xref: /linux/net/ipv4/icmp.c (revision a15c89c7)
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 
100 struct 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 
116 const 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 };
182 EXPORT_SYMBOL(icmp_err_convert);
183 
184 /*
185  *	ICMP control array. This specifies what to do with each ICMP.
186  */
187 
188 struct icmp_control {
189 	bool (*handler)(struct sk_buff *skb);
190 	short   error;		/* This ICMP is classed as an error message */
191 };
192 
193 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
194 
195 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
196 
197 /* Called with BH disabled */
198 static 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 
214 static 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 
220 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
221 int sysctl_icmp_msgs_burst __read_mostly = 50;
222 
223 static 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  */
238 bool 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 = 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, sysctl_icmp_msgs_burst);
261 	if (credit) {
262 		/* We want to use a credit of one in average, but need to randomize
263 		 * it for security reasons.
264 		 */
265 		credit = max_t(int, credit - prandom_u32_max(3), 0);
266 		rc = true;
267 	}
268 	WRITE_ONCE(icmp_global.credit, credit);
269 	spin_unlock(&icmp_global.lock);
270 	return rc;
271 }
272 EXPORT_SYMBOL(icmp_global_allow);
273 
274 static 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) & net->ipv4.sysctl_icmp_ratemask))
285 		return true;
286 
287 	return false;
288 }
289 
290 static bool icmpv4_global_allow(struct net *net, int type, int code)
291 {
292 	if (icmpv4_mask_allow(net, type, code))
293 		return true;
294 
295 	if (icmp_global_allow())
296 		return true;
297 
298 	return false;
299 }
300 
301 /*
302  *	Send an ICMP frame.
303  */
304 
305 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
306 			       struct flowi4 *fl4, int type, int code)
307 {
308 	struct dst_entry *dst = &rt->dst;
309 	struct inet_peer *peer;
310 	bool rc = true;
311 	int vif;
312 
313 	if (icmpv4_mask_allow(net, type, code))
314 		goto out;
315 
316 	/* No rate limit on loopback */
317 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
318 		goto out;
319 
320 	vif = l3mdev_master_ifindex(dst->dev);
321 	peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
322 	rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
323 	if (peer)
324 		inet_putpeer(peer);
325 out:
326 	return rc;
327 }
328 
329 /*
330  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
331  */
332 void icmp_out_count(struct net *net, unsigned char type)
333 {
334 	ICMPMSGOUT_INC_STATS(net, type);
335 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
336 }
337 
338 /*
339  *	Checksum each fragment, and on the first include the headers and final
340  *	checksum.
341  */
342 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
343 			  struct sk_buff *skb)
344 {
345 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
346 	__wsum csum;
347 
348 	csum = skb_copy_and_csum_bits(icmp_param->skb,
349 				      icmp_param->offset + offset,
350 				      to, len);
351 
352 	skb->csum = csum_block_add(skb->csum, csum, odd);
353 	if (icmp_pointers[icmp_param->data.icmph.type].error)
354 		nf_ct_attach(skb, icmp_param->skb);
355 	return 0;
356 }
357 
358 static void icmp_push_reply(struct sock *sk,
359 			    struct icmp_bxm *icmp_param,
360 			    struct flowi4 *fl4,
361 			    struct ipcm_cookie *ipc, struct rtable **rt)
362 {
363 	struct sk_buff *skb;
364 
365 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
366 			   icmp_param->data_len+icmp_param->head_len,
367 			   icmp_param->head_len,
368 			   ipc, rt, MSG_DONTWAIT) < 0) {
369 		__ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
370 		ip_flush_pending_frames(sk);
371 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
372 		struct icmphdr *icmph = icmp_hdr(skb);
373 		__wsum csum;
374 		struct sk_buff *skb1;
375 
376 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
377 						 (char *)icmph,
378 						 icmp_param->head_len);
379 		skb_queue_walk(&sk->sk_write_queue, skb1) {
380 			csum = csum_add(csum, skb1->csum);
381 		}
382 		icmph->checksum = csum_fold(csum);
383 		skb->ip_summed = CHECKSUM_NONE;
384 		ip_push_pending_frames(sk, fl4);
385 	}
386 }
387 
388 /*
389  *	Driving logic for building and sending ICMP messages.
390  */
391 
392 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
393 {
394 	struct ipcm_cookie ipc;
395 	struct rtable *rt = skb_rtable(skb);
396 	struct net *net = dev_net(rt->dst.dev);
397 	struct flowi4 fl4;
398 	struct sock *sk;
399 	struct inet_sock *inet;
400 	__be32 daddr, saddr;
401 	u32 mark = IP4_REPLY_MARK(net, skb->mark);
402 	int type = icmp_param->data.icmph.type;
403 	int code = icmp_param->data.icmph.code;
404 
405 	if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
406 		return;
407 
408 	/* Needed by both icmp_global_allow and icmp_xmit_lock */
409 	local_bh_disable();
410 
411 	/* global icmp_msgs_per_sec */
412 	if (!icmpv4_global_allow(net, type, code))
413 		goto out_bh_enable;
414 
415 	sk = icmp_xmit_lock(net);
416 	if (!sk)
417 		goto out_bh_enable;
418 	inet = inet_sk(sk);
419 
420 	icmp_param->data.icmph.checksum = 0;
421 
422 	ipcm_init(&ipc);
423 	inet->tos = ip_hdr(skb)->tos;
424 	ipc.sockc.mark = mark;
425 	daddr = ipc.addr = ip_hdr(skb)->saddr;
426 	saddr = fib_compute_spec_dst(skb);
427 
428 	if (icmp_param->replyopts.opt.opt.optlen) {
429 		ipc.opt = &icmp_param->replyopts.opt;
430 		if (ipc.opt->opt.srr)
431 			daddr = icmp_param->replyopts.opt.opt.faddr;
432 	}
433 	memset(&fl4, 0, sizeof(fl4));
434 	fl4.daddr = daddr;
435 	fl4.saddr = saddr;
436 	fl4.flowi4_mark = mark;
437 	fl4.flowi4_uid = sock_net_uid(net, NULL);
438 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
439 	fl4.flowi4_proto = IPPROTO_ICMP;
440 	fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
441 	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
442 	rt = ip_route_output_key(net, &fl4);
443 	if (IS_ERR(rt))
444 		goto out_unlock;
445 	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
446 		icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
447 	ip_rt_put(rt);
448 out_unlock:
449 	icmp_xmit_unlock(sk);
450 out_bh_enable:
451 	local_bh_enable();
452 }
453 
454 /*
455  * The device used for looking up which routing table to use for sending an ICMP
456  * error is preferably the source whenever it is set, which should ensure the
457  * icmp error can be sent to the source host, else lookup using the routing
458  * table of the destination device, else use the main routing table (index 0).
459  */
460 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
461 {
462 	struct net_device *route_lookup_dev = NULL;
463 
464 	if (skb->dev)
465 		route_lookup_dev = skb->dev;
466 	else if (skb_dst(skb))
467 		route_lookup_dev = skb_dst(skb)->dev;
468 	return route_lookup_dev;
469 }
470 
471 static struct rtable *icmp_route_lookup(struct net *net,
472 					struct flowi4 *fl4,
473 					struct sk_buff *skb_in,
474 					const struct iphdr *iph,
475 					__be32 saddr, u8 tos, u32 mark,
476 					int type, int code,
477 					struct icmp_bxm *param)
478 {
479 	struct net_device *route_lookup_dev;
480 	struct rtable *rt, *rt2;
481 	struct flowi4 fl4_dec;
482 	int err;
483 
484 	memset(fl4, 0, sizeof(*fl4));
485 	fl4->daddr = (param->replyopts.opt.opt.srr ?
486 		      param->replyopts.opt.opt.faddr : iph->saddr);
487 	fl4->saddr = saddr;
488 	fl4->flowi4_mark = mark;
489 	fl4->flowi4_uid = sock_net_uid(net, NULL);
490 	fl4->flowi4_tos = RT_TOS(tos);
491 	fl4->flowi4_proto = IPPROTO_ICMP;
492 	fl4->fl4_icmp_type = type;
493 	fl4->fl4_icmp_code = code;
494 	route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
495 	fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
496 
497 	security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
498 	rt = ip_route_output_key_hash(net, fl4, skb_in);
499 	if (IS_ERR(rt))
500 		return rt;
501 
502 	/* No need to clone since we're just using its address. */
503 	rt2 = rt;
504 
505 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
506 					   flowi4_to_flowi(fl4), NULL, 0);
507 	if (!IS_ERR(rt)) {
508 		if (rt != rt2)
509 			return rt;
510 	} else if (PTR_ERR(rt) == -EPERM) {
511 		rt = NULL;
512 	} else
513 		return rt;
514 
515 	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
516 	if (err)
517 		goto relookup_failed;
518 
519 	if (inet_addr_type_dev_table(net, route_lookup_dev,
520 				     fl4_dec.saddr) == RTN_LOCAL) {
521 		rt2 = __ip_route_output_key(net, &fl4_dec);
522 		if (IS_ERR(rt2))
523 			err = PTR_ERR(rt2);
524 	} else {
525 		struct flowi4 fl4_2 = {};
526 		unsigned long orefdst;
527 
528 		fl4_2.daddr = fl4_dec.saddr;
529 		rt2 = ip_route_output_key(net, &fl4_2);
530 		if (IS_ERR(rt2)) {
531 			err = PTR_ERR(rt2);
532 			goto relookup_failed;
533 		}
534 		/* Ugh! */
535 		orefdst = skb_in->_skb_refdst; /* save old refdst */
536 		skb_dst_set(skb_in, NULL);
537 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
538 				     RT_TOS(tos), rt2->dst.dev);
539 
540 		dst_release(&rt2->dst);
541 		rt2 = skb_rtable(skb_in);
542 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
543 	}
544 
545 	if (err)
546 		goto relookup_failed;
547 
548 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
549 					    flowi4_to_flowi(&fl4_dec), NULL,
550 					    XFRM_LOOKUP_ICMP);
551 	if (!IS_ERR(rt2)) {
552 		dst_release(&rt->dst);
553 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
554 		rt = rt2;
555 	} else if (PTR_ERR(rt2) == -EPERM) {
556 		if (rt)
557 			dst_release(&rt->dst);
558 		return rt2;
559 	} else {
560 		err = PTR_ERR(rt2);
561 		goto relookup_failed;
562 	}
563 	return rt;
564 
565 relookup_failed:
566 	if (rt)
567 		return rt;
568 	return ERR_PTR(err);
569 }
570 
571 /*
572  *	Send an ICMP message in response to a situation
573  *
574  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
575  *		  MAY send more (we do).
576  *			MUST NOT change this header information.
577  *			MUST NOT reply to a multicast/broadcast IP address.
578  *			MUST NOT reply to a multicast/broadcast MAC address.
579  *			MUST reply to only the first fragment.
580  */
581 
582 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
583 		 const struct ip_options *opt)
584 {
585 	struct iphdr *iph;
586 	int room;
587 	struct icmp_bxm icmp_param;
588 	struct rtable *rt = skb_rtable(skb_in);
589 	struct ipcm_cookie ipc;
590 	struct flowi4 fl4;
591 	__be32 saddr;
592 	u8  tos;
593 	u32 mark;
594 	struct net *net;
595 	struct sock *sk;
596 
597 	if (!rt)
598 		goto out;
599 
600 	if (rt->dst.dev)
601 		net = dev_net(rt->dst.dev);
602 	else if (skb_in->dev)
603 		net = dev_net(skb_in->dev);
604 	else
605 		goto out;
606 
607 	/*
608 	 *	Find the original header. It is expected to be valid, of course.
609 	 *	Check this, icmp_send is called from the most obscure devices
610 	 *	sometimes.
611 	 */
612 	iph = ip_hdr(skb_in);
613 
614 	if ((u8 *)iph < skb_in->head ||
615 	    (skb_network_header(skb_in) + sizeof(*iph)) >
616 	    skb_tail_pointer(skb_in))
617 		goto out;
618 
619 	/*
620 	 *	No replies to physical multicast/broadcast
621 	 */
622 	if (skb_in->pkt_type != PACKET_HOST)
623 		goto out;
624 
625 	/*
626 	 *	Now check at the protocol level
627 	 */
628 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
629 		goto out;
630 
631 	/*
632 	 *	Only reply to fragment 0. We byte re-order the constant
633 	 *	mask for efficiency.
634 	 */
635 	if (iph->frag_off & htons(IP_OFFSET))
636 		goto out;
637 
638 	/*
639 	 *	If we send an ICMP error to an ICMP error a mess would result..
640 	 */
641 	if (icmp_pointers[type].error) {
642 		/*
643 		 *	We are an error, check if we are replying to an
644 		 *	ICMP error
645 		 */
646 		if (iph->protocol == IPPROTO_ICMP) {
647 			u8 _inner_type, *itp;
648 
649 			itp = skb_header_pointer(skb_in,
650 						 skb_network_header(skb_in) +
651 						 (iph->ihl << 2) +
652 						 offsetof(struct icmphdr,
653 							  type) -
654 						 skb_in->data,
655 						 sizeof(_inner_type),
656 						 &_inner_type);
657 			if (!itp)
658 				goto out;
659 
660 			/*
661 			 *	Assume any unknown ICMP type is an error. This
662 			 *	isn't specified by the RFC, but think about it..
663 			 */
664 			if (*itp > NR_ICMP_TYPES ||
665 			    icmp_pointers[*itp].error)
666 				goto out;
667 		}
668 	}
669 
670 	/* Needed by both icmp_global_allow and icmp_xmit_lock */
671 	local_bh_disable();
672 
673 	/* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
674 	 * incoming dev is loopback.  If outgoing dev change to not be
675 	 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
676 	 */
677 	if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
678 	      !icmpv4_global_allow(net, type, code))
679 		goto out_bh_enable;
680 
681 	sk = icmp_xmit_lock(net);
682 	if (!sk)
683 		goto out_bh_enable;
684 
685 	/*
686 	 *	Construct source address and options.
687 	 */
688 
689 	saddr = iph->daddr;
690 	if (!(rt->rt_flags & RTCF_LOCAL)) {
691 		struct net_device *dev = NULL;
692 
693 		rcu_read_lock();
694 		if (rt_is_input_route(rt) &&
695 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
696 			dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
697 
698 		if (dev)
699 			saddr = inet_select_addr(dev, iph->saddr,
700 						 RT_SCOPE_LINK);
701 		else
702 			saddr = 0;
703 		rcu_read_unlock();
704 	}
705 
706 	tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
707 					   IPTOS_PREC_INTERNETCONTROL) :
708 					   iph->tos;
709 	mark = IP4_REPLY_MARK(net, skb_in->mark);
710 
711 	if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
712 		goto out_unlock;
713 
714 
715 	/*
716 	 *	Prepare data for ICMP header.
717 	 */
718 
719 	icmp_param.data.icmph.type	 = type;
720 	icmp_param.data.icmph.code	 = code;
721 	icmp_param.data.icmph.un.gateway = info;
722 	icmp_param.data.icmph.checksum	 = 0;
723 	icmp_param.skb	  = skb_in;
724 	icmp_param.offset = skb_network_offset(skb_in);
725 	inet_sk(sk)->tos = tos;
726 	ipcm_init(&ipc);
727 	ipc.addr = iph->saddr;
728 	ipc.opt = &icmp_param.replyopts.opt;
729 	ipc.sockc.mark = mark;
730 
731 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
732 			       type, code, &icmp_param);
733 	if (IS_ERR(rt))
734 		goto out_unlock;
735 
736 	/* peer icmp_ratelimit */
737 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
738 		goto ende;
739 
740 	/* RFC says return as much as we can without exceeding 576 bytes. */
741 
742 	room = dst_mtu(&rt->dst);
743 	if (room > 576)
744 		room = 576;
745 	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
746 	room -= sizeof(struct icmphdr);
747 
748 	icmp_param.data_len = skb_in->len - icmp_param.offset;
749 	if (icmp_param.data_len > room)
750 		icmp_param.data_len = room;
751 	icmp_param.head_len = sizeof(struct icmphdr);
752 
753 	/* if we don't have a source address at this point, fall back to the
754 	 * dummy address instead of sending out a packet with a source address
755 	 * of 0.0.0.0
756 	 */
757 	if (!fl4.saddr)
758 		fl4.saddr = htonl(INADDR_DUMMY);
759 
760 	icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
761 ende:
762 	ip_rt_put(rt);
763 out_unlock:
764 	icmp_xmit_unlock(sk);
765 out_bh_enable:
766 	local_bh_enable();
767 out:;
768 }
769 EXPORT_SYMBOL(__icmp_send);
770 
771 #if IS_ENABLED(CONFIG_NF_NAT)
772 #include <net/netfilter/nf_conntrack.h>
773 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
774 {
775 	struct sk_buff *cloned_skb = NULL;
776 	struct ip_options opts = { 0 };
777 	enum ip_conntrack_info ctinfo;
778 	struct nf_conn *ct;
779 	__be32 orig_ip;
780 
781 	ct = nf_ct_get(skb_in, &ctinfo);
782 	if (!ct || !(ct->status & IPS_SRC_NAT)) {
783 		__icmp_send(skb_in, type, code, info, &opts);
784 		return;
785 	}
786 
787 	if (skb_shared(skb_in))
788 		skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
789 
790 	if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
791 	    (skb_network_header(skb_in) + sizeof(struct iphdr)) >
792 	    skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
793 	    skb_network_offset(skb_in) + sizeof(struct iphdr))))
794 		goto out;
795 
796 	orig_ip = ip_hdr(skb_in)->saddr;
797 	ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
798 	__icmp_send(skb_in, type, code, info, &opts);
799 	ip_hdr(skb_in)->saddr = orig_ip;
800 out:
801 	consume_skb(cloned_skb);
802 }
803 EXPORT_SYMBOL(icmp_ndo_send);
804 #endif
805 
806 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
807 {
808 	const struct iphdr *iph = (const struct iphdr *)skb->data;
809 	const struct net_protocol *ipprot;
810 	int protocol = iph->protocol;
811 
812 	/* Checkin full IP header plus 8 bytes of protocol to
813 	 * avoid additional coding at protocol handlers.
814 	 */
815 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
816 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
817 		return;
818 	}
819 
820 	raw_icmp_error(skb, protocol, info);
821 
822 	ipprot = rcu_dereference(inet_protos[protocol]);
823 	if (ipprot && ipprot->err_handler)
824 		ipprot->err_handler(skb, info);
825 }
826 
827 static bool icmp_tag_validation(int proto)
828 {
829 	bool ok;
830 
831 	rcu_read_lock();
832 	ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
833 	rcu_read_unlock();
834 	return ok;
835 }
836 
837 /*
838  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
839  *	ICMP_PARAMETERPROB.
840  */
841 
842 static bool icmp_unreach(struct sk_buff *skb)
843 {
844 	const struct iphdr *iph;
845 	struct icmphdr *icmph;
846 	struct net *net;
847 	u32 info = 0;
848 
849 	net = dev_net(skb_dst(skb)->dev);
850 
851 	/*
852 	 *	Incomplete header ?
853 	 * 	Only checks for the IP header, there should be an
854 	 *	additional check for longer headers in upper levels.
855 	 */
856 
857 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
858 		goto out_err;
859 
860 	icmph = icmp_hdr(skb);
861 	iph   = (const struct iphdr *)skb->data;
862 
863 	if (iph->ihl < 5) /* Mangled header, drop. */
864 		goto out_err;
865 
866 	switch (icmph->type) {
867 	case ICMP_DEST_UNREACH:
868 		switch (icmph->code & 15) {
869 		case ICMP_NET_UNREACH:
870 		case ICMP_HOST_UNREACH:
871 		case ICMP_PROT_UNREACH:
872 		case ICMP_PORT_UNREACH:
873 			break;
874 		case ICMP_FRAG_NEEDED:
875 			/* for documentation of the ip_no_pmtu_disc
876 			 * values please see
877 			 * Documentation/networking/ip-sysctl.rst
878 			 */
879 			switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
880 			default:
881 				net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
882 						    &iph->daddr);
883 				break;
884 			case 2:
885 				goto out;
886 			case 3:
887 				if (!icmp_tag_validation(iph->protocol))
888 					goto out;
889 				fallthrough;
890 			case 0:
891 				info = ntohs(icmph->un.frag.mtu);
892 			}
893 			break;
894 		case ICMP_SR_FAILED:
895 			net_dbg_ratelimited("%pI4: Source Route Failed\n",
896 					    &iph->daddr);
897 			break;
898 		default:
899 			break;
900 		}
901 		if (icmph->code > NR_ICMP_UNREACH)
902 			goto out;
903 		break;
904 	case ICMP_PARAMETERPROB:
905 		info = ntohl(icmph->un.gateway) >> 24;
906 		break;
907 	case ICMP_TIME_EXCEEDED:
908 		__ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
909 		if (icmph->code == ICMP_EXC_FRAGTIME)
910 			goto out;
911 		break;
912 	}
913 
914 	/*
915 	 *	Throw it at our lower layers
916 	 *
917 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
918 	 *		  header.
919 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
920 	 *		  transport layer.
921 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
922 	 *		  transport layer.
923 	 */
924 
925 	/*
926 	 *	Check the other end isn't violating RFC 1122. Some routers send
927 	 *	bogus responses to broadcast frames. If you see this message
928 	 *	first check your netmask matches at both ends, if it does then
929 	 *	get the other vendor to fix their kit.
930 	 */
931 
932 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
933 	    inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
934 		net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
935 				     &ip_hdr(skb)->saddr,
936 				     icmph->type, icmph->code,
937 				     &iph->daddr, skb->dev->name);
938 		goto out;
939 	}
940 
941 	icmp_socket_deliver(skb, info);
942 
943 out:
944 	return true;
945 out_err:
946 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
947 	return false;
948 }
949 
950 
951 /*
952  *	Handle ICMP_REDIRECT.
953  */
954 
955 static bool icmp_redirect(struct sk_buff *skb)
956 {
957 	if (skb->len < sizeof(struct iphdr)) {
958 		__ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
959 		return false;
960 	}
961 
962 	if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
963 		/* there aught to be a stat */
964 		return false;
965 	}
966 
967 	icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
968 	return true;
969 }
970 
971 /*
972  *	Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
973  *
974  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
975  *		  requests.
976  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
977  *		  included in the reply.
978  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
979  *		  echo requests, MUST have default=NOT.
980  *	RFC 8335: 8 MUST have a config option to enable/disable ICMP
981  *		  Extended Echo Functionality, MUST be disabled by default
982  *	See also WRT handling of options once they are done and working.
983  */
984 
985 static bool icmp_echo(struct sk_buff *skb)
986 {
987 	struct icmp_bxm icmp_param;
988 	struct net *net;
989 
990 	net = dev_net(skb_dst(skb)->dev);
991 	/* should there be an ICMP stat for ignored echos? */
992 	if (net->ipv4.sysctl_icmp_echo_ignore_all)
993 		return true;
994 
995 	icmp_param.data.icmph	   = *icmp_hdr(skb);
996 	icmp_param.skb		   = skb;
997 	icmp_param.offset	   = 0;
998 	icmp_param.data_len	   = skb->len;
999 	icmp_param.head_len	   = sizeof(struct icmphdr);
1000 
1001 	if (icmp_param.data.icmph.type == ICMP_ECHO)
1002 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1003 	else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1004 		return true;
1005 
1006 	icmp_reply(&icmp_param, skb);
1007 	return true;
1008 }
1009 
1010 /*	Helper for icmp_echo and icmpv6_echo_reply.
1011  *	Searches for net_device that matches PROBE interface identifier
1012  *		and builds PROBE reply message in icmphdr.
1013  *
1014  *	Returns false if PROBE responses are disabled via sysctl
1015  */
1016 
1017 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1018 {
1019 	struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1020 	struct icmp_ext_echo_iio *iio, _iio;
1021 	struct net *net = dev_net(skb->dev);
1022 	struct net_device *dev;
1023 	char buff[IFNAMSIZ];
1024 	u16 ident_len;
1025 	u8 status;
1026 
1027 	if (!net->ipv4.sysctl_icmp_echo_enable_probe)
1028 		return false;
1029 
1030 	/* We currently only support probing interfaces on the proxy node
1031 	 * Check to ensure L-bit is set
1032 	 */
1033 	if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1034 		return false;
1035 	/* Clear status bits in reply message */
1036 	icmphdr->un.echo.sequence &= htons(0xFF00);
1037 	if (icmphdr->type == ICMP_EXT_ECHO)
1038 		icmphdr->type = ICMP_EXT_ECHOREPLY;
1039 	else
1040 		icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1041 	ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1042 	/* Size of iio is class_type dependent.
1043 	 * Only check header here and assign length based on ctype in the switch statement
1044 	 */
1045 	iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1046 	if (!ext_hdr || !iio)
1047 		goto send_mal_query;
1048 	if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1049 	    ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1050 		goto send_mal_query;
1051 	ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1052 	iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1053 				 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1054 	if (!iio)
1055 		goto send_mal_query;
1056 
1057 	status = 0;
1058 	dev = NULL;
1059 	switch (iio->extobj_hdr.class_type) {
1060 	case ICMP_EXT_ECHO_CTYPE_NAME:
1061 		if (ident_len >= IFNAMSIZ)
1062 			goto send_mal_query;
1063 		memset(buff, 0, sizeof(buff));
1064 		memcpy(buff, &iio->ident.name, ident_len);
1065 		dev = dev_get_by_name(net, buff);
1066 		break;
1067 	case ICMP_EXT_ECHO_CTYPE_INDEX:
1068 		if (ident_len != sizeof(iio->ident.ifindex))
1069 			goto send_mal_query;
1070 		dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1071 		break;
1072 	case ICMP_EXT_ECHO_CTYPE_ADDR:
1073 		if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1074 		    ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1075 				 iio->ident.addr.ctype3_hdr.addrlen)
1076 			goto send_mal_query;
1077 		switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1078 		case ICMP_AFI_IP:
1079 			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1080 				goto send_mal_query;
1081 			dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1082 			break;
1083 #if IS_ENABLED(CONFIG_IPV6)
1084 		case ICMP_AFI_IP6:
1085 			if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1086 				goto send_mal_query;
1087 			dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1088 			dev_hold(dev);
1089 			break;
1090 #endif
1091 		default:
1092 			goto send_mal_query;
1093 		}
1094 		break;
1095 	default:
1096 		goto send_mal_query;
1097 	}
1098 	if (!dev) {
1099 		icmphdr->code = ICMP_EXT_CODE_NO_IF;
1100 		return true;
1101 	}
1102 	/* Fill bits in reply message */
1103 	if (dev->flags & IFF_UP)
1104 		status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1105 	if (__in_dev_get_rcu(dev) && __in_dev_get_rcu(dev)->ifa_list)
1106 		status |= ICMP_EXT_ECHOREPLY_IPV4;
1107 	if (!list_empty(&rcu_dereference(dev->ip6_ptr)->addr_list))
1108 		status |= ICMP_EXT_ECHOREPLY_IPV6;
1109 	dev_put(dev);
1110 	icmphdr->un.echo.sequence |= htons(status);
1111 	return true;
1112 send_mal_query:
1113 	icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1114 	return true;
1115 }
1116 EXPORT_SYMBOL_GPL(icmp_build_probe);
1117 
1118 /*
1119  *	Handle ICMP Timestamp requests.
1120  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1121  *		  SHOULD be in the kernel for minimum random latency.
1122  *		  MUST be accurate to a few minutes.
1123  *		  MUST be updated at least at 15Hz.
1124  */
1125 static bool icmp_timestamp(struct sk_buff *skb)
1126 {
1127 	struct icmp_bxm icmp_param;
1128 	/*
1129 	 *	Too short.
1130 	 */
1131 	if (skb->len < 4)
1132 		goto out_err;
1133 
1134 	/*
1135 	 *	Fill in the current time as ms since midnight UT:
1136 	 */
1137 	icmp_param.data.times[1] = inet_current_timestamp();
1138 	icmp_param.data.times[2] = icmp_param.data.times[1];
1139 
1140 	BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1141 
1142 	icmp_param.data.icmph	   = *icmp_hdr(skb);
1143 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1144 	icmp_param.data.icmph.code = 0;
1145 	icmp_param.skb		   = skb;
1146 	icmp_param.offset	   = 0;
1147 	icmp_param.data_len	   = 0;
1148 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
1149 	icmp_reply(&icmp_param, skb);
1150 	return true;
1151 
1152 out_err:
1153 	__ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1154 	return false;
1155 }
1156 
1157 static bool icmp_discard(struct sk_buff *skb)
1158 {
1159 	/* pretend it was a success */
1160 	return true;
1161 }
1162 
1163 /*
1164  *	Deal with incoming ICMP packets.
1165  */
1166 int icmp_rcv(struct sk_buff *skb)
1167 {
1168 	struct icmphdr *icmph;
1169 	struct rtable *rt = skb_rtable(skb);
1170 	struct net *net = dev_net(rt->dst.dev);
1171 	bool success;
1172 
1173 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1174 		struct sec_path *sp = skb_sec_path(skb);
1175 		int nh;
1176 
1177 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1178 				 XFRM_STATE_ICMP))
1179 			goto drop;
1180 
1181 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1182 			goto drop;
1183 
1184 		nh = skb_network_offset(skb);
1185 		skb_set_network_header(skb, sizeof(*icmph));
1186 
1187 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1188 			goto drop;
1189 
1190 		skb_set_network_header(skb, nh);
1191 	}
1192 
1193 	__ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1194 
1195 	if (skb_checksum_simple_validate(skb))
1196 		goto csum_error;
1197 
1198 	if (!pskb_pull(skb, sizeof(*icmph)))
1199 		goto error;
1200 
1201 	icmph = icmp_hdr(skb);
1202 
1203 	ICMPMSGIN_INC_STATS(net, icmph->type);
1204 
1205 	/* Check for ICMP Extended Echo (PROBE) messages */
1206 	if (icmph->type == ICMP_EXT_ECHO) {
1207 		/* We can't use icmp_pointers[].handler() because it is an array of
1208 		 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1209 		 */
1210 		success = icmp_echo(skb);
1211 		goto success_check;
1212 	}
1213 
1214 	if (icmph->type == ICMP_EXT_ECHOREPLY) {
1215 		success = ping_rcv(skb);
1216 		goto success_check;
1217 	}
1218 
1219 	/*
1220 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1221 	 *
1222 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1223 	 *		  discarded.
1224 	 */
1225 	if (icmph->type > NR_ICMP_TYPES)
1226 		goto error;
1227 
1228 	/*
1229 	 *	Parse the ICMP message
1230 	 */
1231 
1232 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1233 		/*
1234 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1235 		 *	  silently ignored (we let user decide with a sysctl).
1236 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1237 		 *	  discarded if to broadcast/multicast.
1238 		 */
1239 		if ((icmph->type == ICMP_ECHO ||
1240 		     icmph->type == ICMP_TIMESTAMP) &&
1241 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1242 			goto error;
1243 		}
1244 		if (icmph->type != ICMP_ECHO &&
1245 		    icmph->type != ICMP_TIMESTAMP &&
1246 		    icmph->type != ICMP_ADDRESS &&
1247 		    icmph->type != ICMP_ADDRESSREPLY) {
1248 			goto error;
1249 		}
1250 	}
1251 
1252 	success = icmp_pointers[icmph->type].handler(skb);
1253 success_check:
1254 	if (success)  {
1255 		consume_skb(skb);
1256 		return NET_RX_SUCCESS;
1257 	}
1258 
1259 drop:
1260 	kfree_skb(skb);
1261 	return NET_RX_DROP;
1262 csum_error:
1263 	__ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1264 error:
1265 	__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1266 	goto drop;
1267 }
1268 
1269 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1270 {
1271 	struct icmp_extobj_hdr *objh, _objh;
1272 	struct icmp_ext_hdr *exth, _exth;
1273 	u16 olen;
1274 
1275 	exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1276 	if (!exth)
1277 		return false;
1278 	if (exth->version != 2)
1279 		return true;
1280 
1281 	if (exth->checksum &&
1282 	    csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1283 		return false;
1284 
1285 	off += sizeof(_exth);
1286 	while (off < skb->len) {
1287 		objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1288 		if (!objh)
1289 			return false;
1290 
1291 		olen = ntohs(objh->length);
1292 		if (olen < sizeof(_objh))
1293 			return false;
1294 
1295 		off += olen;
1296 		if (off > skb->len)
1297 			return false;
1298 	}
1299 
1300 	return true;
1301 }
1302 
1303 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1304 			   struct sock_ee_data_rfc4884 *out,
1305 			   int thlen, int off)
1306 {
1307 	int hlen;
1308 
1309 	/* original datagram headers: end of icmph to payload (skb->data) */
1310 	hlen = -skb_transport_offset(skb) - thlen;
1311 
1312 	/* per rfc 4884: minimal datagram length of 128 bytes */
1313 	if (off < 128 || off < hlen)
1314 		return;
1315 
1316 	/* kernel has stripped headers: return payload offset in bytes */
1317 	off -= hlen;
1318 	if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1319 		return;
1320 
1321 	out->len = off;
1322 
1323 	if (!ip_icmp_error_rfc4884_validate(skb, off))
1324 		out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1325 }
1326 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1327 
1328 int icmp_err(struct sk_buff *skb, u32 info)
1329 {
1330 	struct iphdr *iph = (struct iphdr *)skb->data;
1331 	int offset = iph->ihl<<2;
1332 	struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1333 	int type = icmp_hdr(skb)->type;
1334 	int code = icmp_hdr(skb)->code;
1335 	struct net *net = dev_net(skb->dev);
1336 
1337 	/*
1338 	 * Use ping_err to handle all icmp errors except those
1339 	 * triggered by ICMP_ECHOREPLY which sent from kernel.
1340 	 */
1341 	if (icmph->type != ICMP_ECHOREPLY) {
1342 		ping_err(skb, offset, info);
1343 		return 0;
1344 	}
1345 
1346 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1347 		ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1348 	else if (type == ICMP_REDIRECT)
1349 		ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1350 
1351 	return 0;
1352 }
1353 
1354 /*
1355  *	This table is the definition of how we handle ICMP.
1356  */
1357 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1358 	[ICMP_ECHOREPLY] = {
1359 		.handler = ping_rcv,
1360 	},
1361 	[1] = {
1362 		.handler = icmp_discard,
1363 		.error = 1,
1364 	},
1365 	[2] = {
1366 		.handler = icmp_discard,
1367 		.error = 1,
1368 	},
1369 	[ICMP_DEST_UNREACH] = {
1370 		.handler = icmp_unreach,
1371 		.error = 1,
1372 	},
1373 	[ICMP_SOURCE_QUENCH] = {
1374 		.handler = icmp_unreach,
1375 		.error = 1,
1376 	},
1377 	[ICMP_REDIRECT] = {
1378 		.handler = icmp_redirect,
1379 		.error = 1,
1380 	},
1381 	[6] = {
1382 		.handler = icmp_discard,
1383 		.error = 1,
1384 	},
1385 	[7] = {
1386 		.handler = icmp_discard,
1387 		.error = 1,
1388 	},
1389 	[ICMP_ECHO] = {
1390 		.handler = icmp_echo,
1391 	},
1392 	[9] = {
1393 		.handler = icmp_discard,
1394 		.error = 1,
1395 	},
1396 	[10] = {
1397 		.handler = icmp_discard,
1398 		.error = 1,
1399 	},
1400 	[ICMP_TIME_EXCEEDED] = {
1401 		.handler = icmp_unreach,
1402 		.error = 1,
1403 	},
1404 	[ICMP_PARAMETERPROB] = {
1405 		.handler = icmp_unreach,
1406 		.error = 1,
1407 	},
1408 	[ICMP_TIMESTAMP] = {
1409 		.handler = icmp_timestamp,
1410 	},
1411 	[ICMP_TIMESTAMPREPLY] = {
1412 		.handler = icmp_discard,
1413 	},
1414 	[ICMP_INFO_REQUEST] = {
1415 		.handler = icmp_discard,
1416 	},
1417 	[ICMP_INFO_REPLY] = {
1418 		.handler = icmp_discard,
1419 	},
1420 	[ICMP_ADDRESS] = {
1421 		.handler = icmp_discard,
1422 	},
1423 	[ICMP_ADDRESSREPLY] = {
1424 		.handler = icmp_discard,
1425 	},
1426 };
1427 
1428 static int __net_init icmp_sk_init(struct net *net)
1429 {
1430 	/* Control parameters for ECHO replies. */
1431 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1432 	net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1433 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1434 
1435 	/* Control parameter - ignore bogus broadcast responses? */
1436 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1437 
1438 	/*
1439 	 * 	Configurable global rate limit.
1440 	 *
1441 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1442 	 *	bucket ratemask defines which icmp types are ratelimited by
1443 	 *	setting	it's bit position.
1444 	 *
1445 	 *	default:
1446 	 *	dest unreachable (3), source quench (4),
1447 	 *	time exceeded (11), parameter problem (12)
1448 	 */
1449 
1450 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1451 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1452 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1453 
1454 	return 0;
1455 }
1456 
1457 static struct pernet_operations __net_initdata icmp_sk_ops = {
1458        .init = icmp_sk_init,
1459 };
1460 
1461 int __init icmp_init(void)
1462 {
1463 	int err, i;
1464 
1465 	for_each_possible_cpu(i) {
1466 		struct sock *sk;
1467 
1468 		err = inet_ctl_sock_create(&sk, PF_INET,
1469 					   SOCK_RAW, IPPROTO_ICMP, &init_net);
1470 		if (err < 0)
1471 			return err;
1472 
1473 		per_cpu(ipv4_icmp_sk, i) = sk;
1474 
1475 		/* Enough space for 2 64K ICMP packets, including
1476 		 * sk_buff/skb_shared_info struct overhead.
1477 		 */
1478 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1479 
1480 		/*
1481 		 * Speedup sock_wfree()
1482 		 */
1483 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1484 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1485 	}
1486 	return register_pernet_subsys(&icmp_sk_ops);
1487 }
1488