xref: /linux/net/ipv4/icmp.c (revision 7f3edee81fbd49114c28057512906f169caa0bed)
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 
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 /*
115  *	Statistics
116  */
117 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly;
118 DEFINE_SNMP_STAT(struct icmpmsg_mib, icmpmsg_statistics) __read_mostly;
119 
120 /* An array of errno for error messages from dest unreach. */
121 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
122 
123 struct icmp_err icmp_err_convert[] = {
124 	{
125 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
126 		.fatal = 0,
127 	},
128 	{
129 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
130 		.fatal = 0,
131 	},
132 	{
133 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
134 		.fatal = 1,
135 	},
136 	{
137 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
138 		.fatal = 1,
139 	},
140 	{
141 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
142 		.fatal = 0,
143 	},
144 	{
145 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
146 		.fatal = 0,
147 	},
148 	{
149 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
150 		.fatal = 1,
151 	},
152 	{
153 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
154 		.fatal = 1,
155 	},
156 	{
157 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
158 		.fatal = 1,
159 	},
160 	{
161 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
162 		.fatal = 1,
163 	},
164 	{
165 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
166 		.fatal = 1,
167 	},
168 	{
169 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
170 		.fatal = 0,
171 	},
172 	{
173 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
174 		.fatal = 0,
175 	},
176 	{
177 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
178 		.fatal = 1,
179 	},
180 	{
181 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
182 		.fatal = 1,
183 	},
184 	{
185 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
186 		.fatal = 1,
187 	},
188 };
189 
190 /* Control parameters for ECHO replies. */
191 int sysctl_icmp_echo_ignore_all __read_mostly;
192 int sysctl_icmp_echo_ignore_broadcasts __read_mostly = 1;
193 
194 /* Control parameter - ignore bogus broadcast responses? */
195 int sysctl_icmp_ignore_bogus_error_responses __read_mostly = 1;
196 
197 /*
198  * 	Configurable global rate limit.
199  *
200  *	ratelimit defines tokens/packet consumed for dst->rate_token bucket
201  *	ratemask defines which icmp types are ratelimited by setting
202  * 	it's bit position.
203  *
204  *	default:
205  *	dest unreachable (3), source quench (4),
206  *	time exceeded (11), parameter problem (12)
207  */
208 
209 int sysctl_icmp_ratelimit __read_mostly = 1 * HZ;
210 int sysctl_icmp_ratemask __read_mostly = 0x1818;
211 int sysctl_icmp_errors_use_inbound_ifaddr __read_mostly;
212 
213 /*
214  *	ICMP control array. This specifies what to do with each ICMP.
215  */
216 
217 struct icmp_control {
218 	void (*handler)(struct sk_buff *skb);
219 	short   error;		/* This ICMP is classed as an error message */
220 };
221 
222 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
223 
224 /*
225  *	The ICMP socket(s). This is the most convenient way to flow control
226  *	our ICMP output as well as maintain a clean interface throughout
227  *	all layers. All Socketless IP sends will soon be gone.
228  *
229  *	On SMP we have one ICMP socket per-cpu.
230  */
231 static DEFINE_PER_CPU(struct socket *, __icmp_socket) = NULL;
232 #define icmp_socket	__get_cpu_var(__icmp_socket)
233 
234 static __inline__ int icmp_xmit_lock(void)
235 {
236 	local_bh_disable();
237 
238 	if (unlikely(!spin_trylock(&icmp_socket->sk->sk_lock.slock))) {
239 		/* This can happen if the output path signals a
240 		 * dst_link_failure() for an outgoing ICMP packet.
241 		 */
242 		local_bh_enable();
243 		return 1;
244 	}
245 	return 0;
246 }
247 
248 static void icmp_xmit_unlock(void)
249 {
250 	spin_unlock_bh(&icmp_socket->sk->sk_lock.slock);
251 }
252 
253 /*
254  *	Send an ICMP frame.
255  */
256 
257 /*
258  *	Check transmit rate limitation for given message.
259  *	The rate information is held in the destination cache now.
260  *	This function is generic and could be used for other purposes
261  *	too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
262  *
263  *	Note that the same dst_entry fields are modified by functions in
264  *	route.c too, but these work for packet destinations while xrlim_allow
265  *	works for icmp destinations. This means the rate limiting information
266  *	for one "ip object" is shared - and these ICMPs are twice limited:
267  *	by source and by destination.
268  *
269  *	RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
270  *			  SHOULD allow setting of rate limits
271  *
272  * 	Shared between ICMPv4 and ICMPv6.
273  */
274 #define XRLIM_BURST_FACTOR 6
275 int xrlim_allow(struct dst_entry *dst, int timeout)
276 {
277 	unsigned long now;
278 	int rc = 0;
279 
280 	now = jiffies;
281 	dst->rate_tokens += now - dst->rate_last;
282 	dst->rate_last = now;
283 	if (dst->rate_tokens > XRLIM_BURST_FACTOR * timeout)
284 		dst->rate_tokens = XRLIM_BURST_FACTOR * timeout;
285 	if (dst->rate_tokens >= timeout) {
286 		dst->rate_tokens -= timeout;
287 		rc = 1;
288 	}
289 	return rc;
290 }
291 
292 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
293 {
294 	struct dst_entry *dst = &rt->u.dst;
295 	int rc = 1;
296 
297 	if (type > NR_ICMP_TYPES)
298 		goto out;
299 
300 	/* Don't limit PMTU discovery. */
301 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
302 		goto out;
303 
304 	/* No rate limit on loopback */
305 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
306 		goto out;
307 
308 	/* Limit if icmp type is enabled in ratemask. */
309 	if ((1 << type) & sysctl_icmp_ratemask)
310 		rc = xrlim_allow(dst, sysctl_icmp_ratelimit);
311 out:
312 	return rc;
313 }
314 
315 /*
316  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
317  */
318 void icmp_out_count(unsigned char type)
319 {
320 	ICMPMSGOUT_INC_STATS(type);
321 	ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
322 }
323 
324 /*
325  *	Checksum each fragment, and on the first include the headers and final
326  *	checksum.
327  */
328 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
329 			  struct sk_buff *skb)
330 {
331 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
332 	__wsum csum;
333 
334 	csum = skb_copy_and_csum_bits(icmp_param->skb,
335 				      icmp_param->offset + offset,
336 				      to, len, 0);
337 
338 	skb->csum = csum_block_add(skb->csum, csum, odd);
339 	if (icmp_pointers[icmp_param->data.icmph.type].error)
340 		nf_ct_attach(skb, icmp_param->skb);
341 	return 0;
342 }
343 
344 static void icmp_push_reply(struct icmp_bxm *icmp_param,
345 			    struct ipcm_cookie *ipc, struct rtable *rt)
346 {
347 	struct sk_buff *skb;
348 
349 	if (ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param,
350 			   icmp_param->data_len+icmp_param->head_len,
351 			   icmp_param->head_len,
352 			   ipc, rt, MSG_DONTWAIT) < 0)
353 		ip_flush_pending_frames(icmp_socket->sk);
354 	else if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) {
355 		struct icmphdr *icmph = icmp_hdr(skb);
356 		__wsum csum = 0;
357 		struct sk_buff *skb1;
358 
359 		skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) {
360 			csum = csum_add(csum, skb1->csum);
361 		}
362 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
363 						 (char *)icmph,
364 						 icmp_param->head_len, csum);
365 		icmph->checksum = csum_fold(csum);
366 		skb->ip_summed = CHECKSUM_NONE;
367 		ip_push_pending_frames(icmp_socket->sk);
368 	}
369 }
370 
371 /*
372  *	Driving logic for building and sending ICMP messages.
373  */
374 
375 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
376 {
377 	struct sock *sk = icmp_socket->sk;
378 	struct inet_sock *inet = inet_sk(sk);
379 	struct ipcm_cookie ipc;
380 	struct rtable *rt = (struct rtable *)skb->dst;
381 	__be32 daddr;
382 
383 	if (ip_options_echo(&icmp_param->replyopts, skb))
384 		return;
385 
386 	if (icmp_xmit_lock())
387 		return;
388 
389 	icmp_param->data.icmph.checksum = 0;
390 
391 	inet->tos = ip_hdr(skb)->tos;
392 	daddr = ipc.addr = rt->rt_src;
393 	ipc.opt = NULL;
394 	if (icmp_param->replyopts.optlen) {
395 		ipc.opt = &icmp_param->replyopts;
396 		if (ipc.opt->srr)
397 			daddr = icmp_param->replyopts.faddr;
398 	}
399 	{
400 		struct flowi fl = { .nl_u = { .ip4_u =
401 					      { .daddr = daddr,
402 						.saddr = rt->rt_spec_dst,
403 						.tos = RT_TOS(ip_hdr(skb)->tos) } },
404 				    .proto = IPPROTO_ICMP };
405 		security_skb_classify_flow(skb, &fl);
406 		if (ip_route_output_key(&rt, &fl))
407 			goto out_unlock;
408 	}
409 	if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
410 			       icmp_param->data.icmph.code))
411 		icmp_push_reply(icmp_param, &ipc, rt);
412 	ip_rt_put(rt);
413 out_unlock:
414 	icmp_xmit_unlock();
415 }
416 
417 
418 /*
419  *	Send an ICMP message in response to a situation
420  *
421  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
422  *		  MAY send more (we do).
423  *			MUST NOT change this header information.
424  *			MUST NOT reply to a multicast/broadcast IP address.
425  *			MUST NOT reply to a multicast/broadcast MAC address.
426  *			MUST reply to only the first fragment.
427  */
428 
429 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
430 {
431 	struct iphdr *iph;
432 	int room;
433 	struct icmp_bxm icmp_param;
434 	struct rtable *rt = (struct rtable *)skb_in->dst;
435 	struct ipcm_cookie ipc;
436 	__be32 saddr;
437 	u8  tos;
438 
439 	if (!rt)
440 		goto out;
441 
442 	/*
443 	 *	Find the original header. It is expected to be valid, of course.
444 	 *	Check this, icmp_send is called from the most obscure devices
445 	 *	sometimes.
446 	 */
447 	iph = ip_hdr(skb_in);
448 
449 	if ((u8 *)iph < skb_in->head ||
450 	    (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
451 		goto out;
452 
453 	/*
454 	 *	No replies to physical multicast/broadcast
455 	 */
456 	if (skb_in->pkt_type != PACKET_HOST)
457 		goto out;
458 
459 	/*
460 	 *	Now check at the protocol level
461 	 */
462 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
463 		goto out;
464 
465 	/*
466 	 *	Only reply to fragment 0. We byte re-order the constant
467 	 *	mask for efficiency.
468 	 */
469 	if (iph->frag_off & htons(IP_OFFSET))
470 		goto out;
471 
472 	/*
473 	 *	If we send an ICMP error to an ICMP error a mess would result..
474 	 */
475 	if (icmp_pointers[type].error) {
476 		/*
477 		 *	We are an error, check if we are replying to an
478 		 *	ICMP error
479 		 */
480 		if (iph->protocol == IPPROTO_ICMP) {
481 			u8 _inner_type, *itp;
482 
483 			itp = skb_header_pointer(skb_in,
484 						 skb_network_header(skb_in) +
485 						 (iph->ihl << 2) +
486 						 offsetof(struct icmphdr,
487 							  type) -
488 						 skb_in->data,
489 						 sizeof(_inner_type),
490 						 &_inner_type);
491 			if (itp == NULL)
492 				goto out;
493 
494 			/*
495 			 *	Assume any unknown ICMP type is an error. This
496 			 *	isn't specified by the RFC, but think about it..
497 			 */
498 			if (*itp > NR_ICMP_TYPES ||
499 			    icmp_pointers[*itp].error)
500 				goto out;
501 		}
502 	}
503 
504 	if (icmp_xmit_lock())
505 		return;
506 
507 	/*
508 	 *	Construct source address and options.
509 	 */
510 
511 	saddr = iph->daddr;
512 	if (!(rt->rt_flags & RTCF_LOCAL)) {
513 		struct net_device *dev = NULL;
514 
515 		if (rt->fl.iif && sysctl_icmp_errors_use_inbound_ifaddr)
516 			dev = dev_get_by_index(&init_net, rt->fl.iif);
517 
518 		if (dev) {
519 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
520 			dev_put(dev);
521 		} else
522 			saddr = 0;
523 	}
524 
525 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
526 					   IPTOS_PREC_INTERNETCONTROL) :
527 					  iph->tos;
528 
529 	if (ip_options_echo(&icmp_param.replyopts, skb_in))
530 		goto out_unlock;
531 
532 
533 	/*
534 	 *	Prepare data for ICMP header.
535 	 */
536 
537 	icmp_param.data.icmph.type	 = type;
538 	icmp_param.data.icmph.code	 = code;
539 	icmp_param.data.icmph.un.gateway = info;
540 	icmp_param.data.icmph.checksum	 = 0;
541 	icmp_param.skb	  = skb_in;
542 	icmp_param.offset = skb_network_offset(skb_in);
543 	inet_sk(icmp_socket->sk)->tos = tos;
544 	ipc.addr = iph->saddr;
545 	ipc.opt = &icmp_param.replyopts;
546 
547 	{
548 		struct flowi fl = {
549 			.nl_u = {
550 				.ip4_u = {
551 					.daddr = icmp_param.replyopts.srr ?
552 						icmp_param.replyopts.faddr :
553 						iph->saddr,
554 					.saddr = saddr,
555 					.tos = RT_TOS(tos)
556 				}
557 			},
558 			.proto = IPPROTO_ICMP,
559 			.uli_u = {
560 				.icmpt = {
561 					.type = type,
562 					.code = code
563 				}
564 			}
565 		};
566 		security_skb_classify_flow(skb_in, &fl);
567 		if (ip_route_output_key(&rt, &fl))
568 			goto out_unlock;
569 	}
570 
571 	if (!icmpv4_xrlim_allow(rt, type, code))
572 		goto ende;
573 
574 	/* RFC says return as much as we can without exceeding 576 bytes. */
575 
576 	room = dst_mtu(&rt->u.dst);
577 	if (room > 576)
578 		room = 576;
579 	room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
580 	room -= sizeof(struct icmphdr);
581 
582 	icmp_param.data_len = skb_in->len - icmp_param.offset;
583 	if (icmp_param.data_len > room)
584 		icmp_param.data_len = room;
585 	icmp_param.head_len = sizeof(struct icmphdr);
586 
587 	icmp_push_reply(&icmp_param, &ipc, rt);
588 ende:
589 	ip_rt_put(rt);
590 out_unlock:
591 	icmp_xmit_unlock();
592 out:;
593 }
594 
595 
596 /*
597  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
598  */
599 
600 static void icmp_unreach(struct sk_buff *skb)
601 {
602 	struct iphdr *iph;
603 	struct icmphdr *icmph;
604 	int hash, protocol;
605 	struct net_protocol *ipprot;
606 	struct sock *raw_sk;
607 	u32 info = 0;
608 
609 	/*
610 	 *	Incomplete header ?
611 	 * 	Only checks for the IP header, there should be an
612 	 *	additional check for longer headers in upper levels.
613 	 */
614 
615 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
616 		goto out_err;
617 
618 	icmph = icmp_hdr(skb);
619 	iph   = (struct iphdr *)skb->data;
620 
621 	if (iph->ihl < 5) /* Mangled header, drop. */
622 		goto out_err;
623 
624 	if (icmph->type == ICMP_DEST_UNREACH) {
625 		switch (icmph->code & 15) {
626 		case ICMP_NET_UNREACH:
627 		case ICMP_HOST_UNREACH:
628 		case ICMP_PROT_UNREACH:
629 		case ICMP_PORT_UNREACH:
630 			break;
631 		case ICMP_FRAG_NEEDED:
632 			if (ipv4_config.no_pmtu_disc) {
633 				LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
634 							 "fragmentation needed "
635 							 "and DF set.\n",
636 					       NIPQUAD(iph->daddr));
637 			} else {
638 				info = ip_rt_frag_needed(iph,
639 						     ntohs(icmph->un.frag.mtu));
640 				if (!info)
641 					goto out;
642 			}
643 			break;
644 		case ICMP_SR_FAILED:
645 			LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
646 						 "Route Failed.\n",
647 				       NIPQUAD(iph->daddr));
648 			break;
649 		default:
650 			break;
651 		}
652 		if (icmph->code > NR_ICMP_UNREACH)
653 			goto out;
654 	} else if (icmph->type == ICMP_PARAMETERPROB)
655 		info = ntohl(icmph->un.gateway) >> 24;
656 
657 	/*
658 	 *	Throw it at our lower layers
659 	 *
660 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
661 	 *		  header.
662 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
663 	 *		  transport layer.
664 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
665 	 *		  transport layer.
666 	 */
667 
668 	/*
669 	 *	Check the other end isnt violating RFC 1122. Some routers send
670 	 *	bogus responses to broadcast frames. If you see this message
671 	 *	first check your netmask matches at both ends, if it does then
672 	 *	get the other vendor to fix their kit.
673 	 */
674 
675 	if (!sysctl_icmp_ignore_bogus_error_responses &&
676 	    inet_addr_type(iph->daddr) == RTN_BROADCAST) {
677 		if (net_ratelimit())
678 			printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
679 					    "type %u, code %u "
680 					    "error to a broadcast: %u.%u.%u.%u on %s\n",
681 			       NIPQUAD(ip_hdr(skb)->saddr),
682 			       icmph->type, icmph->code,
683 			       NIPQUAD(iph->daddr),
684 			       skb->dev->name);
685 		goto out;
686 	}
687 
688 	/* Checkin full IP header plus 8 bytes of protocol to
689 	 * avoid additional coding at protocol handlers.
690 	 */
691 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
692 		goto out;
693 
694 	iph = (struct iphdr *)skb->data;
695 	protocol = iph->protocol;
696 
697 	/*
698 	 *	Deliver ICMP message to raw sockets. Pretty useless feature?
699 	 */
700 
701 	/* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */
702 	hash = protocol & (MAX_INET_PROTOS - 1);
703 	read_lock(&raw_v4_lock);
704 	if ((raw_sk = sk_head(&raw_v4_htable[hash])) != NULL) {
705 		while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr,
706 						 iph->saddr,
707 						 skb->dev->ifindex)) != NULL) {
708 			raw_err(raw_sk, skb, info);
709 			raw_sk = sk_next(raw_sk);
710 			iph = (struct iphdr *)skb->data;
711 		}
712 	}
713 	read_unlock(&raw_v4_lock);
714 
715 	rcu_read_lock();
716 	ipprot = rcu_dereference(inet_protos[hash]);
717 	if (ipprot && ipprot->err_handler)
718 		ipprot->err_handler(skb, info);
719 	rcu_read_unlock();
720 
721 out:
722 	return;
723 out_err:
724 	ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
725 	goto out;
726 }
727 
728 
729 /*
730  *	Handle ICMP_REDIRECT.
731  */
732 
733 static void icmp_redirect(struct sk_buff *skb)
734 {
735 	struct iphdr *iph;
736 
737 	if (skb->len < sizeof(struct iphdr))
738 		goto out_err;
739 
740 	/*
741 	 *	Get the copied header of the packet that caused the redirect
742 	 */
743 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
744 		goto out;
745 
746 	iph = (struct iphdr *)skb->data;
747 
748 	switch (icmp_hdr(skb)->code & 7) {
749 	case ICMP_REDIR_NET:
750 	case ICMP_REDIR_NETTOS:
751 		/*
752 		 * As per RFC recommendations now handle it as a host redirect.
753 		 */
754 	case ICMP_REDIR_HOST:
755 	case ICMP_REDIR_HOSTTOS:
756 		ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
757 			       icmp_hdr(skb)->un.gateway,
758 			       iph->saddr, skb->dev);
759 		break;
760 	}
761 out:
762 	return;
763 out_err:
764 	ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
765 	goto out;
766 }
767 
768 /*
769  *	Handle ICMP_ECHO ("ping") requests.
770  *
771  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
772  *		  requests.
773  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
774  *		  included in the reply.
775  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
776  *		  echo requests, MUST have default=NOT.
777  *	See also WRT handling of options once they are done and working.
778  */
779 
780 static void icmp_echo(struct sk_buff *skb)
781 {
782 	if (!sysctl_icmp_echo_ignore_all) {
783 		struct icmp_bxm icmp_param;
784 
785 		icmp_param.data.icmph	   = *icmp_hdr(skb);
786 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
787 		icmp_param.skb		   = skb;
788 		icmp_param.offset	   = 0;
789 		icmp_param.data_len	   = skb->len;
790 		icmp_param.head_len	   = sizeof(struct icmphdr);
791 		icmp_reply(&icmp_param, skb);
792 	}
793 }
794 
795 /*
796  *	Handle ICMP Timestamp requests.
797  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
798  *		  SHOULD be in the kernel for minimum random latency.
799  *		  MUST be accurate to a few minutes.
800  *		  MUST be updated at least at 15Hz.
801  */
802 static void icmp_timestamp(struct sk_buff *skb)
803 {
804 	struct timeval tv;
805 	struct icmp_bxm icmp_param;
806 	/*
807 	 *	Too short.
808 	 */
809 	if (skb->len < 4)
810 		goto out_err;
811 
812 	/*
813 	 *	Fill in the current time as ms since midnight UT:
814 	 */
815 	do_gettimeofday(&tv);
816 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
817 					 tv.tv_usec / 1000);
818 	icmp_param.data.times[2] = icmp_param.data.times[1];
819 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
820 		BUG();
821 	icmp_param.data.icmph	   = *icmp_hdr(skb);
822 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
823 	icmp_param.data.icmph.code = 0;
824 	icmp_param.skb		   = skb;
825 	icmp_param.offset	   = 0;
826 	icmp_param.data_len	   = 0;
827 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
828 	icmp_reply(&icmp_param, skb);
829 out:
830 	return;
831 out_err:
832 	ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
833 	goto out;
834 }
835 
836 
837 /*
838  *	Handle ICMP_ADDRESS_MASK requests.  (RFC950)
839  *
840  * RFC1122 (3.2.2.9).  A host MUST only send replies to
841  * ADDRESS_MASK requests if it's been configured as an address mask
842  * agent.  Receiving a request doesn't constitute implicit permission to
843  * act as one. Of course, implementing this correctly requires (SHOULD)
844  * a way to turn the functionality on and off.  Another one for sysctl(),
845  * I guess. -- MS
846  *
847  * RFC1812 (4.3.3.9).	A router MUST implement it.
848  *			A router SHOULD have switch turning it on/off.
849  *		      	This switch MUST be ON by default.
850  *
851  * Gratuitous replies, zero-source replies are not implemented,
852  * that complies with RFC. DO NOT implement them!!! All the idea
853  * of broadcast addrmask replies as specified in RFC950 is broken.
854  * The problem is that it is not uncommon to have several prefixes
855  * on one physical interface. Moreover, addrmask agent can even be
856  * not aware of existing another prefixes.
857  * If source is zero, addrmask agent cannot choose correct prefix.
858  * Gratuitous mask announcements suffer from the same problem.
859  * RFC1812 explains it, but still allows to use ADDRMASK,
860  * that is pretty silly. --ANK
861  *
862  * All these rules are so bizarre, that I removed kernel addrmask
863  * support at all. It is wrong, it is obsolete, nobody uses it in
864  * any case. --ANK
865  *
866  * Furthermore you can do it with a usermode address agent program
867  * anyway...
868  */
869 
870 static void icmp_address(struct sk_buff *skb)
871 {
872 #if 0
873 	if (net_ratelimit())
874 		printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
875 #endif
876 }
877 
878 /*
879  * RFC1812 (4.3.3.9).	A router SHOULD listen all replies, and complain
880  *			loudly if an inconsistency is found.
881  */
882 
883 static void icmp_address_reply(struct sk_buff *skb)
884 {
885 	struct rtable *rt = (struct rtable *)skb->dst;
886 	struct net_device *dev = skb->dev;
887 	struct in_device *in_dev;
888 	struct in_ifaddr *ifa;
889 
890 	if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
891 		goto out;
892 
893 	in_dev = in_dev_get(dev);
894 	if (!in_dev)
895 		goto out;
896 	rcu_read_lock();
897 	if (in_dev->ifa_list &&
898 	    IN_DEV_LOG_MARTIANS(in_dev) &&
899 	    IN_DEV_FORWARD(in_dev)) {
900 		__be32 _mask, *mp;
901 
902 		mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
903 		BUG_ON(mp == NULL);
904 		for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
905 			if (*mp == ifa->ifa_mask &&
906 			    inet_ifa_match(rt->rt_src, ifa))
907 				break;
908 		}
909 		if (!ifa && net_ratelimit()) {
910 			printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
911 					 "%s/%u.%u.%u.%u\n",
912 			       NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
913 		}
914 	}
915 	rcu_read_unlock();
916 	in_dev_put(in_dev);
917 out:;
918 }
919 
920 static void icmp_discard(struct sk_buff *skb)
921 {
922 }
923 
924 /*
925  *	Deal with incoming ICMP packets.
926  */
927 int icmp_rcv(struct sk_buff *skb)
928 {
929 	struct icmphdr *icmph;
930 	struct rtable *rt = (struct rtable *)skb->dst;
931 
932 	ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
933 
934 	switch (skb->ip_summed) {
935 	case CHECKSUM_COMPLETE:
936 		if (!csum_fold(skb->csum))
937 			break;
938 		/* fall through */
939 	case CHECKSUM_NONE:
940 		skb->csum = 0;
941 		if (__skb_checksum_complete(skb))
942 			goto error;
943 	}
944 
945 	if (!pskb_pull(skb, sizeof(struct icmphdr)))
946 		goto error;
947 
948 	icmph = icmp_hdr(skb);
949 
950 	ICMPMSGIN_INC_STATS_BH(icmph->type);
951 	/*
952 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
953 	 *
954 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
955 	 *		  discarded.
956 	 */
957 	if (icmph->type > NR_ICMP_TYPES)
958 		goto error;
959 
960 
961 	/*
962 	 *	Parse the ICMP message
963 	 */
964 
965 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
966 		/*
967 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
968 		 *	  silently ignored (we let user decide with a sysctl).
969 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
970 		 *	  discarded if to broadcast/multicast.
971 		 */
972 		if ((icmph->type == ICMP_ECHO ||
973 		     icmph->type == ICMP_TIMESTAMP) &&
974 		    sysctl_icmp_echo_ignore_broadcasts) {
975 			goto error;
976 		}
977 		if (icmph->type != ICMP_ECHO &&
978 		    icmph->type != ICMP_TIMESTAMP &&
979 		    icmph->type != ICMP_ADDRESS &&
980 		    icmph->type != ICMP_ADDRESSREPLY) {
981 			goto error;
982 		}
983 	}
984 
985 	icmp_pointers[icmph->type].handler(skb);
986 
987 drop:
988 	kfree_skb(skb);
989 	return 0;
990 error:
991 	ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
992 	goto drop;
993 }
994 
995 /*
996  *	This table is the definition of how we handle ICMP.
997  */
998 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
999 	[ICMP_ECHOREPLY] = {
1000 		.handler = icmp_discard,
1001 	},
1002 	[1] = {
1003 		.handler = icmp_discard,
1004 		.error = 1,
1005 	},
1006 	[2] = {
1007 		.handler = icmp_discard,
1008 		.error = 1,
1009 	},
1010 	[ICMP_DEST_UNREACH] = {
1011 		.handler = icmp_unreach,
1012 		.error = 1,
1013 	},
1014 	[ICMP_SOURCE_QUENCH] = {
1015 		.handler = icmp_unreach,
1016 		.error = 1,
1017 	},
1018 	[ICMP_REDIRECT] = {
1019 		.handler = icmp_redirect,
1020 		.error = 1,
1021 	},
1022 	[6] = {
1023 		.handler = icmp_discard,
1024 		.error = 1,
1025 	},
1026 	[7] = {
1027 		.handler = icmp_discard,
1028 		.error = 1,
1029 	},
1030 	[ICMP_ECHO] = {
1031 		.handler = icmp_echo,
1032 	},
1033 	[9] = {
1034 		.handler = icmp_discard,
1035 		.error = 1,
1036 	},
1037 	[10] = {
1038 		.handler = icmp_discard,
1039 		.error = 1,
1040 	},
1041 	[ICMP_TIME_EXCEEDED] = {
1042 		.handler = icmp_unreach,
1043 		.error = 1,
1044 	},
1045 	[ICMP_PARAMETERPROB] = {
1046 		.handler = icmp_unreach,
1047 		.error = 1,
1048 	},
1049 	[ICMP_TIMESTAMP] = {
1050 		.handler = icmp_timestamp,
1051 	},
1052 	[ICMP_TIMESTAMPREPLY] = {
1053 		.handler = icmp_discard,
1054 	},
1055 	[ICMP_INFO_REQUEST] = {
1056 		.handler = icmp_discard,
1057 	},
1058 	[ICMP_INFO_REPLY] = {
1059 		.handler = icmp_discard,
1060 	},
1061 	[ICMP_ADDRESS] = {
1062 		.handler = icmp_address,
1063 	},
1064 	[ICMP_ADDRESSREPLY] = {
1065 		.handler = icmp_address_reply,
1066 	},
1067 };
1068 
1069 void __init icmp_init(struct net_proto_family *ops)
1070 {
1071 	struct inet_sock *inet;
1072 	int i;
1073 
1074 	for_each_possible_cpu(i) {
1075 		int err;
1076 
1077 		err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP,
1078 				       &per_cpu(__icmp_socket, i));
1079 
1080 		if (err < 0)
1081 			panic("Failed to create the ICMP control socket.\n");
1082 
1083 		per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC;
1084 
1085 		/* Enough space for 2 64K ICMP packets, including
1086 		 * sk_buff struct overhead.
1087 		 */
1088 		per_cpu(__icmp_socket, i)->sk->sk_sndbuf =
1089 			(2 * ((64 * 1024) + sizeof(struct sk_buff)));
1090 
1091 		inet = inet_sk(per_cpu(__icmp_socket, i)->sk);
1092 		inet->uc_ttl = -1;
1093 		inet->pmtudisc = IP_PMTUDISC_DONT;
1094 
1095 		/* Unhash it so that IP input processing does not even
1096 		 * see it, we do not wish this socket to see incoming
1097 		 * packets.
1098 		 */
1099 		per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk);
1100 	}
1101 }
1102 
1103 EXPORT_SYMBOL(icmp_err_convert);
1104 EXPORT_SYMBOL(icmp_send);
1105 EXPORT_SYMBOL(icmp_statistics);
1106 EXPORT_SYMBOL(xrlim_allow);
1107