xref: /linux/net/ipv4/icmp.c (revision ed3174d93c342b8b2eeba6bbd124707d55304a7b)
1 /*
2  *	NET3:	Implementation of the ICMP protocol layer.
3  *
4  *		Alan Cox, <alan@redhat.com>
5  *
6  *	Version: $Id: icmp.c,v 1.85 2002/02/01 22:01:03 davem Exp $
7  *
8  *	This program is free software; you can redistribute it and/or
9  *	modify it under the terms of the GNU General Public License
10  *	as published by the Free Software Foundation; either version
11  *	2 of the License, or (at your option) any later version.
12  *
13  *	Some of the function names and the icmp unreach table for this
14  *	module were derived from [icmp.c 1.0.11 06/02/93] by
15  *	Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
16  *	Other than that this module is a complete rewrite.
17  *
18  *	Fixes:
19  *	Clemens Fruhwirth	:	introduce global icmp rate limiting
20  *					with icmp type masking ability instead
21  *					of broken per type icmp timeouts.
22  *		Mike Shaver	:	RFC1122 checks.
23  *		Alan Cox	:	Multicast ping reply as self.
24  *		Alan Cox	:	Fix atomicity lockup in ip_build_xmit
25  *					call.
26  *		Alan Cox	:	Added 216,128 byte paths to the MTU
27  *					code.
28  *		Martin Mares	:	RFC1812 checks.
29  *		Martin Mares	:	Can be configured to follow redirects
30  *					if acting as a router _without_ a
31  *					routing protocol (RFC 1812).
32  *		Martin Mares	:	Echo requests may be configured to
33  *					be ignored (RFC 1812).
34  *		Martin Mares	:	Limitation of ICMP error message
35  *					transmit rate (RFC 1812).
36  *		Martin Mares	:	TOS and Precedence set correctly
37  *					(RFC 1812).
38  *		Martin Mares	:	Now copying as much data from the
39  *					original packet as we can without
40  *					exceeding 576 bytes (RFC 1812).
41  *	Willy Konynenberg	:	Transparent proxying support.
42  *		Keith Owens	:	RFC1191 correction for 4.2BSD based
43  *					path MTU bug.
44  *		Thomas Quinot	:	ICMP Dest Unreach codes up to 15 are
45  *					valid (RFC 1812).
46  *		Andi Kleen	:	Check all packet lengths properly
47  *					and moved all kfree_skb() up to
48  *					icmp_rcv.
49  *		Andi Kleen	:	Move the rate limit bookkeeping
50  *					into the dest entry and use a token
51  *					bucket filter (thanks to ANK). Make
52  *					the rates sysctl configurable.
53  *		Yu Tianli	:	Fixed two ugly bugs in icmp_send
54  *					- IP option length was accounted wrongly
55  *					- ICMP header length was not accounted
56  *					  at all.
57  *              Tristan Greaves :       Added sysctl option to ignore bogus
58  *              			broadcast responses from broken routers.
59  *
60  * To Fix:
61  *
62  *	- Should use skb_pull() instead of all the manual checking.
63  *	  This would also greatly simply some upper layer error handlers. --AK
64  *
65  */
66 
67 #include <linux/module.h>
68 #include <linux/types.h>
69 #include <linux/jiffies.h>
70 #include <linux/kernel.h>
71 #include <linux/fcntl.h>
72 #include <linux/socket.h>
73 #include <linux/in.h>
74 #include <linux/inet.h>
75 #include <linux/inetdevice.h>
76 #include <linux/netdevice.h>
77 #include <linux/string.h>
78 #include <linux/netfilter_ipv4.h>
79 #include <net/snmp.h>
80 #include <net/ip.h>
81 #include <net/route.h>
82 #include <net/protocol.h>
83 #include <net/icmp.h>
84 #include <net/tcp.h>
85 #include <net/udp.h>
86 #include <net/raw.h>
87 #include <linux/skbuff.h>
88 #include <net/sock.h>
89 #include <linux/errno.h>
90 #include <linux/timer.h>
91 #include <linux/init.h>
92 #include <asm/system.h>
93 #include <asm/uaccess.h>
94 #include <net/checksum.h>
95 #include <net/xfrm.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 replyopts;
112 	unsigned char  optbuf[40];
113 };
114 
115 /*
116  *	Statistics
117  */
118 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly;
119 DEFINE_SNMP_STAT(struct icmpmsg_mib, icmpmsg_statistics) __read_mostly;
120 
121 /* An array of errno for error messages from dest unreach. */
122 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
123 
124 struct icmp_err icmp_err_convert[] = {
125 	{
126 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
127 		.fatal = 0,
128 	},
129 	{
130 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
131 		.fatal = 0,
132 	},
133 	{
134 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
135 		.fatal = 1,
136 	},
137 	{
138 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
139 		.fatal = 1,
140 	},
141 	{
142 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
143 		.fatal = 0,
144 	},
145 	{
146 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
147 		.fatal = 0,
148 	},
149 	{
150 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
151 		.fatal = 1,
152 	},
153 	{
154 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
155 		.fatal = 1,
156 	},
157 	{
158 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
159 		.fatal = 1,
160 	},
161 	{
162 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
163 		.fatal = 1,
164 	},
165 	{
166 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
167 		.fatal = 1,
168 	},
169 	{
170 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
171 		.fatal = 0,
172 	},
173 	{
174 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
175 		.fatal = 0,
176 	},
177 	{
178 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
179 		.fatal = 1,
180 	},
181 	{
182 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
183 		.fatal = 1,
184 	},
185 	{
186 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
187 		.fatal = 1,
188 	},
189 };
190 
191 /* Control parameters for ECHO replies. */
192 int sysctl_icmp_echo_ignore_all __read_mostly;
193 int sysctl_icmp_echo_ignore_broadcasts __read_mostly = 1;
194 
195 /* Control parameter - ignore bogus broadcast responses? */
196 int sysctl_icmp_ignore_bogus_error_responses __read_mostly = 1;
197 
198 /*
199  * 	Configurable global rate limit.
200  *
201  *	ratelimit defines tokens/packet consumed for dst->rate_token bucket
202  *	ratemask defines which icmp types are ratelimited by setting
203  * 	it's bit position.
204  *
205  *	default:
206  *	dest unreachable (3), source quench (4),
207  *	time exceeded (11), parameter problem (12)
208  */
209 
210 int sysctl_icmp_ratelimit __read_mostly = 1 * HZ;
211 int sysctl_icmp_ratemask __read_mostly = 0x1818;
212 int sysctl_icmp_errors_use_inbound_ifaddr __read_mostly;
213 
214 /*
215  *	ICMP control array. This specifies what to do with each ICMP.
216  */
217 
218 struct icmp_control {
219 	void (*handler)(struct sk_buff *skb);
220 	short   error;		/* This ICMP is classed as an error message */
221 };
222 
223 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
224 
225 /*
226  *	The ICMP socket(s). This is the most convenient way to flow control
227  *	our ICMP output as well as maintain a clean interface throughout
228  *	all layers. All Socketless IP sends will soon be gone.
229  *
230  *	On SMP we have one ICMP socket per-cpu.
231  */
232 static DEFINE_PER_CPU(struct socket *, __icmp_socket) = NULL;
233 #define icmp_socket	__get_cpu_var(__icmp_socket)
234 
235 static inline int icmp_xmit_lock(void)
236 {
237 	local_bh_disable();
238 
239 	if (unlikely(!spin_trylock(&icmp_socket->sk->sk_lock.slock))) {
240 		/* This can happen if the output path signals a
241 		 * dst_link_failure() for an outgoing ICMP packet.
242 		 */
243 		local_bh_enable();
244 		return 1;
245 	}
246 	return 0;
247 }
248 
249 static inline void icmp_xmit_unlock(void)
250 {
251 	spin_unlock_bh(&icmp_socket->sk->sk_lock.slock);
252 }
253 
254 /*
255  *	Send an ICMP frame.
256  */
257 
258 /*
259  *	Check transmit rate limitation for given message.
260  *	The rate information is held in the destination cache now.
261  *	This function is generic and could be used for other purposes
262  *	too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
263  *
264  *	Note that the same dst_entry fields are modified by functions in
265  *	route.c too, but these work for packet destinations while xrlim_allow
266  *	works for icmp destinations. This means the rate limiting information
267  *	for one "ip object" is shared - and these ICMPs are twice limited:
268  *	by source and by destination.
269  *
270  *	RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
271  *			  SHOULD allow setting of rate limits
272  *
273  * 	Shared between ICMPv4 and ICMPv6.
274  */
275 #define XRLIM_BURST_FACTOR 6
276 int xrlim_allow(struct dst_entry *dst, int timeout)
277 {
278 	unsigned long now, token = dst->rate_tokens;
279 	int rc = 0;
280 
281 	now = jiffies;
282 	token += now - dst->rate_last;
283 	dst->rate_last = now;
284 	if (token > XRLIM_BURST_FACTOR * timeout)
285 		token = XRLIM_BURST_FACTOR * timeout;
286 	if (token >= timeout) {
287 		token -= timeout;
288 		rc = 1;
289 	}
290 	dst->rate_tokens = token;
291 	return rc;
292 }
293 
294 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
295 {
296 	struct dst_entry *dst = &rt->u.dst;
297 	int rc = 1;
298 
299 	if (type > NR_ICMP_TYPES)
300 		goto out;
301 
302 	/* Don't limit PMTU discovery. */
303 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
304 		goto out;
305 
306 	/* No rate limit on loopback */
307 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
308 		goto out;
309 
310 	/* Limit if icmp type is enabled in ratemask. */
311 	if ((1 << type) & sysctl_icmp_ratemask)
312 		rc = xrlim_allow(dst, sysctl_icmp_ratelimit);
313 out:
314 	return rc;
315 }
316 
317 /*
318  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
319  */
320 void icmp_out_count(unsigned char type)
321 {
322 	ICMPMSGOUT_INC_STATS(type);
323 	ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
324 }
325 
326 /*
327  *	Checksum each fragment, and on the first include the headers and final
328  *	checksum.
329  */
330 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
331 			  struct sk_buff *skb)
332 {
333 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
334 	__wsum csum;
335 
336 	csum = skb_copy_and_csum_bits(icmp_param->skb,
337 				      icmp_param->offset + offset,
338 				      to, len, 0);
339 
340 	skb->csum = csum_block_add(skb->csum, csum, odd);
341 	if (icmp_pointers[icmp_param->data.icmph.type].error)
342 		nf_ct_attach(skb, icmp_param->skb);
343 	return 0;
344 }
345 
346 static void icmp_push_reply(struct icmp_bxm *icmp_param,
347 			    struct ipcm_cookie *ipc, struct rtable *rt)
348 {
349 	struct sk_buff *skb;
350 
351 	if (ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param,
352 			   icmp_param->data_len+icmp_param->head_len,
353 			   icmp_param->head_len,
354 			   ipc, rt, MSG_DONTWAIT) < 0)
355 		ip_flush_pending_frames(icmp_socket->sk);
356 	else if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) {
357 		struct icmphdr *icmph = icmp_hdr(skb);
358 		__wsum csum = 0;
359 		struct sk_buff *skb1;
360 
361 		skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) {
362 			csum = csum_add(csum, skb1->csum);
363 		}
364 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
365 						 (char *)icmph,
366 						 icmp_param->head_len, csum);
367 		icmph->checksum = csum_fold(csum);
368 		skb->ip_summed = CHECKSUM_NONE;
369 		ip_push_pending_frames(icmp_socket->sk);
370 	}
371 }
372 
373 /*
374  *	Driving logic for building and sending ICMP messages.
375  */
376 
377 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
378 {
379 	struct sock *sk = icmp_socket->sk;
380 	struct inet_sock *inet = inet_sk(sk);
381 	struct ipcm_cookie ipc;
382 	struct rtable *rt = (struct rtable *)skb->dst;
383 	__be32 daddr;
384 
385 	if (ip_options_echo(&icmp_param->replyopts, skb))
386 		return;
387 
388 	if (icmp_xmit_lock())
389 		return;
390 
391 	icmp_param->data.icmph.checksum = 0;
392 
393 	inet->tos = ip_hdr(skb)->tos;
394 	daddr = ipc.addr = rt->rt_src;
395 	ipc.opt = NULL;
396 	if (icmp_param->replyopts.optlen) {
397 		ipc.opt = &icmp_param->replyopts;
398 		if (ipc.opt->srr)
399 			daddr = icmp_param->replyopts.faddr;
400 	}
401 	{
402 		struct flowi fl = { .nl_u = { .ip4_u =
403 					      { .daddr = daddr,
404 						.saddr = rt->rt_spec_dst,
405 						.tos = RT_TOS(ip_hdr(skb)->tos) } },
406 				    .proto = IPPROTO_ICMP };
407 		security_skb_classify_flow(skb, &fl);
408 		if (ip_route_output_key(rt->u.dst.dev->nd_net, &rt, &fl))
409 			goto out_unlock;
410 	}
411 	if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
412 			       icmp_param->data.icmph.code))
413 		icmp_push_reply(icmp_param, &ipc, rt);
414 	ip_rt_put(rt);
415 out_unlock:
416 	icmp_xmit_unlock();
417 }
418 
419 
420 /*
421  *	Send an ICMP message in response to a situation
422  *
423  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
424  *		  MAY send more (we do).
425  *			MUST NOT change this header information.
426  *			MUST NOT reply to a multicast/broadcast IP address.
427  *			MUST NOT reply to a multicast/broadcast MAC address.
428  *			MUST reply to only the first fragment.
429  */
430 
431 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
432 {
433 	struct iphdr *iph;
434 	int room;
435 	struct icmp_bxm icmp_param;
436 	struct rtable *rt = (struct rtable *)skb_in->dst;
437 	struct ipcm_cookie ipc;
438 	__be32 saddr;
439 	u8  tos;
440 	struct net *net;
441 
442 	if (!rt)
443 		goto out;
444 	net = rt->u.dst.dev->nd_net;
445 
446 	/*
447 	 *	Find the original header. It is expected to be valid, of course.
448 	 *	Check this, icmp_send is called from the most obscure devices
449 	 *	sometimes.
450 	 */
451 	iph = ip_hdr(skb_in);
452 
453 	if ((u8 *)iph < skb_in->head ||
454 	    (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
455 		goto out;
456 
457 	/*
458 	 *	No replies to physical multicast/broadcast
459 	 */
460 	if (skb_in->pkt_type != PACKET_HOST)
461 		goto out;
462 
463 	/*
464 	 *	Now check at the protocol level
465 	 */
466 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
467 		goto out;
468 
469 	/*
470 	 *	Only reply to fragment 0. We byte re-order the constant
471 	 *	mask for efficiency.
472 	 */
473 	if (iph->frag_off & htons(IP_OFFSET))
474 		goto out;
475 
476 	/*
477 	 *	If we send an ICMP error to an ICMP error a mess would result..
478 	 */
479 	if (icmp_pointers[type].error) {
480 		/*
481 		 *	We are an error, check if we are replying to an
482 		 *	ICMP error
483 		 */
484 		if (iph->protocol == IPPROTO_ICMP) {
485 			u8 _inner_type, *itp;
486 
487 			itp = skb_header_pointer(skb_in,
488 						 skb_network_header(skb_in) +
489 						 (iph->ihl << 2) +
490 						 offsetof(struct icmphdr,
491 							  type) -
492 						 skb_in->data,
493 						 sizeof(_inner_type),
494 						 &_inner_type);
495 			if (itp == NULL)
496 				goto out;
497 
498 			/*
499 			 *	Assume any unknown ICMP type is an error. This
500 			 *	isn't specified by the RFC, but think about it..
501 			 */
502 			if (*itp > NR_ICMP_TYPES ||
503 			    icmp_pointers[*itp].error)
504 				goto out;
505 		}
506 	}
507 
508 	if (icmp_xmit_lock())
509 		return;
510 
511 	/*
512 	 *	Construct source address and options.
513 	 */
514 
515 	saddr = iph->daddr;
516 	if (!(rt->rt_flags & RTCF_LOCAL)) {
517 		struct net_device *dev = NULL;
518 
519 		if (rt->fl.iif && sysctl_icmp_errors_use_inbound_ifaddr)
520 			dev = dev_get_by_index(net, rt->fl.iif);
521 
522 		if (dev) {
523 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
524 			dev_put(dev);
525 		} else
526 			saddr = 0;
527 	}
528 
529 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
530 					   IPTOS_PREC_INTERNETCONTROL) :
531 					  iph->tos;
532 
533 	if (ip_options_echo(&icmp_param.replyopts, skb_in))
534 		goto out_unlock;
535 
536 
537 	/*
538 	 *	Prepare data for ICMP header.
539 	 */
540 
541 	icmp_param.data.icmph.type	 = type;
542 	icmp_param.data.icmph.code	 = code;
543 	icmp_param.data.icmph.un.gateway = info;
544 	icmp_param.data.icmph.checksum	 = 0;
545 	icmp_param.skb	  = skb_in;
546 	icmp_param.offset = skb_network_offset(skb_in);
547 	inet_sk(icmp_socket->sk)->tos = tos;
548 	ipc.addr = iph->saddr;
549 	ipc.opt = &icmp_param.replyopts;
550 
551 	{
552 		struct flowi fl = {
553 			.nl_u = {
554 				.ip4_u = {
555 					.daddr = icmp_param.replyopts.srr ?
556 						icmp_param.replyopts.faddr :
557 						iph->saddr,
558 					.saddr = saddr,
559 					.tos = RT_TOS(tos)
560 				}
561 			},
562 			.proto = IPPROTO_ICMP,
563 			.uli_u = {
564 				.icmpt = {
565 					.type = type,
566 					.code = code
567 				}
568 			}
569 		};
570 		int err;
571 		struct rtable *rt2;
572 
573 		security_skb_classify_flow(skb_in, &fl);
574 		if (__ip_route_output_key(net, &rt, &fl))
575 			goto out_unlock;
576 
577 		/* No need to clone since we're just using its address. */
578 		rt2 = rt;
579 
580 		err = xfrm_lookup((struct dst_entry **)&rt, &fl, NULL, 0);
581 		switch (err) {
582 		case 0:
583 			if (rt != rt2)
584 				goto route_done;
585 			break;
586 		case -EPERM:
587 			rt = NULL;
588 			break;
589 		default:
590 			goto out_unlock;
591 		}
592 
593 		if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
594 			goto out_unlock;
595 
596 		if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
597 			err = __ip_route_output_key(net, &rt2, &fl);
598 		else {
599 			struct flowi fl2 = {};
600 			struct dst_entry *odst;
601 
602 			fl2.fl4_dst = fl.fl4_src;
603 			if (ip_route_output_key(net, &rt2, &fl2))
604 				goto out_unlock;
605 
606 			/* Ugh! */
607 			odst = skb_in->dst;
608 			err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
609 					     RT_TOS(tos), rt2->u.dst.dev);
610 
611 			dst_release(&rt2->u.dst);
612 			rt2 = (struct rtable *)skb_in->dst;
613 			skb_in->dst = odst;
614 		}
615 
616 		if (err)
617 			goto out_unlock;
618 
619 		err = xfrm_lookup((struct dst_entry **)&rt2, &fl, NULL,
620 				  XFRM_LOOKUP_ICMP);
621 		if (err == -ENOENT) {
622 			if (!rt)
623 				goto out_unlock;
624 			goto route_done;
625 		}
626 
627 		dst_release(&rt->u.dst);
628 		rt = rt2;
629 
630 		if (err)
631 			goto out_unlock;
632 	}
633 
634 route_done:
635 	if (!icmpv4_xrlim_allow(rt, type, code))
636 		goto ende;
637 
638 	/* RFC says return as much as we can without exceeding 576 bytes. */
639 
640 	room = dst_mtu(&rt->u.dst);
641 	if (room > 576)
642 		room = 576;
643 	room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
644 	room -= sizeof(struct icmphdr);
645 
646 	icmp_param.data_len = skb_in->len - icmp_param.offset;
647 	if (icmp_param.data_len > room)
648 		icmp_param.data_len = room;
649 	icmp_param.head_len = sizeof(struct icmphdr);
650 
651 	icmp_push_reply(&icmp_param, &ipc, rt);
652 ende:
653 	ip_rt_put(rt);
654 out_unlock:
655 	icmp_xmit_unlock();
656 out:;
657 }
658 
659 
660 /*
661  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
662  */
663 
664 static void icmp_unreach(struct sk_buff *skb)
665 {
666 	struct iphdr *iph;
667 	struct icmphdr *icmph;
668 	int hash, protocol;
669 	struct net_protocol *ipprot;
670 	u32 info = 0;
671 	struct net *net;
672 
673 	net = skb->dst->dev->nd_net;
674 
675 	/*
676 	 *	Incomplete header ?
677 	 * 	Only checks for the IP header, there should be an
678 	 *	additional check for longer headers in upper levels.
679 	 */
680 
681 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
682 		goto out_err;
683 
684 	icmph = icmp_hdr(skb);
685 	iph   = (struct iphdr *)skb->data;
686 
687 	if (iph->ihl < 5) /* Mangled header, drop. */
688 		goto out_err;
689 
690 	if (icmph->type == ICMP_DEST_UNREACH) {
691 		switch (icmph->code & 15) {
692 		case ICMP_NET_UNREACH:
693 		case ICMP_HOST_UNREACH:
694 		case ICMP_PROT_UNREACH:
695 		case ICMP_PORT_UNREACH:
696 			break;
697 		case ICMP_FRAG_NEEDED:
698 			if (ipv4_config.no_pmtu_disc) {
699 				LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
700 							 "fragmentation needed "
701 							 "and DF set.\n",
702 					       NIPQUAD(iph->daddr));
703 			} else {
704 				info = ip_rt_frag_needed(net, iph,
705 						     ntohs(icmph->un.frag.mtu));
706 				if (!info)
707 					goto out;
708 			}
709 			break;
710 		case ICMP_SR_FAILED:
711 			LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
712 						 "Route Failed.\n",
713 				       NIPQUAD(iph->daddr));
714 			break;
715 		default:
716 			break;
717 		}
718 		if (icmph->code > NR_ICMP_UNREACH)
719 			goto out;
720 	} else if (icmph->type == ICMP_PARAMETERPROB)
721 		info = ntohl(icmph->un.gateway) >> 24;
722 
723 	/*
724 	 *	Throw it at our lower layers
725 	 *
726 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
727 	 *		  header.
728 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
729 	 *		  transport layer.
730 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
731 	 *		  transport layer.
732 	 */
733 
734 	/*
735 	 *	Check the other end isnt violating RFC 1122. Some routers send
736 	 *	bogus responses to broadcast frames. If you see this message
737 	 *	first check your netmask matches at both ends, if it does then
738 	 *	get the other vendor to fix their kit.
739 	 */
740 
741 	if (!sysctl_icmp_ignore_bogus_error_responses &&
742 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
743 		if (net_ratelimit())
744 			printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
745 					    "type %u, code %u "
746 					    "error to a broadcast: %u.%u.%u.%u on %s\n",
747 			       NIPQUAD(ip_hdr(skb)->saddr),
748 			       icmph->type, icmph->code,
749 			       NIPQUAD(iph->daddr),
750 			       skb->dev->name);
751 		goto out;
752 	}
753 
754 	/* Checkin full IP header plus 8 bytes of protocol to
755 	 * avoid additional coding at protocol handlers.
756 	 */
757 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
758 		goto out;
759 
760 	iph = (struct iphdr *)skb->data;
761 	protocol = iph->protocol;
762 
763 	/*
764 	 *	Deliver ICMP message to raw sockets. Pretty useless feature?
765 	 */
766 	raw_icmp_error(skb, protocol, info);
767 
768 	hash = protocol & (MAX_INET_PROTOS - 1);
769 	rcu_read_lock();
770 	ipprot = rcu_dereference(inet_protos[hash]);
771 	if (ipprot && ipprot->err_handler)
772 		ipprot->err_handler(skb, info);
773 	rcu_read_unlock();
774 
775 out:
776 	return;
777 out_err:
778 	ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
779 	goto out;
780 }
781 
782 
783 /*
784  *	Handle ICMP_REDIRECT.
785  */
786 
787 static void icmp_redirect(struct sk_buff *skb)
788 {
789 	struct iphdr *iph;
790 
791 	if (skb->len < sizeof(struct iphdr))
792 		goto out_err;
793 
794 	/*
795 	 *	Get the copied header of the packet that caused the redirect
796 	 */
797 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
798 		goto out;
799 
800 	iph = (struct iphdr *)skb->data;
801 
802 	switch (icmp_hdr(skb)->code & 7) {
803 	case ICMP_REDIR_NET:
804 	case ICMP_REDIR_NETTOS:
805 		/*
806 		 * As per RFC recommendations now handle it as a host redirect.
807 		 */
808 	case ICMP_REDIR_HOST:
809 	case ICMP_REDIR_HOSTTOS:
810 		ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
811 			       icmp_hdr(skb)->un.gateway,
812 			       iph->saddr, skb->dev);
813 		break;
814 	}
815 out:
816 	return;
817 out_err:
818 	ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
819 	goto out;
820 }
821 
822 /*
823  *	Handle ICMP_ECHO ("ping") requests.
824  *
825  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
826  *		  requests.
827  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
828  *		  included in the reply.
829  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
830  *		  echo requests, MUST have default=NOT.
831  *	See also WRT handling of options once they are done and working.
832  */
833 
834 static void icmp_echo(struct sk_buff *skb)
835 {
836 	if (!sysctl_icmp_echo_ignore_all) {
837 		struct icmp_bxm icmp_param;
838 
839 		icmp_param.data.icmph	   = *icmp_hdr(skb);
840 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
841 		icmp_param.skb		   = skb;
842 		icmp_param.offset	   = 0;
843 		icmp_param.data_len	   = skb->len;
844 		icmp_param.head_len	   = sizeof(struct icmphdr);
845 		icmp_reply(&icmp_param, skb);
846 	}
847 }
848 
849 /*
850  *	Handle ICMP Timestamp requests.
851  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
852  *		  SHOULD be in the kernel for minimum random latency.
853  *		  MUST be accurate to a few minutes.
854  *		  MUST be updated at least at 15Hz.
855  */
856 static void icmp_timestamp(struct sk_buff *skb)
857 {
858 	struct timeval tv;
859 	struct icmp_bxm icmp_param;
860 	/*
861 	 *	Too short.
862 	 */
863 	if (skb->len < 4)
864 		goto out_err;
865 
866 	/*
867 	 *	Fill in the current time as ms since midnight UT:
868 	 */
869 	do_gettimeofday(&tv);
870 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
871 					 tv.tv_usec / 1000);
872 	icmp_param.data.times[2] = icmp_param.data.times[1];
873 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
874 		BUG();
875 	icmp_param.data.icmph	   = *icmp_hdr(skb);
876 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
877 	icmp_param.data.icmph.code = 0;
878 	icmp_param.skb		   = skb;
879 	icmp_param.offset	   = 0;
880 	icmp_param.data_len	   = 0;
881 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
882 	icmp_reply(&icmp_param, skb);
883 out:
884 	return;
885 out_err:
886 	ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
887 	goto out;
888 }
889 
890 
891 /*
892  *	Handle ICMP_ADDRESS_MASK requests.  (RFC950)
893  *
894  * RFC1122 (3.2.2.9).  A host MUST only send replies to
895  * ADDRESS_MASK requests if it's been configured as an address mask
896  * agent.  Receiving a request doesn't constitute implicit permission to
897  * act as one. Of course, implementing this correctly requires (SHOULD)
898  * a way to turn the functionality on and off.  Another one for sysctl(),
899  * I guess. -- MS
900  *
901  * RFC1812 (4.3.3.9).	A router MUST implement it.
902  *			A router SHOULD have switch turning it on/off.
903  *		      	This switch MUST be ON by default.
904  *
905  * Gratuitous replies, zero-source replies are not implemented,
906  * that complies with RFC. DO NOT implement them!!! All the idea
907  * of broadcast addrmask replies as specified in RFC950 is broken.
908  * The problem is that it is not uncommon to have several prefixes
909  * on one physical interface. Moreover, addrmask agent can even be
910  * not aware of existing another prefixes.
911  * If source is zero, addrmask agent cannot choose correct prefix.
912  * Gratuitous mask announcements suffer from the same problem.
913  * RFC1812 explains it, but still allows to use ADDRMASK,
914  * that is pretty silly. --ANK
915  *
916  * All these rules are so bizarre, that I removed kernel addrmask
917  * support at all. It is wrong, it is obsolete, nobody uses it in
918  * any case. --ANK
919  *
920  * Furthermore you can do it with a usermode address agent program
921  * anyway...
922  */
923 
924 static void icmp_address(struct sk_buff *skb)
925 {
926 #if 0
927 	if (net_ratelimit())
928 		printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
929 #endif
930 }
931 
932 /*
933  * RFC1812 (4.3.3.9).	A router SHOULD listen all replies, and complain
934  *			loudly if an inconsistency is found.
935  */
936 
937 static void icmp_address_reply(struct sk_buff *skb)
938 {
939 	struct rtable *rt = (struct rtable *)skb->dst;
940 	struct net_device *dev = skb->dev;
941 	struct in_device *in_dev;
942 	struct in_ifaddr *ifa;
943 
944 	if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
945 		goto out;
946 
947 	in_dev = in_dev_get(dev);
948 	if (!in_dev)
949 		goto out;
950 	rcu_read_lock();
951 	if (in_dev->ifa_list &&
952 	    IN_DEV_LOG_MARTIANS(in_dev) &&
953 	    IN_DEV_FORWARD(in_dev)) {
954 		__be32 _mask, *mp;
955 
956 		mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
957 		BUG_ON(mp == NULL);
958 		for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
959 			if (*mp == ifa->ifa_mask &&
960 			    inet_ifa_match(rt->rt_src, ifa))
961 				break;
962 		}
963 		if (!ifa && net_ratelimit()) {
964 			printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
965 					 "%s/%u.%u.%u.%u\n",
966 			       NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
967 		}
968 	}
969 	rcu_read_unlock();
970 	in_dev_put(in_dev);
971 out:;
972 }
973 
974 static void icmp_discard(struct sk_buff *skb)
975 {
976 }
977 
978 /*
979  *	Deal with incoming ICMP packets.
980  */
981 int icmp_rcv(struct sk_buff *skb)
982 {
983 	struct icmphdr *icmph;
984 	struct rtable *rt = (struct rtable *)skb->dst;
985 
986 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
987 		int nh;
988 
989 		if (!(skb->sp && skb->sp->xvec[skb->sp->len - 1]->props.flags &
990 				 XFRM_STATE_ICMP))
991 			goto drop;
992 
993 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
994 			goto drop;
995 
996 		nh = skb_network_offset(skb);
997 		skb_set_network_header(skb, sizeof(*icmph));
998 
999 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1000 			goto drop;
1001 
1002 		skb_set_network_header(skb, nh);
1003 	}
1004 
1005 	ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
1006 
1007 	switch (skb->ip_summed) {
1008 	case CHECKSUM_COMPLETE:
1009 		if (!csum_fold(skb->csum))
1010 			break;
1011 		/* fall through */
1012 	case CHECKSUM_NONE:
1013 		skb->csum = 0;
1014 		if (__skb_checksum_complete(skb))
1015 			goto error;
1016 	}
1017 
1018 	if (!pskb_pull(skb, sizeof(*icmph)))
1019 		goto error;
1020 
1021 	icmph = icmp_hdr(skb);
1022 
1023 	ICMPMSGIN_INC_STATS_BH(icmph->type);
1024 	/*
1025 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1026 	 *
1027 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1028 	 *		  discarded.
1029 	 */
1030 	if (icmph->type > NR_ICMP_TYPES)
1031 		goto error;
1032 
1033 
1034 	/*
1035 	 *	Parse the ICMP message
1036 	 */
1037 
1038 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1039 		/*
1040 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1041 		 *	  silently ignored (we let user decide with a sysctl).
1042 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1043 		 *	  discarded if to broadcast/multicast.
1044 		 */
1045 		if ((icmph->type == ICMP_ECHO ||
1046 		     icmph->type == ICMP_TIMESTAMP) &&
1047 		    sysctl_icmp_echo_ignore_broadcasts) {
1048 			goto error;
1049 		}
1050 		if (icmph->type != ICMP_ECHO &&
1051 		    icmph->type != ICMP_TIMESTAMP &&
1052 		    icmph->type != ICMP_ADDRESS &&
1053 		    icmph->type != ICMP_ADDRESSREPLY) {
1054 			goto error;
1055 		}
1056 	}
1057 
1058 	icmp_pointers[icmph->type].handler(skb);
1059 
1060 drop:
1061 	kfree_skb(skb);
1062 	return 0;
1063 error:
1064 	ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
1065 	goto drop;
1066 }
1067 
1068 /*
1069  *	This table is the definition of how we handle ICMP.
1070  */
1071 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1072 	[ICMP_ECHOREPLY] = {
1073 		.handler = icmp_discard,
1074 	},
1075 	[1] = {
1076 		.handler = icmp_discard,
1077 		.error = 1,
1078 	},
1079 	[2] = {
1080 		.handler = icmp_discard,
1081 		.error = 1,
1082 	},
1083 	[ICMP_DEST_UNREACH] = {
1084 		.handler = icmp_unreach,
1085 		.error = 1,
1086 	},
1087 	[ICMP_SOURCE_QUENCH] = {
1088 		.handler = icmp_unreach,
1089 		.error = 1,
1090 	},
1091 	[ICMP_REDIRECT] = {
1092 		.handler = icmp_redirect,
1093 		.error = 1,
1094 	},
1095 	[6] = {
1096 		.handler = icmp_discard,
1097 		.error = 1,
1098 	},
1099 	[7] = {
1100 		.handler = icmp_discard,
1101 		.error = 1,
1102 	},
1103 	[ICMP_ECHO] = {
1104 		.handler = icmp_echo,
1105 	},
1106 	[9] = {
1107 		.handler = icmp_discard,
1108 		.error = 1,
1109 	},
1110 	[10] = {
1111 		.handler = icmp_discard,
1112 		.error = 1,
1113 	},
1114 	[ICMP_TIME_EXCEEDED] = {
1115 		.handler = icmp_unreach,
1116 		.error = 1,
1117 	},
1118 	[ICMP_PARAMETERPROB] = {
1119 		.handler = icmp_unreach,
1120 		.error = 1,
1121 	},
1122 	[ICMP_TIMESTAMP] = {
1123 		.handler = icmp_timestamp,
1124 	},
1125 	[ICMP_TIMESTAMPREPLY] = {
1126 		.handler = icmp_discard,
1127 	},
1128 	[ICMP_INFO_REQUEST] = {
1129 		.handler = icmp_discard,
1130 	},
1131 	[ICMP_INFO_REPLY] = {
1132 		.handler = icmp_discard,
1133 	},
1134 	[ICMP_ADDRESS] = {
1135 		.handler = icmp_address,
1136 	},
1137 	[ICMP_ADDRESSREPLY] = {
1138 		.handler = icmp_address_reply,
1139 	},
1140 };
1141 
1142 void __init icmp_init(struct net_proto_family *ops)
1143 {
1144 	struct inet_sock *inet;
1145 	int i;
1146 
1147 	for_each_possible_cpu(i) {
1148 		int err;
1149 
1150 		err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP,
1151 				       &per_cpu(__icmp_socket, i));
1152 
1153 		if (err < 0)
1154 			panic("Failed to create the ICMP control socket.\n");
1155 
1156 		per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC;
1157 
1158 		/* Enough space for 2 64K ICMP packets, including
1159 		 * sk_buff struct overhead.
1160 		 */
1161 		per_cpu(__icmp_socket, i)->sk->sk_sndbuf =
1162 			(2 * ((64 * 1024) + sizeof(struct sk_buff)));
1163 
1164 		inet = inet_sk(per_cpu(__icmp_socket, i)->sk);
1165 		inet->uc_ttl = -1;
1166 		inet->pmtudisc = IP_PMTUDISC_DONT;
1167 
1168 		/* Unhash it so that IP input processing does not even
1169 		 * see it, we do not wish this socket to see incoming
1170 		 * packets.
1171 		 */
1172 		per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk);
1173 	}
1174 }
1175 
1176 EXPORT_SYMBOL(icmp_err_convert);
1177 EXPORT_SYMBOL(icmp_send);
1178 EXPORT_SYMBOL(icmp_statistics);
1179 EXPORT_SYMBOL(xrlim_allow);
1180