xref: /linux/net/ipv4/icmp.c (revision 52479b62)
1 /*
2  *	NET3:	Implementation of the ICMP protocol layer.
3  *
4  *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
5  *
6  *	This program is free software; you can redistribute it and/or
7  *	modify it under the terms of the GNU General Public License
8  *	as published by the Free Software Foundation; either version
9  *	2 of the License, or (at your option) any later version.
10  *
11  *	Some of the function names and the icmp unreach table for this
12  *	module were derived from [icmp.c 1.0.11 06/02/93] by
13  *	Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
14  *	Other than that this module is a complete rewrite.
15  *
16  *	Fixes:
17  *	Clemens Fruhwirth	:	introduce global icmp rate limiting
18  *					with icmp type masking ability instead
19  *					of broken per type icmp timeouts.
20  *		Mike Shaver	:	RFC1122 checks.
21  *		Alan Cox	:	Multicast ping reply as self.
22  *		Alan Cox	:	Fix atomicity lockup in ip_build_xmit
23  *					call.
24  *		Alan Cox	:	Added 216,128 byte paths to the MTU
25  *					code.
26  *		Martin Mares	:	RFC1812 checks.
27  *		Martin Mares	:	Can be configured to follow redirects
28  *					if acting as a router _without_ a
29  *					routing protocol (RFC 1812).
30  *		Martin Mares	:	Echo requests may be configured to
31  *					be ignored (RFC 1812).
32  *		Martin Mares	:	Limitation of ICMP error message
33  *					transmit rate (RFC 1812).
34  *		Martin Mares	:	TOS and Precedence set correctly
35  *					(RFC 1812).
36  *		Martin Mares	:	Now copying as much data from the
37  *					original packet as we can without
38  *					exceeding 576 bytes (RFC 1812).
39  *	Willy Konynenberg	:	Transparent proxying support.
40  *		Keith Owens	:	RFC1191 correction for 4.2BSD based
41  *					path MTU bug.
42  *		Thomas Quinot	:	ICMP Dest Unreach codes up to 15 are
43  *					valid (RFC 1812).
44  *		Andi Kleen	:	Check all packet lengths properly
45  *					and moved all kfree_skb() up to
46  *					icmp_rcv.
47  *		Andi Kleen	:	Move the rate limit bookkeeping
48  *					into the dest entry and use a token
49  *					bucket filter (thanks to ANK). Make
50  *					the rates sysctl configurable.
51  *		Yu Tianli	:	Fixed two ugly bugs in icmp_send
52  *					- IP option length was accounted wrongly
53  *					- ICMP header length was not accounted
54  *					  at all.
55  *              Tristan Greaves :       Added sysctl option to ignore bogus
56  *              			broadcast responses from broken routers.
57  *
58  * To Fix:
59  *
60  *	- Should use skb_pull() instead of all the manual checking.
61  *	  This would also greatly simply some upper layer error handlers. --AK
62  *
63  */
64 
65 #include <linux/module.h>
66 #include <linux/types.h>
67 #include <linux/jiffies.h>
68 #include <linux/kernel.h>
69 #include <linux/fcntl.h>
70 #include <linux/socket.h>
71 #include <linux/in.h>
72 #include <linux/inet.h>
73 #include <linux/inetdevice.h>
74 #include <linux/netdevice.h>
75 #include <linux/string.h>
76 #include <linux/netfilter_ipv4.h>
77 #include <net/snmp.h>
78 #include <net/ip.h>
79 #include <net/route.h>
80 #include <net/protocol.h>
81 #include <net/icmp.h>
82 #include <net/tcp.h>
83 #include <net/udp.h>
84 #include <net/raw.h>
85 #include <linux/skbuff.h>
86 #include <net/sock.h>
87 #include <linux/errno.h>
88 #include <linux/timer.h>
89 #include <linux/init.h>
90 #include <asm/system.h>
91 #include <asm/uaccess.h>
92 #include <net/checksum.h>
93 #include <net/xfrm.h>
94 #include <net/inet_common.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 replyopts;
111 	unsigned char  optbuf[40];
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 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 
184 /*
185  *	ICMP control array. This specifies what to do with each ICMP.
186  */
187 
188 struct icmp_control {
189 	void (*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 /*
196  *	The ICMP socket(s). This is the most convenient way to flow control
197  *	our ICMP output as well as maintain a clean interface throughout
198  *	all layers. All Socketless IP sends will soon be gone.
199  *
200  *	On SMP we have one ICMP socket per-cpu.
201  */
202 static struct sock *icmp_sk(struct net *net)
203 {
204 	return net->ipv4.icmp_sk[smp_processor_id()];
205 }
206 
207 static inline struct sock *icmp_xmit_lock(struct net *net)
208 {
209 	struct sock *sk;
210 
211 	local_bh_disable();
212 
213 	sk = icmp_sk(net);
214 
215 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
216 		/* This can happen if the output path signals a
217 		 * dst_link_failure() for an outgoing ICMP packet.
218 		 */
219 		local_bh_enable();
220 		return NULL;
221 	}
222 	return sk;
223 }
224 
225 static inline void icmp_xmit_unlock(struct sock *sk)
226 {
227 	spin_unlock_bh(&sk->sk_lock.slock);
228 }
229 
230 /*
231  *	Send an ICMP frame.
232  */
233 
234 /*
235  *	Check transmit rate limitation for given message.
236  *	The rate information is held in the destination cache now.
237  *	This function is generic and could be used for other purposes
238  *	too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
239  *
240  *	Note that the same dst_entry fields are modified by functions in
241  *	route.c too, but these work for packet destinations while xrlim_allow
242  *	works for icmp destinations. This means the rate limiting information
243  *	for one "ip object" is shared - and these ICMPs are twice limited:
244  *	by source and by destination.
245  *
246  *	RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
247  *			  SHOULD allow setting of rate limits
248  *
249  * 	Shared between ICMPv4 and ICMPv6.
250  */
251 #define XRLIM_BURST_FACTOR 6
252 int xrlim_allow(struct dst_entry *dst, int timeout)
253 {
254 	unsigned long now, token = dst->rate_tokens;
255 	int rc = 0;
256 
257 	now = jiffies;
258 	token += now - dst->rate_last;
259 	dst->rate_last = now;
260 	if (token > XRLIM_BURST_FACTOR * timeout)
261 		token = XRLIM_BURST_FACTOR * timeout;
262 	if (token >= timeout) {
263 		token -= timeout;
264 		rc = 1;
265 	}
266 	dst->rate_tokens = token;
267 	return rc;
268 }
269 
270 static inline int icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
271 		int type, int code)
272 {
273 	struct dst_entry *dst = &rt->u.dst;
274 	int rc = 1;
275 
276 	if (type > NR_ICMP_TYPES)
277 		goto out;
278 
279 	/* Don't limit PMTU discovery. */
280 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
281 		goto out;
282 
283 	/* No rate limit on loopback */
284 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
285 		goto out;
286 
287 	/* Limit if icmp type is enabled in ratemask. */
288 	if ((1 << type) & net->ipv4.sysctl_icmp_ratemask)
289 		rc = xrlim_allow(dst, net->ipv4.sysctl_icmp_ratelimit);
290 out:
291 	return rc;
292 }
293 
294 /*
295  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
296  */
297 void icmp_out_count(struct net *net, unsigned char type)
298 {
299 	ICMPMSGOUT_INC_STATS(net, type);
300 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
301 }
302 
303 /*
304  *	Checksum each fragment, and on the first include the headers and final
305  *	checksum.
306  */
307 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
308 			  struct sk_buff *skb)
309 {
310 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
311 	__wsum csum;
312 
313 	csum = skb_copy_and_csum_bits(icmp_param->skb,
314 				      icmp_param->offset + offset,
315 				      to, len, 0);
316 
317 	skb->csum = csum_block_add(skb->csum, csum, odd);
318 	if (icmp_pointers[icmp_param->data.icmph.type].error)
319 		nf_ct_attach(skb, icmp_param->skb);
320 	return 0;
321 }
322 
323 static void icmp_push_reply(struct icmp_bxm *icmp_param,
324 			    struct ipcm_cookie *ipc, struct rtable **rt)
325 {
326 	struct sock *sk;
327 	struct sk_buff *skb;
328 
329 	sk = icmp_sk(dev_net((*rt)->u.dst.dev));
330 	if (ip_append_data(sk, icmp_glue_bits, icmp_param,
331 			   icmp_param->data_len+icmp_param->head_len,
332 			   icmp_param->head_len,
333 			   ipc, rt, MSG_DONTWAIT) < 0)
334 		ip_flush_pending_frames(sk);
335 	else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
336 		struct icmphdr *icmph = icmp_hdr(skb);
337 		__wsum csum = 0;
338 		struct sk_buff *skb1;
339 
340 		skb_queue_walk(&sk->sk_write_queue, skb1) {
341 			csum = csum_add(csum, skb1->csum);
342 		}
343 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
344 						 (char *)icmph,
345 						 icmp_param->head_len, csum);
346 		icmph->checksum = csum_fold(csum);
347 		skb->ip_summed = CHECKSUM_NONE;
348 		ip_push_pending_frames(sk);
349 	}
350 }
351 
352 /*
353  *	Driving logic for building and sending ICMP messages.
354  */
355 
356 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
357 {
358 	struct ipcm_cookie ipc;
359 	struct rtable *rt = skb->rtable;
360 	struct net *net = dev_net(rt->u.dst.dev);
361 	struct sock *sk;
362 	struct inet_sock *inet;
363 	__be32 daddr;
364 
365 	if (ip_options_echo(&icmp_param->replyopts, skb))
366 		return;
367 
368 	sk = icmp_xmit_lock(net);
369 	if (sk == NULL)
370 		return;
371 	inet = inet_sk(sk);
372 
373 	icmp_param->data.icmph.checksum = 0;
374 
375 	inet->tos = ip_hdr(skb)->tos;
376 	daddr = ipc.addr = rt->rt_src;
377 	ipc.opt = NULL;
378 	if (icmp_param->replyopts.optlen) {
379 		ipc.opt = &icmp_param->replyopts;
380 		if (ipc.opt->srr)
381 			daddr = icmp_param->replyopts.faddr;
382 	}
383 	{
384 		struct flowi fl = { .nl_u = { .ip4_u =
385 					      { .daddr = daddr,
386 						.saddr = rt->rt_spec_dst,
387 						.tos = RT_TOS(ip_hdr(skb)->tos) } },
388 				    .proto = IPPROTO_ICMP };
389 		security_skb_classify_flow(skb, &fl);
390 		if (ip_route_output_key(net, &rt, &fl))
391 			goto out_unlock;
392 	}
393 	if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
394 			       icmp_param->data.icmph.code))
395 		icmp_push_reply(icmp_param, &ipc, &rt);
396 	ip_rt_put(rt);
397 out_unlock:
398 	icmp_xmit_unlock(sk);
399 }
400 
401 
402 /*
403  *	Send an ICMP message in response to a situation
404  *
405  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
406  *		  MAY send more (we do).
407  *			MUST NOT change this header information.
408  *			MUST NOT reply to a multicast/broadcast IP address.
409  *			MUST NOT reply to a multicast/broadcast MAC address.
410  *			MUST reply to only the first fragment.
411  */
412 
413 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
414 {
415 	struct iphdr *iph;
416 	int room;
417 	struct icmp_bxm icmp_param;
418 	struct rtable *rt = skb_in->rtable;
419 	struct ipcm_cookie ipc;
420 	__be32 saddr;
421 	u8  tos;
422 	struct net *net;
423 	struct sock *sk;
424 
425 	if (!rt)
426 		goto out;
427 	net = dev_net(rt->u.dst.dev);
428 
429 	/*
430 	 *	Find the original header. It is expected to be valid, of course.
431 	 *	Check this, icmp_send is called from the most obscure devices
432 	 *	sometimes.
433 	 */
434 	iph = ip_hdr(skb_in);
435 
436 	if ((u8 *)iph < skb_in->head ||
437 	    (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
438 		goto out;
439 
440 	/*
441 	 *	No replies to physical multicast/broadcast
442 	 */
443 	if (skb_in->pkt_type != PACKET_HOST)
444 		goto out;
445 
446 	/*
447 	 *	Now check at the protocol level
448 	 */
449 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
450 		goto out;
451 
452 	/*
453 	 *	Only reply to fragment 0. We byte re-order the constant
454 	 *	mask for efficiency.
455 	 */
456 	if (iph->frag_off & htons(IP_OFFSET))
457 		goto out;
458 
459 	/*
460 	 *	If we send an ICMP error to an ICMP error a mess would result..
461 	 */
462 	if (icmp_pointers[type].error) {
463 		/*
464 		 *	We are an error, check if we are replying to an
465 		 *	ICMP error
466 		 */
467 		if (iph->protocol == IPPROTO_ICMP) {
468 			u8 _inner_type, *itp;
469 
470 			itp = skb_header_pointer(skb_in,
471 						 skb_network_header(skb_in) +
472 						 (iph->ihl << 2) +
473 						 offsetof(struct icmphdr,
474 							  type) -
475 						 skb_in->data,
476 						 sizeof(_inner_type),
477 						 &_inner_type);
478 			if (itp == NULL)
479 				goto out;
480 
481 			/*
482 			 *	Assume any unknown ICMP type is an error. This
483 			 *	isn't specified by the RFC, but think about it..
484 			 */
485 			if (*itp > NR_ICMP_TYPES ||
486 			    icmp_pointers[*itp].error)
487 				goto out;
488 		}
489 	}
490 
491 	sk = icmp_xmit_lock(net);
492 	if (sk == NULL)
493 		return;
494 
495 	/*
496 	 *	Construct source address and options.
497 	 */
498 
499 	saddr = iph->daddr;
500 	if (!(rt->rt_flags & RTCF_LOCAL)) {
501 		struct net_device *dev = NULL;
502 
503 		if (rt->fl.iif &&
504 			net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
505 			dev = dev_get_by_index(net, rt->fl.iif);
506 
507 		if (dev) {
508 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
509 			dev_put(dev);
510 		} else
511 			saddr = 0;
512 	}
513 
514 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
515 					   IPTOS_PREC_INTERNETCONTROL) :
516 					  iph->tos;
517 
518 	if (ip_options_echo(&icmp_param.replyopts, skb_in))
519 		goto out_unlock;
520 
521 
522 	/*
523 	 *	Prepare data for ICMP header.
524 	 */
525 
526 	icmp_param.data.icmph.type	 = type;
527 	icmp_param.data.icmph.code	 = code;
528 	icmp_param.data.icmph.un.gateway = info;
529 	icmp_param.data.icmph.checksum	 = 0;
530 	icmp_param.skb	  = skb_in;
531 	icmp_param.offset = skb_network_offset(skb_in);
532 	inet_sk(sk)->tos = tos;
533 	ipc.addr = iph->saddr;
534 	ipc.opt = &icmp_param.replyopts;
535 
536 	{
537 		struct flowi fl = {
538 			.nl_u = {
539 				.ip4_u = {
540 					.daddr = icmp_param.replyopts.srr ?
541 						icmp_param.replyopts.faddr :
542 						iph->saddr,
543 					.saddr = saddr,
544 					.tos = RT_TOS(tos)
545 				}
546 			},
547 			.proto = IPPROTO_ICMP,
548 			.uli_u = {
549 				.icmpt = {
550 					.type = type,
551 					.code = code
552 				}
553 			}
554 		};
555 		int err;
556 		struct rtable *rt2;
557 
558 		security_skb_classify_flow(skb_in, &fl);
559 		if (__ip_route_output_key(net, &rt, &fl))
560 			goto out_unlock;
561 
562 		/* No need to clone since we're just using its address. */
563 		rt2 = rt;
564 
565 		err = xfrm_lookup(net, (struct dst_entry **)&rt, &fl, NULL, 0);
566 		switch (err) {
567 		case 0:
568 			if (rt != rt2)
569 				goto route_done;
570 			break;
571 		case -EPERM:
572 			rt = NULL;
573 			break;
574 		default:
575 			goto out_unlock;
576 		}
577 
578 		if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
579 			goto relookup_failed;
580 
581 		if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
582 			err = __ip_route_output_key(net, &rt2, &fl);
583 		else {
584 			struct flowi fl2 = {};
585 			struct dst_entry *odst;
586 
587 			fl2.fl4_dst = fl.fl4_src;
588 			if (ip_route_output_key(net, &rt2, &fl2))
589 				goto relookup_failed;
590 
591 			/* Ugh! */
592 			odst = skb_in->dst;
593 			err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
594 					     RT_TOS(tos), rt2->u.dst.dev);
595 
596 			dst_release(&rt2->u.dst);
597 			rt2 = skb_in->rtable;
598 			skb_in->dst = odst;
599 		}
600 
601 		if (err)
602 			goto relookup_failed;
603 
604 		err = xfrm_lookup(net, (struct dst_entry **)&rt2, &fl, NULL,
605 				  XFRM_LOOKUP_ICMP);
606 		switch (err) {
607 		case 0:
608 			dst_release(&rt->u.dst);
609 			rt = rt2;
610 			break;
611 		case -EPERM:
612 			goto ende;
613 		default:
614 relookup_failed:
615 			if (!rt)
616 				goto out_unlock;
617 			break;
618 		}
619 	}
620 
621 route_done:
622 	if (!icmpv4_xrlim_allow(net, rt, type, code))
623 		goto ende;
624 
625 	/* RFC says return as much as we can without exceeding 576 bytes. */
626 
627 	room = dst_mtu(&rt->u.dst);
628 	if (room > 576)
629 		room = 576;
630 	room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
631 	room -= sizeof(struct icmphdr);
632 
633 	icmp_param.data_len = skb_in->len - icmp_param.offset;
634 	if (icmp_param.data_len > room)
635 		icmp_param.data_len = room;
636 	icmp_param.head_len = sizeof(struct icmphdr);
637 
638 	icmp_push_reply(&icmp_param, &ipc, &rt);
639 ende:
640 	ip_rt_put(rt);
641 out_unlock:
642 	icmp_xmit_unlock(sk);
643 out:;
644 }
645 
646 
647 /*
648  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
649  */
650 
651 static void icmp_unreach(struct sk_buff *skb)
652 {
653 	struct iphdr *iph;
654 	struct icmphdr *icmph;
655 	int hash, protocol;
656 	struct net_protocol *ipprot;
657 	u32 info = 0;
658 	struct net *net;
659 
660 	net = dev_net(skb->dst->dev);
661 
662 	/*
663 	 *	Incomplete header ?
664 	 * 	Only checks for the IP header, there should be an
665 	 *	additional check for longer headers in upper levels.
666 	 */
667 
668 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
669 		goto out_err;
670 
671 	icmph = icmp_hdr(skb);
672 	iph   = (struct iphdr *)skb->data;
673 
674 	if (iph->ihl < 5) /* Mangled header, drop. */
675 		goto out_err;
676 
677 	if (icmph->type == ICMP_DEST_UNREACH) {
678 		switch (icmph->code & 15) {
679 		case ICMP_NET_UNREACH:
680 		case ICMP_HOST_UNREACH:
681 		case ICMP_PROT_UNREACH:
682 		case ICMP_PORT_UNREACH:
683 			break;
684 		case ICMP_FRAG_NEEDED:
685 			if (ipv4_config.no_pmtu_disc) {
686 				LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n",
687 					       &iph->daddr);
688 			} else {
689 				info = ip_rt_frag_needed(net, iph,
690 							 ntohs(icmph->un.frag.mtu),
691 							 skb->dev);
692 				if (!info)
693 					goto out;
694 			}
695 			break;
696 		case ICMP_SR_FAILED:
697 			LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n",
698 				       &iph->daddr);
699 			break;
700 		default:
701 			break;
702 		}
703 		if (icmph->code > NR_ICMP_UNREACH)
704 			goto out;
705 	} else if (icmph->type == ICMP_PARAMETERPROB)
706 		info = ntohl(icmph->un.gateway) >> 24;
707 
708 	/*
709 	 *	Throw it at our lower layers
710 	 *
711 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
712 	 *		  header.
713 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
714 	 *		  transport layer.
715 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
716 	 *		  transport layer.
717 	 */
718 
719 	/*
720 	 *	Check the other end isnt violating RFC 1122. Some routers send
721 	 *	bogus responses to broadcast frames. If you see this message
722 	 *	first check your netmask matches at both ends, if it does then
723 	 *	get the other vendor to fix their kit.
724 	 */
725 
726 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
727 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
728 		if (net_ratelimit())
729 			printk(KERN_WARNING "%pI4 sent an invalid ICMP "
730 					    "type %u, code %u "
731 					    "error to a broadcast: %pI4 on %s\n",
732 			       &ip_hdr(skb)->saddr,
733 			       icmph->type, icmph->code,
734 			       &iph->daddr,
735 			       skb->dev->name);
736 		goto out;
737 	}
738 
739 	/* Checkin full IP header plus 8 bytes of protocol to
740 	 * avoid additional coding at protocol handlers.
741 	 */
742 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
743 		goto out;
744 
745 	iph = (struct iphdr *)skb->data;
746 	protocol = iph->protocol;
747 
748 	/*
749 	 *	Deliver ICMP message to raw sockets. Pretty useless feature?
750 	 */
751 	raw_icmp_error(skb, protocol, info);
752 
753 	hash = protocol & (MAX_INET_PROTOS - 1);
754 	rcu_read_lock();
755 	ipprot = rcu_dereference(inet_protos[hash]);
756 	if (ipprot && ipprot->err_handler)
757 		ipprot->err_handler(skb, info);
758 	rcu_read_unlock();
759 
760 out:
761 	return;
762 out_err:
763 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
764 	goto out;
765 }
766 
767 
768 /*
769  *	Handle ICMP_REDIRECT.
770  */
771 
772 static void icmp_redirect(struct sk_buff *skb)
773 {
774 	struct iphdr *iph;
775 
776 	if (skb->len < sizeof(struct iphdr))
777 		goto out_err;
778 
779 	/*
780 	 *	Get the copied header of the packet that caused the redirect
781 	 */
782 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
783 		goto out;
784 
785 	iph = (struct iphdr *)skb->data;
786 
787 	switch (icmp_hdr(skb)->code & 7) {
788 	case ICMP_REDIR_NET:
789 	case ICMP_REDIR_NETTOS:
790 		/*
791 		 * As per RFC recommendations now handle it as a host redirect.
792 		 */
793 	case ICMP_REDIR_HOST:
794 	case ICMP_REDIR_HOSTTOS:
795 		ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
796 			       icmp_hdr(skb)->un.gateway,
797 			       iph->saddr, skb->dev);
798 		break;
799 	}
800 out:
801 	return;
802 out_err:
803 	ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
804 	goto out;
805 }
806 
807 /*
808  *	Handle ICMP_ECHO ("ping") requests.
809  *
810  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
811  *		  requests.
812  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
813  *		  included in the reply.
814  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
815  *		  echo requests, MUST have default=NOT.
816  *	See also WRT handling of options once they are done and working.
817  */
818 
819 static void icmp_echo(struct sk_buff *skb)
820 {
821 	struct net *net;
822 
823 	net = dev_net(skb->dst->dev);
824 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
825 		struct icmp_bxm icmp_param;
826 
827 		icmp_param.data.icmph	   = *icmp_hdr(skb);
828 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
829 		icmp_param.skb		   = skb;
830 		icmp_param.offset	   = 0;
831 		icmp_param.data_len	   = skb->len;
832 		icmp_param.head_len	   = sizeof(struct icmphdr);
833 		icmp_reply(&icmp_param, skb);
834 	}
835 }
836 
837 /*
838  *	Handle ICMP Timestamp requests.
839  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
840  *		  SHOULD be in the kernel for minimum random latency.
841  *		  MUST be accurate to a few minutes.
842  *		  MUST be updated at least at 15Hz.
843  */
844 static void icmp_timestamp(struct sk_buff *skb)
845 {
846 	struct timespec tv;
847 	struct icmp_bxm icmp_param;
848 	/*
849 	 *	Too short.
850 	 */
851 	if (skb->len < 4)
852 		goto out_err;
853 
854 	/*
855 	 *	Fill in the current time as ms since midnight UT:
856 	 */
857 	getnstimeofday(&tv);
858 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
859 					 tv.tv_nsec / NSEC_PER_MSEC);
860 	icmp_param.data.times[2] = icmp_param.data.times[1];
861 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
862 		BUG();
863 	icmp_param.data.icmph	   = *icmp_hdr(skb);
864 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
865 	icmp_param.data.icmph.code = 0;
866 	icmp_param.skb		   = skb;
867 	icmp_param.offset	   = 0;
868 	icmp_param.data_len	   = 0;
869 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
870 	icmp_reply(&icmp_param, skb);
871 out:
872 	return;
873 out_err:
874 	ICMP_INC_STATS_BH(dev_net(skb->dst->dev), ICMP_MIB_INERRORS);
875 	goto out;
876 }
877 
878 
879 /*
880  *	Handle ICMP_ADDRESS_MASK requests.  (RFC950)
881  *
882  * RFC1122 (3.2.2.9).  A host MUST only send replies to
883  * ADDRESS_MASK requests if it's been configured as an address mask
884  * agent.  Receiving a request doesn't constitute implicit permission to
885  * act as one. Of course, implementing this correctly requires (SHOULD)
886  * a way to turn the functionality on and off.  Another one for sysctl(),
887  * I guess. -- MS
888  *
889  * RFC1812 (4.3.3.9).	A router MUST implement it.
890  *			A router SHOULD have switch turning it on/off.
891  *		      	This switch MUST be ON by default.
892  *
893  * Gratuitous replies, zero-source replies are not implemented,
894  * that complies with RFC. DO NOT implement them!!! All the idea
895  * of broadcast addrmask replies as specified in RFC950 is broken.
896  * The problem is that it is not uncommon to have several prefixes
897  * on one physical interface. Moreover, addrmask agent can even be
898  * not aware of existing another prefixes.
899  * If source is zero, addrmask agent cannot choose correct prefix.
900  * Gratuitous mask announcements suffer from the same problem.
901  * RFC1812 explains it, but still allows to use ADDRMASK,
902  * that is pretty silly. --ANK
903  *
904  * All these rules are so bizarre, that I removed kernel addrmask
905  * support at all. It is wrong, it is obsolete, nobody uses it in
906  * any case. --ANK
907  *
908  * Furthermore you can do it with a usermode address agent program
909  * anyway...
910  */
911 
912 static void icmp_address(struct sk_buff *skb)
913 {
914 #if 0
915 	if (net_ratelimit())
916 		printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
917 #endif
918 }
919 
920 /*
921  * RFC1812 (4.3.3.9).	A router SHOULD listen all replies, and complain
922  *			loudly if an inconsistency is found.
923  */
924 
925 static void icmp_address_reply(struct sk_buff *skb)
926 {
927 	struct rtable *rt = skb->rtable;
928 	struct net_device *dev = skb->dev;
929 	struct in_device *in_dev;
930 	struct in_ifaddr *ifa;
931 
932 	if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
933 		goto out;
934 
935 	in_dev = in_dev_get(dev);
936 	if (!in_dev)
937 		goto out;
938 	rcu_read_lock();
939 	if (in_dev->ifa_list &&
940 	    IN_DEV_LOG_MARTIANS(in_dev) &&
941 	    IN_DEV_FORWARD(in_dev)) {
942 		__be32 _mask, *mp;
943 
944 		mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
945 		BUG_ON(mp == NULL);
946 		for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
947 			if (*mp == ifa->ifa_mask &&
948 			    inet_ifa_match(rt->rt_src, ifa))
949 				break;
950 		}
951 		if (!ifa && net_ratelimit()) {
952 			printk(KERN_INFO "Wrong address mask %pI4 from %s/%pI4\n",
953 			       mp, dev->name, &rt->rt_src);
954 		}
955 	}
956 	rcu_read_unlock();
957 	in_dev_put(in_dev);
958 out:;
959 }
960 
961 static void icmp_discard(struct sk_buff *skb)
962 {
963 }
964 
965 /*
966  *	Deal with incoming ICMP packets.
967  */
968 int icmp_rcv(struct sk_buff *skb)
969 {
970 	struct icmphdr *icmph;
971 	struct rtable *rt = skb->rtable;
972 	struct net *net = dev_net(rt->u.dst.dev);
973 
974 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
975 		struct sec_path *sp = skb_sec_path(skb);
976 		int nh;
977 
978 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
979 				 XFRM_STATE_ICMP))
980 			goto drop;
981 
982 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
983 			goto drop;
984 
985 		nh = skb_network_offset(skb);
986 		skb_set_network_header(skb, sizeof(*icmph));
987 
988 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
989 			goto drop;
990 
991 		skb_set_network_header(skb, nh);
992 	}
993 
994 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
995 
996 	switch (skb->ip_summed) {
997 	case CHECKSUM_COMPLETE:
998 		if (!csum_fold(skb->csum))
999 			break;
1000 		/* fall through */
1001 	case CHECKSUM_NONE:
1002 		skb->csum = 0;
1003 		if (__skb_checksum_complete(skb))
1004 			goto error;
1005 	}
1006 
1007 	if (!pskb_pull(skb, sizeof(*icmph)))
1008 		goto error;
1009 
1010 	icmph = icmp_hdr(skb);
1011 
1012 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1013 	/*
1014 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1015 	 *
1016 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1017 	 *		  discarded.
1018 	 */
1019 	if (icmph->type > NR_ICMP_TYPES)
1020 		goto error;
1021 
1022 
1023 	/*
1024 	 *	Parse the ICMP message
1025 	 */
1026 
1027 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1028 		/*
1029 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1030 		 *	  silently ignored (we let user decide with a sysctl).
1031 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1032 		 *	  discarded if to broadcast/multicast.
1033 		 */
1034 		if ((icmph->type == ICMP_ECHO ||
1035 		     icmph->type == ICMP_TIMESTAMP) &&
1036 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1037 			goto error;
1038 		}
1039 		if (icmph->type != ICMP_ECHO &&
1040 		    icmph->type != ICMP_TIMESTAMP &&
1041 		    icmph->type != ICMP_ADDRESS &&
1042 		    icmph->type != ICMP_ADDRESSREPLY) {
1043 			goto error;
1044 		}
1045 	}
1046 
1047 	icmp_pointers[icmph->type].handler(skb);
1048 
1049 drop:
1050 	kfree_skb(skb);
1051 	return 0;
1052 error:
1053 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1054 	goto drop;
1055 }
1056 
1057 /*
1058  *	This table is the definition of how we handle ICMP.
1059  */
1060 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1061 	[ICMP_ECHOREPLY] = {
1062 		.handler = icmp_discard,
1063 	},
1064 	[1] = {
1065 		.handler = icmp_discard,
1066 		.error = 1,
1067 	},
1068 	[2] = {
1069 		.handler = icmp_discard,
1070 		.error = 1,
1071 	},
1072 	[ICMP_DEST_UNREACH] = {
1073 		.handler = icmp_unreach,
1074 		.error = 1,
1075 	},
1076 	[ICMP_SOURCE_QUENCH] = {
1077 		.handler = icmp_unreach,
1078 		.error = 1,
1079 	},
1080 	[ICMP_REDIRECT] = {
1081 		.handler = icmp_redirect,
1082 		.error = 1,
1083 	},
1084 	[6] = {
1085 		.handler = icmp_discard,
1086 		.error = 1,
1087 	},
1088 	[7] = {
1089 		.handler = icmp_discard,
1090 		.error = 1,
1091 	},
1092 	[ICMP_ECHO] = {
1093 		.handler = icmp_echo,
1094 	},
1095 	[9] = {
1096 		.handler = icmp_discard,
1097 		.error = 1,
1098 	},
1099 	[10] = {
1100 		.handler = icmp_discard,
1101 		.error = 1,
1102 	},
1103 	[ICMP_TIME_EXCEEDED] = {
1104 		.handler = icmp_unreach,
1105 		.error = 1,
1106 	},
1107 	[ICMP_PARAMETERPROB] = {
1108 		.handler = icmp_unreach,
1109 		.error = 1,
1110 	},
1111 	[ICMP_TIMESTAMP] = {
1112 		.handler = icmp_timestamp,
1113 	},
1114 	[ICMP_TIMESTAMPREPLY] = {
1115 		.handler = icmp_discard,
1116 	},
1117 	[ICMP_INFO_REQUEST] = {
1118 		.handler = icmp_discard,
1119 	},
1120 	[ICMP_INFO_REPLY] = {
1121 		.handler = icmp_discard,
1122 	},
1123 	[ICMP_ADDRESS] = {
1124 		.handler = icmp_address,
1125 	},
1126 	[ICMP_ADDRESSREPLY] = {
1127 		.handler = icmp_address_reply,
1128 	},
1129 };
1130 
1131 static void __net_exit icmp_sk_exit(struct net *net)
1132 {
1133 	int i;
1134 
1135 	for_each_possible_cpu(i)
1136 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1137 	kfree(net->ipv4.icmp_sk);
1138 	net->ipv4.icmp_sk = NULL;
1139 }
1140 
1141 static int __net_init icmp_sk_init(struct net *net)
1142 {
1143 	int i, err;
1144 
1145 	net->ipv4.icmp_sk =
1146 		kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1147 	if (net->ipv4.icmp_sk == NULL)
1148 		return -ENOMEM;
1149 
1150 	for_each_possible_cpu(i) {
1151 		struct sock *sk;
1152 
1153 		err = inet_ctl_sock_create(&sk, PF_INET,
1154 					   SOCK_RAW, IPPROTO_ICMP, net);
1155 		if (err < 0)
1156 			goto fail;
1157 
1158 		net->ipv4.icmp_sk[i] = sk;
1159 
1160 		/* Enough space for 2 64K ICMP packets, including
1161 		 * sk_buff struct overhead.
1162 		 */
1163 		sk->sk_sndbuf =
1164 			(2 * ((64 * 1024) + sizeof(struct sk_buff)));
1165 
1166 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1167 	}
1168 
1169 	/* Control parameters for ECHO replies. */
1170 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1171 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1172 
1173 	/* Control parameter - ignore bogus broadcast responses? */
1174 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1175 
1176 	/*
1177 	 * 	Configurable global rate limit.
1178 	 *
1179 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1180 	 *	bucket ratemask defines which icmp types are ratelimited by
1181 	 *	setting	it's bit position.
1182 	 *
1183 	 *	default:
1184 	 *	dest unreachable (3), source quench (4),
1185 	 *	time exceeded (11), parameter problem (12)
1186 	 */
1187 
1188 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1189 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1190 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1191 
1192 	return 0;
1193 
1194 fail:
1195 	for_each_possible_cpu(i)
1196 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1197 	kfree(net->ipv4.icmp_sk);
1198 	return err;
1199 }
1200 
1201 static struct pernet_operations __net_initdata icmp_sk_ops = {
1202        .init = icmp_sk_init,
1203        .exit = icmp_sk_exit,
1204 };
1205 
1206 int __init icmp_init(void)
1207 {
1208 	return register_pernet_device(&icmp_sk_ops);
1209 }
1210 
1211 EXPORT_SYMBOL(icmp_err_convert);
1212 EXPORT_SYMBOL(icmp_send);
1213 EXPORT_SYMBOL(xrlim_allow);
1214