xref: /linux/net/ipv4/icmp.c (revision ec2212088c42ff7d1362629ec26dda4f3e8bdad3)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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 <linux/slab.h>
80 #include <net/snmp.h>
81 #include <net/ip.h>
82 #include <net/route.h>
83 #include <net/protocol.h>
84 #include <net/icmp.h>
85 #include <net/tcp.h>
86 #include <net/udp.h>
87 #include <net/raw.h>
88 #include <net/ping.h>
89 #include <linux/skbuff.h>
90 #include <net/sock.h>
91 #include <linux/errno.h>
92 #include <linux/timer.h>
93 #include <linux/init.h>
94 #include <asm/system.h>
95 #include <asm/uaccess.h>
96 #include <net/checksum.h>
97 #include <net/xfrm.h>
98 #include <net/inet_common.h>
99 
100 /*
101  *	Build xmit assembly blocks
102  */
103 
104 struct icmp_bxm {
105 	struct sk_buff *skb;
106 	int offset;
107 	int data_len;
108 
109 	struct {
110 		struct icmphdr icmph;
111 		__be32	       times[3];
112 	} data;
113 	int head_len;
114 	struct ip_options_data replyopts;
115 };
116 
117 /* An array of errno for error messages from dest unreach. */
118 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
119 
120 const struct icmp_err icmp_err_convert[] = {
121 	{
122 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
123 		.fatal = 0,
124 	},
125 	{
126 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
127 		.fatal = 0,
128 	},
129 	{
130 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
131 		.fatal = 1,
132 	},
133 	{
134 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
135 		.fatal = 1,
136 	},
137 	{
138 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
139 		.fatal = 0,
140 	},
141 	{
142 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
143 		.fatal = 0,
144 	},
145 	{
146 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
147 		.fatal = 1,
148 	},
149 	{
150 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
151 		.fatal = 1,
152 	},
153 	{
154 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
155 		.fatal = 1,
156 	},
157 	{
158 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
159 		.fatal = 1,
160 	},
161 	{
162 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
163 		.fatal = 1,
164 	},
165 	{
166 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
167 		.fatal = 0,
168 	},
169 	{
170 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
171 		.fatal = 0,
172 	},
173 	{
174 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
175 		.fatal = 1,
176 	},
177 	{
178 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
179 		.fatal = 1,
180 	},
181 	{
182 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
183 		.fatal = 1,
184 	},
185 };
186 EXPORT_SYMBOL(icmp_err_convert);
187 
188 /*
189  *	ICMP control array. This specifies what to do with each ICMP.
190  */
191 
192 struct icmp_control {
193 	void (*handler)(struct sk_buff *skb);
194 	short   error;		/* This ICMP is classed as an error message */
195 };
196 
197 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
198 
199 /*
200  *	The ICMP socket(s). This is the most convenient way to flow control
201  *	our ICMP output as well as maintain a clean interface throughout
202  *	all layers. All Socketless IP sends will soon be gone.
203  *
204  *	On SMP we have one ICMP socket per-cpu.
205  */
206 static struct sock *icmp_sk(struct net *net)
207 {
208 	return net->ipv4.icmp_sk[smp_processor_id()];
209 }
210 
211 static inline struct sock *icmp_xmit_lock(struct net *net)
212 {
213 	struct sock *sk;
214 
215 	local_bh_disable();
216 
217 	sk = icmp_sk(net);
218 
219 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
220 		/* This can happen if the output path signals a
221 		 * dst_link_failure() for an outgoing ICMP packet.
222 		 */
223 		local_bh_enable();
224 		return NULL;
225 	}
226 	return sk;
227 }
228 
229 static inline void icmp_xmit_unlock(struct sock *sk)
230 {
231 	spin_unlock_bh(&sk->sk_lock.slock);
232 }
233 
234 /*
235  *	Send an ICMP frame.
236  */
237 
238 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
239 				      struct flowi4 *fl4, int type, int code)
240 {
241 	struct dst_entry *dst = &rt->dst;
242 	bool rc = true;
243 
244 	if (type > NR_ICMP_TYPES)
245 		goto out;
246 
247 	/* Don't limit PMTU discovery. */
248 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
249 		goto out;
250 
251 	/* No rate limit on loopback */
252 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
253 		goto out;
254 
255 	/* Limit if icmp type is enabled in ratemask. */
256 	if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
257 		if (!rt->peer)
258 			rt_bind_peer(rt, fl4->daddr, 1);
259 		rc = inet_peer_xrlim_allow(rt->peer,
260 					   net->ipv4.sysctl_icmp_ratelimit);
261 	}
262 out:
263 	return rc;
264 }
265 
266 /*
267  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
268  */
269 void icmp_out_count(struct net *net, unsigned char type)
270 {
271 	ICMPMSGOUT_INC_STATS(net, type);
272 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
273 }
274 
275 /*
276  *	Checksum each fragment, and on the first include the headers and final
277  *	checksum.
278  */
279 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
280 			  struct sk_buff *skb)
281 {
282 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
283 	__wsum csum;
284 
285 	csum = skb_copy_and_csum_bits(icmp_param->skb,
286 				      icmp_param->offset + offset,
287 				      to, len, 0);
288 
289 	skb->csum = csum_block_add(skb->csum, csum, odd);
290 	if (icmp_pointers[icmp_param->data.icmph.type].error)
291 		nf_ct_attach(skb, icmp_param->skb);
292 	return 0;
293 }
294 
295 static void icmp_push_reply(struct icmp_bxm *icmp_param,
296 			    struct flowi4 *fl4,
297 			    struct ipcm_cookie *ipc, struct rtable **rt)
298 {
299 	struct sock *sk;
300 	struct sk_buff *skb;
301 
302 	sk = icmp_sk(dev_net((*rt)->dst.dev));
303 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
304 			   icmp_param->data_len+icmp_param->head_len,
305 			   icmp_param->head_len,
306 			   ipc, rt, MSG_DONTWAIT) < 0) {
307 		ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
308 		ip_flush_pending_frames(sk);
309 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
310 		struct icmphdr *icmph = icmp_hdr(skb);
311 		__wsum csum = 0;
312 		struct sk_buff *skb1;
313 
314 		skb_queue_walk(&sk->sk_write_queue, skb1) {
315 			csum = csum_add(csum, skb1->csum);
316 		}
317 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
318 						 (char *)icmph,
319 						 icmp_param->head_len, csum);
320 		icmph->checksum = csum_fold(csum);
321 		skb->ip_summed = CHECKSUM_NONE;
322 		ip_push_pending_frames(sk, fl4);
323 	}
324 }
325 
326 /*
327  *	Driving logic for building and sending ICMP messages.
328  */
329 
330 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
331 {
332 	struct ipcm_cookie ipc;
333 	struct rtable *rt = skb_rtable(skb);
334 	struct net *net = dev_net(rt->dst.dev);
335 	struct flowi4 fl4;
336 	struct sock *sk;
337 	struct inet_sock *inet;
338 	__be32 daddr;
339 
340 	if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
341 		return;
342 
343 	sk = icmp_xmit_lock(net);
344 	if (sk == NULL)
345 		return;
346 	inet = inet_sk(sk);
347 
348 	icmp_param->data.icmph.checksum = 0;
349 
350 	inet->tos = ip_hdr(skb)->tos;
351 	daddr = ipc.addr = ip_hdr(skb)->saddr;
352 	ipc.opt = NULL;
353 	ipc.tx_flags = 0;
354 	if (icmp_param->replyopts.opt.opt.optlen) {
355 		ipc.opt = &icmp_param->replyopts.opt;
356 		if (ipc.opt->opt.srr)
357 			daddr = icmp_param->replyopts.opt.opt.faddr;
358 	}
359 	memset(&fl4, 0, sizeof(fl4));
360 	fl4.daddr = daddr;
361 	fl4.saddr = rt->rt_spec_dst;
362 	fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
363 	fl4.flowi4_proto = IPPROTO_ICMP;
364 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
365 	rt = ip_route_output_key(net, &fl4);
366 	if (IS_ERR(rt))
367 		goto out_unlock;
368 	if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
369 			       icmp_param->data.icmph.code))
370 		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
371 	ip_rt_put(rt);
372 out_unlock:
373 	icmp_xmit_unlock(sk);
374 }
375 
376 static struct rtable *icmp_route_lookup(struct net *net,
377 					struct flowi4 *fl4,
378 					struct sk_buff *skb_in,
379 					const struct iphdr *iph,
380 					__be32 saddr, u8 tos,
381 					int type, int code,
382 					struct icmp_bxm *param)
383 {
384 	struct rtable *rt, *rt2;
385 	struct flowi4 fl4_dec;
386 	int err;
387 
388 	memset(fl4, 0, sizeof(*fl4));
389 	fl4->daddr = (param->replyopts.opt.opt.srr ?
390 		      param->replyopts.opt.opt.faddr : iph->saddr);
391 	fl4->saddr = saddr;
392 	fl4->flowi4_tos = RT_TOS(tos);
393 	fl4->flowi4_proto = IPPROTO_ICMP;
394 	fl4->fl4_icmp_type = type;
395 	fl4->fl4_icmp_code = code;
396 	security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
397 	rt = __ip_route_output_key(net, fl4);
398 	if (IS_ERR(rt))
399 		return rt;
400 
401 	/* No need to clone since we're just using its address. */
402 	rt2 = rt;
403 
404 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
405 					   flowi4_to_flowi(fl4), NULL, 0);
406 	if (!IS_ERR(rt)) {
407 		if (rt != rt2)
408 			return rt;
409 	} else if (PTR_ERR(rt) == -EPERM) {
410 		rt = NULL;
411 	} else
412 		return rt;
413 
414 	err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
415 	if (err)
416 		goto relookup_failed;
417 
418 	if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
419 		rt2 = __ip_route_output_key(net, &fl4_dec);
420 		if (IS_ERR(rt2))
421 			err = PTR_ERR(rt2);
422 	} else {
423 		struct flowi4 fl4_2 = {};
424 		unsigned long orefdst;
425 
426 		fl4_2.daddr = fl4_dec.saddr;
427 		rt2 = ip_route_output_key(net, &fl4_2);
428 		if (IS_ERR(rt2)) {
429 			err = PTR_ERR(rt2);
430 			goto relookup_failed;
431 		}
432 		/* Ugh! */
433 		orefdst = skb_in->_skb_refdst; /* save old refdst */
434 		err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
435 				     RT_TOS(tos), rt2->dst.dev);
436 
437 		dst_release(&rt2->dst);
438 		rt2 = skb_rtable(skb_in);
439 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
440 	}
441 
442 	if (err)
443 		goto relookup_failed;
444 
445 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
446 					    flowi4_to_flowi(&fl4_dec), NULL,
447 					    XFRM_LOOKUP_ICMP);
448 	if (!IS_ERR(rt2)) {
449 		dst_release(&rt->dst);
450 		memcpy(fl4, &fl4_dec, sizeof(*fl4));
451 		rt = rt2;
452 	} else if (PTR_ERR(rt2) == -EPERM) {
453 		if (rt)
454 			dst_release(&rt->dst);
455 		return rt2;
456 	} else {
457 		err = PTR_ERR(rt2);
458 		goto relookup_failed;
459 	}
460 	return rt;
461 
462 relookup_failed:
463 	if (rt)
464 		return rt;
465 	return ERR_PTR(err);
466 }
467 
468 /*
469  *	Send an ICMP message in response to a situation
470  *
471  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
472  *		  MAY send more (we do).
473  *			MUST NOT change this header information.
474  *			MUST NOT reply to a multicast/broadcast IP address.
475  *			MUST NOT reply to a multicast/broadcast MAC address.
476  *			MUST reply to only the first fragment.
477  */
478 
479 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
480 {
481 	struct iphdr *iph;
482 	int room;
483 	struct icmp_bxm icmp_param;
484 	struct rtable *rt = skb_rtable(skb_in);
485 	struct ipcm_cookie ipc;
486 	struct flowi4 fl4;
487 	__be32 saddr;
488 	u8  tos;
489 	struct net *net;
490 	struct sock *sk;
491 
492 	if (!rt)
493 		goto out;
494 	net = dev_net(rt->dst.dev);
495 
496 	/*
497 	 *	Find the original header. It is expected to be valid, of course.
498 	 *	Check this, icmp_send is called from the most obscure devices
499 	 *	sometimes.
500 	 */
501 	iph = ip_hdr(skb_in);
502 
503 	if ((u8 *)iph < skb_in->head ||
504 	    (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
505 		goto out;
506 
507 	/*
508 	 *	No replies to physical multicast/broadcast
509 	 */
510 	if (skb_in->pkt_type != PACKET_HOST)
511 		goto out;
512 
513 	/*
514 	 *	Now check at the protocol level
515 	 */
516 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
517 		goto out;
518 
519 	/*
520 	 *	Only reply to fragment 0. We byte re-order the constant
521 	 *	mask for efficiency.
522 	 */
523 	if (iph->frag_off & htons(IP_OFFSET))
524 		goto out;
525 
526 	/*
527 	 *	If we send an ICMP error to an ICMP error a mess would result..
528 	 */
529 	if (icmp_pointers[type].error) {
530 		/*
531 		 *	We are an error, check if we are replying to an
532 		 *	ICMP error
533 		 */
534 		if (iph->protocol == IPPROTO_ICMP) {
535 			u8 _inner_type, *itp;
536 
537 			itp = skb_header_pointer(skb_in,
538 						 skb_network_header(skb_in) +
539 						 (iph->ihl << 2) +
540 						 offsetof(struct icmphdr,
541 							  type) -
542 						 skb_in->data,
543 						 sizeof(_inner_type),
544 						 &_inner_type);
545 			if (itp == NULL)
546 				goto out;
547 
548 			/*
549 			 *	Assume any unknown ICMP type is an error. This
550 			 *	isn't specified by the RFC, but think about it..
551 			 */
552 			if (*itp > NR_ICMP_TYPES ||
553 			    icmp_pointers[*itp].error)
554 				goto out;
555 		}
556 	}
557 
558 	sk = icmp_xmit_lock(net);
559 	if (sk == NULL)
560 		return;
561 
562 	/*
563 	 *	Construct source address and options.
564 	 */
565 
566 	saddr = iph->daddr;
567 	if (!(rt->rt_flags & RTCF_LOCAL)) {
568 		struct net_device *dev = NULL;
569 
570 		rcu_read_lock();
571 		if (rt_is_input_route(rt) &&
572 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
573 			dev = dev_get_by_index_rcu(net, rt->rt_iif);
574 
575 		if (dev)
576 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
577 		else
578 			saddr = 0;
579 		rcu_read_unlock();
580 	}
581 
582 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
583 					   IPTOS_PREC_INTERNETCONTROL) :
584 					  iph->tos;
585 
586 	if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in))
587 		goto out_unlock;
588 
589 
590 	/*
591 	 *	Prepare data for ICMP header.
592 	 */
593 
594 	icmp_param.data.icmph.type	 = type;
595 	icmp_param.data.icmph.code	 = code;
596 	icmp_param.data.icmph.un.gateway = info;
597 	icmp_param.data.icmph.checksum	 = 0;
598 	icmp_param.skb	  = skb_in;
599 	icmp_param.offset = skb_network_offset(skb_in);
600 	inet_sk(sk)->tos = tos;
601 	ipc.addr = iph->saddr;
602 	ipc.opt = &icmp_param.replyopts.opt;
603 	ipc.tx_flags = 0;
604 
605 	rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos,
606 			       type, code, &icmp_param);
607 	if (IS_ERR(rt))
608 		goto out_unlock;
609 
610 	if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
611 		goto ende;
612 
613 	/* RFC says return as much as we can without exceeding 576 bytes. */
614 
615 	room = dst_mtu(&rt->dst);
616 	if (room > 576)
617 		room = 576;
618 	room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
619 	room -= sizeof(struct icmphdr);
620 
621 	icmp_param.data_len = skb_in->len - icmp_param.offset;
622 	if (icmp_param.data_len > room)
623 		icmp_param.data_len = room;
624 	icmp_param.head_len = sizeof(struct icmphdr);
625 
626 	icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
627 ende:
628 	ip_rt_put(rt);
629 out_unlock:
630 	icmp_xmit_unlock(sk);
631 out:;
632 }
633 EXPORT_SYMBOL(icmp_send);
634 
635 
636 /*
637  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
638  */
639 
640 static void icmp_unreach(struct sk_buff *skb)
641 {
642 	const struct iphdr *iph;
643 	struct icmphdr *icmph;
644 	int hash, protocol;
645 	const struct net_protocol *ipprot;
646 	u32 info = 0;
647 	struct net *net;
648 
649 	net = dev_net(skb_dst(skb)->dev);
650 
651 	/*
652 	 *	Incomplete header ?
653 	 * 	Only checks for the IP header, there should be an
654 	 *	additional check for longer headers in upper levels.
655 	 */
656 
657 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
658 		goto out_err;
659 
660 	icmph = icmp_hdr(skb);
661 	iph   = (const struct iphdr *)skb->data;
662 
663 	if (iph->ihl < 5) /* Mangled header, drop. */
664 		goto out_err;
665 
666 	if (icmph->type == ICMP_DEST_UNREACH) {
667 		switch (icmph->code & 15) {
668 		case ICMP_NET_UNREACH:
669 		case ICMP_HOST_UNREACH:
670 		case ICMP_PROT_UNREACH:
671 		case ICMP_PORT_UNREACH:
672 			break;
673 		case ICMP_FRAG_NEEDED:
674 			if (ipv4_config.no_pmtu_disc) {
675 				LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"),
676 					       &iph->daddr);
677 			} else {
678 				info = ip_rt_frag_needed(net, iph,
679 							 ntohs(icmph->un.frag.mtu),
680 							 skb->dev);
681 				if (!info)
682 					goto out;
683 			}
684 			break;
685 		case ICMP_SR_FAILED:
686 			LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"),
687 				       &iph->daddr);
688 			break;
689 		default:
690 			break;
691 		}
692 		if (icmph->code > NR_ICMP_UNREACH)
693 			goto out;
694 	} else if (icmph->type == ICMP_PARAMETERPROB)
695 		info = ntohl(icmph->un.gateway) >> 24;
696 
697 	/*
698 	 *	Throw it at our lower layers
699 	 *
700 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
701 	 *		  header.
702 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
703 	 *		  transport layer.
704 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
705 	 *		  transport layer.
706 	 */
707 
708 	/*
709 	 *	Check the other end isn't violating RFC 1122. Some routers send
710 	 *	bogus responses to broadcast frames. If you see this message
711 	 *	first check your netmask matches at both ends, if it does then
712 	 *	get the other vendor to fix their kit.
713 	 */
714 
715 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
716 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
717 		if (net_ratelimit())
718 			pr_warn("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
719 				&ip_hdr(skb)->saddr,
720 				icmph->type, icmph->code,
721 				&iph->daddr, skb->dev->name);
722 		goto out;
723 	}
724 
725 	/* Checkin full IP header plus 8 bytes of protocol to
726 	 * avoid additional coding at protocol handlers.
727 	 */
728 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
729 		goto out;
730 
731 	iph = (const struct iphdr *)skb->data;
732 	protocol = iph->protocol;
733 
734 	/*
735 	 *	Deliver ICMP message to raw sockets. Pretty useless feature?
736 	 */
737 	raw_icmp_error(skb, protocol, info);
738 
739 	hash = protocol & (MAX_INET_PROTOS - 1);
740 	rcu_read_lock();
741 	ipprot = rcu_dereference(inet_protos[hash]);
742 	if (ipprot && ipprot->err_handler)
743 		ipprot->err_handler(skb, info);
744 	rcu_read_unlock();
745 
746 out:
747 	return;
748 out_err:
749 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
750 	goto out;
751 }
752 
753 
754 /*
755  *	Handle ICMP_REDIRECT.
756  */
757 
758 static void icmp_redirect(struct sk_buff *skb)
759 {
760 	const struct iphdr *iph;
761 
762 	if (skb->len < sizeof(struct iphdr))
763 		goto out_err;
764 
765 	/*
766 	 *	Get the copied header of the packet that caused the redirect
767 	 */
768 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
769 		goto out;
770 
771 	iph = (const struct iphdr *)skb->data;
772 
773 	switch (icmp_hdr(skb)->code & 7) {
774 	case ICMP_REDIR_NET:
775 	case ICMP_REDIR_NETTOS:
776 		/*
777 		 * As per RFC recommendations now handle it as a host redirect.
778 		 */
779 	case ICMP_REDIR_HOST:
780 	case ICMP_REDIR_HOSTTOS:
781 		ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
782 			       icmp_hdr(skb)->un.gateway,
783 			       iph->saddr, skb->dev);
784 		break;
785 	}
786 
787 	/* Ping wants to see redirects.
788          * Let's pretend they are errors of sorts... */
789 	if (iph->protocol == IPPROTO_ICMP &&
790 	    iph->ihl >= 5 &&
791 	    pskb_may_pull(skb, (iph->ihl<<2)+8)) {
792 		ping_err(skb, icmp_hdr(skb)->un.gateway);
793 	}
794 
795 out:
796 	return;
797 out_err:
798 	ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
799 	goto out;
800 }
801 
802 /*
803  *	Handle ICMP_ECHO ("ping") requests.
804  *
805  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
806  *		  requests.
807  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
808  *		  included in the reply.
809  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
810  *		  echo requests, MUST have default=NOT.
811  *	See also WRT handling of options once they are done and working.
812  */
813 
814 static void icmp_echo(struct sk_buff *skb)
815 {
816 	struct net *net;
817 
818 	net = dev_net(skb_dst(skb)->dev);
819 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
820 		struct icmp_bxm icmp_param;
821 
822 		icmp_param.data.icmph	   = *icmp_hdr(skb);
823 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
824 		icmp_param.skb		   = skb;
825 		icmp_param.offset	   = 0;
826 		icmp_param.data_len	   = skb->len;
827 		icmp_param.head_len	   = sizeof(struct icmphdr);
828 		icmp_reply(&icmp_param, skb);
829 	}
830 }
831 
832 /*
833  *	Handle ICMP Timestamp requests.
834  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
835  *		  SHOULD be in the kernel for minimum random latency.
836  *		  MUST be accurate to a few minutes.
837  *		  MUST be updated at least at 15Hz.
838  */
839 static void icmp_timestamp(struct sk_buff *skb)
840 {
841 	struct timespec tv;
842 	struct icmp_bxm icmp_param;
843 	/*
844 	 *	Too short.
845 	 */
846 	if (skb->len < 4)
847 		goto out_err;
848 
849 	/*
850 	 *	Fill in the current time as ms since midnight UT:
851 	 */
852 	getnstimeofday(&tv);
853 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
854 					 tv.tv_nsec / NSEC_PER_MSEC);
855 	icmp_param.data.times[2] = icmp_param.data.times[1];
856 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
857 		BUG();
858 	icmp_param.data.icmph	   = *icmp_hdr(skb);
859 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
860 	icmp_param.data.icmph.code = 0;
861 	icmp_param.skb		   = skb;
862 	icmp_param.offset	   = 0;
863 	icmp_param.data_len	   = 0;
864 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
865 	icmp_reply(&icmp_param, skb);
866 out:
867 	return;
868 out_err:
869 	ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
870 	goto out;
871 }
872 
873 
874 /*
875  *	Handle ICMP_ADDRESS_MASK requests.  (RFC950)
876  *
877  * RFC1122 (3.2.2.9).  A host MUST only send replies to
878  * ADDRESS_MASK requests if it's been configured as an address mask
879  * agent.  Receiving a request doesn't constitute implicit permission to
880  * act as one. Of course, implementing this correctly requires (SHOULD)
881  * a way to turn the functionality on and off.  Another one for sysctl(),
882  * I guess. -- MS
883  *
884  * RFC1812 (4.3.3.9).	A router MUST implement it.
885  *			A router SHOULD have switch turning it on/off.
886  *		      	This switch MUST be ON by default.
887  *
888  * Gratuitous replies, zero-source replies are not implemented,
889  * that complies with RFC. DO NOT implement them!!! All the idea
890  * of broadcast addrmask replies as specified in RFC950 is broken.
891  * The problem is that it is not uncommon to have several prefixes
892  * on one physical interface. Moreover, addrmask agent can even be
893  * not aware of existing another prefixes.
894  * If source is zero, addrmask agent cannot choose correct prefix.
895  * Gratuitous mask announcements suffer from the same problem.
896  * RFC1812 explains it, but still allows to use ADDRMASK,
897  * that is pretty silly. --ANK
898  *
899  * All these rules are so bizarre, that I removed kernel addrmask
900  * support at all. It is wrong, it is obsolete, nobody uses it in
901  * any case. --ANK
902  *
903  * Furthermore you can do it with a usermode address agent program
904  * anyway...
905  */
906 
907 static void icmp_address(struct sk_buff *skb)
908 {
909 #if 0
910 	if (net_ratelimit())
911 		printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
912 #endif
913 }
914 
915 /*
916  * RFC1812 (4.3.3.9).	A router SHOULD listen all replies, and complain
917  *			loudly if an inconsistency is found.
918  * called with rcu_read_lock()
919  */
920 
921 static void icmp_address_reply(struct sk_buff *skb)
922 {
923 	struct rtable *rt = skb_rtable(skb);
924 	struct net_device *dev = skb->dev;
925 	struct in_device *in_dev;
926 	struct in_ifaddr *ifa;
927 
928 	if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
929 		return;
930 
931 	in_dev = __in_dev_get_rcu(dev);
932 	if (!in_dev)
933 		return;
934 
935 	if (in_dev->ifa_list &&
936 	    IN_DEV_LOG_MARTIANS(in_dev) &&
937 	    IN_DEV_FORWARD(in_dev)) {
938 		__be32 _mask, *mp;
939 
940 		mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
941 		BUG_ON(mp == NULL);
942 		for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
943 			if (*mp == ifa->ifa_mask &&
944 			    inet_ifa_match(ip_hdr(skb)->saddr, ifa))
945 				break;
946 		}
947 		if (!ifa && net_ratelimit()) {
948 			pr_info("Wrong address mask %pI4 from %s/%pI4\n",
949 				mp, dev->name, &ip_hdr(skb)->saddr);
950 		}
951 	}
952 }
953 
954 static void icmp_discard(struct sk_buff *skb)
955 {
956 }
957 
958 /*
959  *	Deal with incoming ICMP packets.
960  */
961 int icmp_rcv(struct sk_buff *skb)
962 {
963 	struct icmphdr *icmph;
964 	struct rtable *rt = skb_rtable(skb);
965 	struct net *net = dev_net(rt->dst.dev);
966 
967 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
968 		struct sec_path *sp = skb_sec_path(skb);
969 		int nh;
970 
971 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
972 				 XFRM_STATE_ICMP))
973 			goto drop;
974 
975 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
976 			goto drop;
977 
978 		nh = skb_network_offset(skb);
979 		skb_set_network_header(skb, sizeof(*icmph));
980 
981 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
982 			goto drop;
983 
984 		skb_set_network_header(skb, nh);
985 	}
986 
987 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
988 
989 	switch (skb->ip_summed) {
990 	case CHECKSUM_COMPLETE:
991 		if (!csum_fold(skb->csum))
992 			break;
993 		/* fall through */
994 	case CHECKSUM_NONE:
995 		skb->csum = 0;
996 		if (__skb_checksum_complete(skb))
997 			goto error;
998 	}
999 
1000 	if (!pskb_pull(skb, sizeof(*icmph)))
1001 		goto error;
1002 
1003 	icmph = icmp_hdr(skb);
1004 
1005 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1006 	/*
1007 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1008 	 *
1009 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1010 	 *		  discarded.
1011 	 */
1012 	if (icmph->type > NR_ICMP_TYPES)
1013 		goto error;
1014 
1015 
1016 	/*
1017 	 *	Parse the ICMP message
1018 	 */
1019 
1020 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1021 		/*
1022 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1023 		 *	  silently ignored (we let user decide with a sysctl).
1024 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1025 		 *	  discarded if to broadcast/multicast.
1026 		 */
1027 		if ((icmph->type == ICMP_ECHO ||
1028 		     icmph->type == ICMP_TIMESTAMP) &&
1029 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1030 			goto error;
1031 		}
1032 		if (icmph->type != ICMP_ECHO &&
1033 		    icmph->type != ICMP_TIMESTAMP &&
1034 		    icmph->type != ICMP_ADDRESS &&
1035 		    icmph->type != ICMP_ADDRESSREPLY) {
1036 			goto error;
1037 		}
1038 	}
1039 
1040 	icmp_pointers[icmph->type].handler(skb);
1041 
1042 drop:
1043 	kfree_skb(skb);
1044 	return 0;
1045 error:
1046 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1047 	goto drop;
1048 }
1049 
1050 /*
1051  *	This table is the definition of how we handle ICMP.
1052  */
1053 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1054 	[ICMP_ECHOREPLY] = {
1055 		.handler = ping_rcv,
1056 	},
1057 	[1] = {
1058 		.handler = icmp_discard,
1059 		.error = 1,
1060 	},
1061 	[2] = {
1062 		.handler = icmp_discard,
1063 		.error = 1,
1064 	},
1065 	[ICMP_DEST_UNREACH] = {
1066 		.handler = icmp_unreach,
1067 		.error = 1,
1068 	},
1069 	[ICMP_SOURCE_QUENCH] = {
1070 		.handler = icmp_unreach,
1071 		.error = 1,
1072 	},
1073 	[ICMP_REDIRECT] = {
1074 		.handler = icmp_redirect,
1075 		.error = 1,
1076 	},
1077 	[6] = {
1078 		.handler = icmp_discard,
1079 		.error = 1,
1080 	},
1081 	[7] = {
1082 		.handler = icmp_discard,
1083 		.error = 1,
1084 	},
1085 	[ICMP_ECHO] = {
1086 		.handler = icmp_echo,
1087 	},
1088 	[9] = {
1089 		.handler = icmp_discard,
1090 		.error = 1,
1091 	},
1092 	[10] = {
1093 		.handler = icmp_discard,
1094 		.error = 1,
1095 	},
1096 	[ICMP_TIME_EXCEEDED] = {
1097 		.handler = icmp_unreach,
1098 		.error = 1,
1099 	},
1100 	[ICMP_PARAMETERPROB] = {
1101 		.handler = icmp_unreach,
1102 		.error = 1,
1103 	},
1104 	[ICMP_TIMESTAMP] = {
1105 		.handler = icmp_timestamp,
1106 	},
1107 	[ICMP_TIMESTAMPREPLY] = {
1108 		.handler = icmp_discard,
1109 	},
1110 	[ICMP_INFO_REQUEST] = {
1111 		.handler = icmp_discard,
1112 	},
1113 	[ICMP_INFO_REPLY] = {
1114 		.handler = icmp_discard,
1115 	},
1116 	[ICMP_ADDRESS] = {
1117 		.handler = icmp_address,
1118 	},
1119 	[ICMP_ADDRESSREPLY] = {
1120 		.handler = icmp_address_reply,
1121 	},
1122 };
1123 
1124 static void __net_exit icmp_sk_exit(struct net *net)
1125 {
1126 	int i;
1127 
1128 	for_each_possible_cpu(i)
1129 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1130 	kfree(net->ipv4.icmp_sk);
1131 	net->ipv4.icmp_sk = NULL;
1132 }
1133 
1134 static int __net_init icmp_sk_init(struct net *net)
1135 {
1136 	int i, err;
1137 
1138 	net->ipv4.icmp_sk =
1139 		kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1140 	if (net->ipv4.icmp_sk == NULL)
1141 		return -ENOMEM;
1142 
1143 	for_each_possible_cpu(i) {
1144 		struct sock *sk;
1145 
1146 		err = inet_ctl_sock_create(&sk, PF_INET,
1147 					   SOCK_RAW, IPPROTO_ICMP, net);
1148 		if (err < 0)
1149 			goto fail;
1150 
1151 		net->ipv4.icmp_sk[i] = sk;
1152 
1153 		/* Enough space for 2 64K ICMP packets, including
1154 		 * sk_buff/skb_shared_info struct overhead.
1155 		 */
1156 		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
1157 
1158 		/*
1159 		 * Speedup sock_wfree()
1160 		 */
1161 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1162 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1163 	}
1164 
1165 	/* Control parameters for ECHO replies. */
1166 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1167 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1168 
1169 	/* Control parameter - ignore bogus broadcast responses? */
1170 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1171 
1172 	/*
1173 	 * 	Configurable global rate limit.
1174 	 *
1175 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1176 	 *	bucket ratemask defines which icmp types are ratelimited by
1177 	 *	setting	it's bit position.
1178 	 *
1179 	 *	default:
1180 	 *	dest unreachable (3), source quench (4),
1181 	 *	time exceeded (11), parameter problem (12)
1182 	 */
1183 
1184 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1185 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1186 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1187 
1188 	return 0;
1189 
1190 fail:
1191 	for_each_possible_cpu(i)
1192 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1193 	kfree(net->ipv4.icmp_sk);
1194 	return err;
1195 }
1196 
1197 static struct pernet_operations __net_initdata icmp_sk_ops = {
1198        .init = icmp_sk_init,
1199        .exit = icmp_sk_exit,
1200 };
1201 
1202 int __init icmp_init(void)
1203 {
1204 	return register_pernet_subsys(&icmp_sk_ops);
1205 }
1206