 3:
910 				if (!icmp_tag_validation(iph->protocol))
911 					goto out;
912 				fallthrough;
913 			case 0:
914 				info = ntohs(icmph->un.frag.mtu);
915 			}
916 			break;
917 		case ICMP_SR_FAILED:
918 			net_dbg_ratelimited("%pI4: Source Route Failed\n",
919 					    &iph->daddr);
920 			break;
921 		default:
922 			break;
923 		}
924 		if (icmph->code > NR_ICMP_UNREACH)
925 			goto out;
926 		break;
927 	case ICMP_PARAMETERPROB:
928 		info = ntohl(icmph->un.gateway) >> 24;
929 		break;
930 	case ICMP_TIME_EXCEEDED:
931 		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
932 		if (icmph->code == ICMP_EXC_FRAGTIME)
933 			goto out;
934 		break;
935 	}
936 
937 	/*
938 	 *	Throw it at our lower layers
939 	 *
940 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
941 	 *		  header.
942 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
943 	 *		  transport layer.
944 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
945 	 *		  transport layer.
946 	 */
947 
948 	/*
949 	 *	Check the other end isn't violating RFC 1122. Some routers send
950 	 *	bogus responses to broadcast frames. If you see this message
951 	 *	first check your netmask matches at both ends, if it does then
952 	 *	get the other vendor to fix their kit.
953 	 */
954 
955 	if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
956 	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
957 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
958 				     &ip_hdr(skb)->saddr,
959 				     icmph->type, icmph->code,
960 				     &iph->daddr, skb->dev->name);
961 		goto out;
962 	}
963 
964 	icmp_socket_deliver(skb, info);
965 
966 out:
967 	return reason;
968 out_err:
969 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
970 	return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
971 }
972 
973 
974 /*
975  *	Handle ICMP_REDIRECT.
976  */
977 
978 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
979 {
980 	if (skb->len < sizeof(struct iphdr)) {
981 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
982 		return SKB_DROP_REASON_PKT_TOO_SMALL;
983 	}
984 
985 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
986 		/* there aught to be a stat */
987 		return SKB_DROP_REASON_NOMEM;
988 	}
989 
990 	icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
991 	return SKB_NOT_DROPPED_YET;
992 }
993 
994 /*
995  *	Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
996  *
997  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
998  *		  requests.
999  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
1000  *		  included in the reply.
1001  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
1002  *		  echo requests, MUST have default=NOT.
1003  *	RFC 8335: 8 MUST have a config option to enable/disable ICMP
1004  *		  Extended Echo Functionality, MUST be disabled by default
1005  *	See also WRT handling of options once they are done and working.
1006  */
1007 
1008 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1009 {
1010 	struct icmp_bxm icmp_param;
1011 	struct net *net;
1012 
1013 	net = dev_net(skb_dst(skb)->dev);
1014 	/* should there be an ICMP stat for ignored echos? */
1015 	if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1016 		return SKB_NOT_DROPPED_YET;
1017 
1018 	icmp_param.data.icmph	   = *icmp_hdr(skb);
1019 	icmp_param.skb		   = skb;
1020 	icmp_param.offset	   = 0;
1021 	icmp_param.data_len	   = skb->len;
1022 	icmp_param.head_len	   = sizeof(struct icmphdr);
1023 
1024 	if (icmp_param.data.icmph.type == ICMP_ECHO)
1025 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1026 	else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1027 		return SKB_NOT_DROPPED_YET;
1028 
1029 	icmp_reply(&icmp_param, skb);
1030 	return SKB_NOT_DROPPED_YET;
1031 }
1032 
1033 /*	Helper for icmp_echo and icmpv6_echo_reply.
1034  *	Searches for net_device that matches PROBE interface identifier
1035  *		and builds PROBE reply message in icmphdr.
1036  *
1037  *	Returns false if PROBE responses are disabled via sysctl
1038  */
1039 
1040 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1041 {
1042 	struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1043 	struct icmp_ext_echo_iio *iio, _iio;
1044 	struct net *net = dev_net(skb->dev);
1045 	struct inet6_dev *in6_dev;
1046 	struct in_device *in_dev;
1047 	struct net_device *dev;
1048 	char buff[IFNAMSIZ];
1049 	u16 ident_len;
1050 	u8 status;
1051 
1052 	if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1053 		return false;
1054 
1055 	/* We currently only support probing interfaces on the proxy node
1056 	 * Check to ensure L-bit is set
1057 	 */
1058 	if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1059 		return false;
1060 	/* Clear status bits in reply message */
1061 	icmphdr->un.echo.sequence &= htons(0xFF00);
1062 	if (icmphdr->type == ICMP_EXT_ECHO)
1063 		icmphdr->type = ICMP_EXT_ECHOREPLY;
1064 	else
1065 		icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1066 	ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1067 	/* Size of iio is class_type dependent.
1068 	 * Only check header here and assign length based on ctype in the switch statement
1069 	 */
1070 	iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1071 	if (!ext_hdr || !iio)
1072 		goto send_mal_query;
1073 	if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1074 	    ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1075 		goto send_mal_query;
1076 	ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1077 	iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1078 				 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1079 	if (!iio)
1080 		goto send_mal_query;
1081 
1082 	status = 0;
1083 	dev = NULL;
1084 	switch (iio->extobj_hdr.class_type) {
1085 	case ICMP_EXT_ECHO_CTYPE_NAME:
1086 		if (ident_len >= IFNAMSIZ)
1087 			goto send_mal_query;
1088 		memset(buff, 0, sizeof(buff));
1089 		memcpy(buff, &iio->ident.name, ident_len);
1090 		dev = dev_get_by_name(net, buff);
1091 		break;
1092 	case ICMP_EXT_ECHO_CTYPE_INDEX:
1093 		if (ident_len != sizeof(iio->ident.ifindex))
1094 			goto send_mal_query;
1095 		dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1096 		break;
1097 	case ICMP_EXT_ECHO_CTYPE_ADDR:
1098 		if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1099 		    ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1100 				 iio->ident.addr.ctype3_hdr.addrlen)
1101 			goto send_mal_query;
1102 		switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1103 		case ICMP_AFI_IP:
1104 			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1105 				goto send_mal_query;
1106 			dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1107 			break;
1108 #if IS_ENABLED(CONFIG_IPV6)
1109 		case ICMP_AFI_IP6:
1110 			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1111 				goto send_mal_query;
1112 			dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1113 			dev_hold(dev);
1114 			break;
1115 #endif
1116 		default:
1117 			goto send_mal_query;
1118 		}
1119 		break;
1120 	default:
1121 		goto send_mal_query;
1122 	}
1123 	if (!dev) {
1124 		icmphdr->code = ICMP_EXT_CODE_NO_IF;
1125 		return true;
1126 	}
1127 	/* Fill bits in reply message */
1128 	if (dev->flags & IFF_UP)
1129 		status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1130 
1131 	in_dev = __in_dev_get_rcu(dev);
1132 	if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1133 		status |= ICMP_EXT_ECHOREPLY_IPV4;
1134 
1135 	in6_dev = __in6_dev_get(dev);
1136 	if (in6_dev && !list_empty(&in6_dev->addr_list))
1137 		status |= ICMP_EXT_ECHOREPLY_IPV6;
1138 
1139 	dev_put(dev);
1140 	icmphdr->un.echo.sequence |= htons(status);
1141 	return true;
1142 send_mal_query:
1143 	icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1144 	return true;
1145 }
1146 EXPORT_SYMBOL_GPL(icmp_build_probe);
1147 
1148 /*
1149  *	Handle ICMP Timestamp requests.
1150  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1151  *		  SHOULD be in the kernel for minimum random latency.
1152  *		  MUST be accurate to a few minutes.
1153  *		  MUST be updated at least at 15Hz.
1154  */
1155 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1156 {
1157 	struct icmp_bxm icmp_param;
1158 	/*
1159 	 *	Too short.
1160 	 */
1161 	if (skb->len < 4)
1162 		goto out_err;
1163 
1164 	/*
1165 	 *	Fill in the current time as ms since midnight UT:
1166 	 */
1167 	icmp_param.data.times[1] = inet_current_timestamp();
1168 	icmp_param.data.times[2] = icmp_param.data.times[1];
1169 
1170 	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1171 
1172 	icmp_param.data.icmph	   = *icmp_hdr(skb);
1173 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1174 	icmp_param.data.icmph.code = 0;
1175 	icmp_param.skb		   = skb;
1176 	icmp_param.offset	   = 0;
1177 	icmp_param.data_len	   = 0;
1178 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
1179 	icmp_reply(&icmp_param, skb);
1180 	return SKB_NOT_DROPPED_YET;
1181 
1182 out_err:
1183 	__ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1184 	return SKB_DROP_REASON_PKT_TOO_SMALL;
1185 }
1186 
1187 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1188 {
1189 	/* pretend it was a success */
1190 	return SKB_NOT_DROPPED_YET;
1191 }
1192 
1193 /*
1194  *	Deal with incoming ICMP packets.
1195  */
1196 int icmp_rcv(struct sk_buff *skb)
1197 {
1198 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1199 	struct rtable *rt = skb_rtable(skb);
1200 	struct net *net = dev_net(rt->dst.dev);
1201 	struct icmphdr *icmph;
1202 
1203 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1204 		struct sec_path *sp = skb_sec_path(skb);
1205 		int nh;
1206 
1207 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1208 				 XFRM_STATE_ICMP)) {
1209 			reason = SKB_DROP_REASON_XFRM_POLICY;
1210 			goto drop;
1211 		}
1212 
1213 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1214 			goto drop;
1215 
1216 		nh = skb_network_offset(skb);
1217 		skb_set_network_header(skb, sizeof(*icmph));
1218 
1219 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1220 						skb)) {
1221 			reason = SKB_DROP_REASON_XFRM_POLICY;
1222 			goto drop;
1223 		}
1224 
1225 		skb_set_network_header(skb, nh);
1226 	}
1227 
1228 	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1229 
1230 	if (skb_checksum_simple_validate(skb))
1231 		goto csum_error;
1232 
1233 	if (!pskb_pull(skb, sizeof(*icmph)))
1234 		goto error;
1235 
1236 	icmph = icmp_hdr(skb);
1237 
1238 	ICMPMSGIN_INC_STATS(net, icmph->type);
1239 
1240 	/* Check for ICMP Extended Echo (PROBE) messages */
1241 	if (icmph->type == ICMP_EXT_ECHO) {
1242 		/* We can't use icmp_pointers[].handler() because it is an array of
1243 		 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1244 		 */
1245 		reason = icmp_echo(skb);
1246 		goto reason_check;
1247 	}
1248 
1249 	if (icmph->type == ICMP_EXT_ECHOREPLY) {
1250 		reason = ping_rcv(skb);
1251 		goto reason_check;
1252 	}
1253 
1254 	/*
1255 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1256 	 *
1257 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1258 	 *		  discarded.
1259 	 */
1260 	if (icmph->type > NR_ICMP_TYPES) {
1261 		reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1262 		goto error;
1263 	}
1264 
1265 	/*
1266 	 *	Parse the ICMP message
1267 	 */
1268 
1269 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1270 		/*
1271 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1272 		 *	  silently ignored (we let user decide with a sysctl).
1273 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1274 		 *	  discarded if to broadcast/multicast.
1275 		 */
1276 		if ((icmph->type == ICMP_ECHO ||
1277 		     icmph->type == ICMP_TIMESTAMP) &&
1278 		    READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1279 			reason = SKB_DROP_REASON_INVALID_PROTO;
1280 			goto error;
1281 		}
1282 		if (icmph->type != ICMP_ECHO &&
1283 		    icmph->type != ICMP_TIMESTAMP &&
1284 		    icmph->type != ICMP_ADDRESS &&
1285 		    icmph->type != ICMP_ADDRESSREPLY) {
1286 			reason = SKB_DROP_REASON_INVALID_PROTO;
1287 			goto error;
1288 		}
1289 	}
1290 
1291 	reason = icmp_pointers[icmph->type].handler(skb);
1292 reason_check:
1293 	if (!reason)  {
1294 		consume_skb(skb);
1295 		return NET_RX_SUCCESS;
1296 	}
1297 
1298 drop:
1299 	kfree_skb_reason(skb, reason);
1300 	return NET_RX_DROP;
1301 csum_error:
1302 	reason = SKB_DROP_REASON_ICMP_CSUM;
1303 	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1304 error:
1305 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1306 	goto drop;
1307 }
1308 
1309 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1310 {
1311 	struct icmp_extobj_hdr *objh, _objh;
1312 	struct icmp_ext_hdr *exth, _exth;
1313 	u16 olen;
1314 
1315 	exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1316 	if (!exth)
1317 		return false;
1318 	if (exth->version != 2)
1319 		return true;
1320 
1321 	if (exth->checksum &&
1322 	    csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1323 		return false;
1324 
1325 	off += sizeof(_exth);
1326 	while (off < skb->len) {
1327 		objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1328 		if (!objh)
1329 			return false;
1330 
1331 		olen = ntohs(objh->length);
1332 		if (olen < sizeof(_objh))
1333 			return false;
1334 
1335 		off += olen;
1336 		if (off > skb->len)
1337 			return false;
1338 	}
1339 
1340 	return true;
1341 }
1342 
1343 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1344 			   struct sock_ee_data_rfc4884 *out,
1345 			   int thlen, int off)
1346 {
1347 	int hlen;
1348 
1349 	/* original datagram headers: end of icmph to payload (skb->data) */
1350 	hlen = -skb_transport_offset(skb) - thlen;
1351 
1352 	/* per rfc 4884: minimal datagram length of 128 bytes */
1353 	if (off < 128 || off < hlen)
1354 		return;
1355 
1356 	/* kernel has stripped headers: return payload offset in bytes */
1357 	off -= hlen;
1358 	if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1359 		return;
1360 
1361 	out->len = off;
1362 
1363 	if (!ip_icmp_error_rfc4884_validate(skb, off))
1364 		out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1365 }
1366 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1367 
1368 int icmp_err(struct sk_buff *skb, u32 info)
1369 {
1370 	struct iphdr *iph = (struct iphdr *)skb->data;
1371 	int offset = iph->ihl<<2;
1372 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1373 	int type = icmp_hdr(skb)->type;
1374 	int code = icmp_hdr(skb)->code;
1375 	struct net *net = dev_net(skb->dev);
1376 
1377 	/*
1378 	 * Use ping_err to handle all icmp errors except those
1379 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1380 	 */
1381 	if (icmph->type != ICMP_ECHOREPLY) {
1382 		ping_err(skb, offset, info);
1383 		return 0;
1384 	}
1385 
1386 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1387 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1388 	else if (type == ICMP_REDIRECT)
1389 		ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1390 
1391 	return 0;
1392 }
1393 
1394 /*
1395  *	This table is the definition of how we handle ICMP.
1396  */
1397 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1398 	[ICMP_ECHOREPLY] = {
1399 		.handler = ping_rcv,
1400 	},
1401 	[1] = {
1402 		.handler = icmp_discard,
1403 		.error = 1,
1404 	},
1405 	[2] = {
1406 		.handler = icmp_discard,
1407 		.error = 1,
1408 	},
1409 	[ICMP_DEST_UNREACH] = {
1410 		.handler = icmp_unreach,
1411 		.error = 1,
1412 	},
1413 	[ICMP_SOURCE_QUENCH] = {
1414 		.handler = icmp_unreach,
1415 		.error = 1,
1416 	},
1417 	[ICMP_REDIRECT] = {
1418 		.handler = icmp_redirect,
1419 		.error = 1,
1420 	},
1421 	[6] = {
1422 		.handler = icmp_discard,
1423 		.error = 1,
1424 	},
1425 	[7] = {
1426 		.handler = icmp_discard,
1427 		.error = 1,
1428 	},
1429 	[ICMP_ECHO] = {
1430 		.handler = icmp_echo,
1431 	},
1432 	[9] = {
1433 		.handler = icmp_discard,
1434 		.error = 1,
1435 	},
1436 	[10] = {
1437 		.handler = icmp_discard,
1438 		.error = 1,
1439 	},
1440 	[ICMP_TIME_EXCEEDED] = {
1441 		.handler = icmp_unreach,
1442 		.error = 1,
1443 	},
1444 	[ICMP_PARAMETERPROB] = {
1445 		.handler = icmp_unreach,
1446 		.error = 1,
1447 	},
1448 	[ICMP_TIMESTAMP] = {
1449 		.handler = icmp_timestamp,
1450 	},
1451 	[ICMP_TIMESTAMPREPLY] = {
1452 		.handler = icmp_discard,
1453 	},
1454 	[ICMP_INFO_REQUEST] = {
1455 		.handler = icmp_discard,
1456 	},
1457 	[ICMP_INFO_REPLY] = {
1458 		.handler = icmp_discard,
1459 	},
1460 	[ICMP_ADDRESS] = {
1461 		.handler = icmp_discard,
1462 	},
1463 	[ICMP_ADDRESSREPLY] = {
1464 		.handler = icmp_discard,
1465 	},
1466 };
1467 
1468 static int __net_init icmp_sk_init(struct net *net)
1469 {
1470 	/* Control parameters for ECHO replies. */
1471 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1472 	net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1473 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1474 
1475 	/* Control parameter - ignore bogus broadcast responses? */
1476 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1477 
1478 	/*
1479 	 * 	Configurable global rate limit.
1480 	 *
1481 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1482 	 *	bucket ratemask defines which icmp types are ratelimited by
1483 	 *	setting	it's bit position.
1484 	 *
1485 	 *	default:
1486 	 *	dest unreachable (3), source quench (4),
1487 	 *	time exceeded (11), parameter problem (12)
1488 	 */
1489 
1490 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1491 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1492 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1493 
1494 	return 0;
1495 }
1496 
1497 static struct pernet_operations __net_initdata icmp_sk_ops = {
1498        .init = icmp_sk_init,
1499 };
1500 
1501 int __init icmp_init(void)
1502 {
1503 	int err, i;
1504 
1505 	for_each_possible_cpu(i) {
1506 		struct sock *sk;
1507 
1508 		err = inet_ctl_sock_create(&sk, PF_INET,
1509 					   SOCK_RAW, IPPROTO_ICMP, &init_net);
1510 		if (err < 0)
1511 			return err;
1512 
1513 		per_cpu(ipv4_icmp_sk, i) = sk;
1514 
1515 		/* Enough space for 2 64K ICMP packets, including
1516 		 * sk_buff/skb_shared_info struct overhead.
1517 		 */
1518 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1519 
1520 		/*
1521 		 * Speedup sock_wfree()
1522 		 */
1523 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1524 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1525 	}
1526 	return register_pernet_subsys(&icmp_sk_ops);
1527 }
1528