xref: /linux/net/ipv4/icmp.c (revision 25985edc)
1 /*
2  *	NET3:	Implementation of the ICMP protocol layer.
3  *
4  *		Alan Cox, <alan@lxorguk.ukuu.org.uk>
5  *
6  *	This program is free software; you can redistribute it and/or
7  *	modify it under the terms of the GNU General Public License
8  *	as published by the Free Software Foundation; either version
9  *	2 of the License, or (at your option) any later version.
10  *
11  *	Some of the function names and the icmp unreach table for this
12  *	module were derived from [icmp.c 1.0.11 06/02/93] by
13  *	Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
14  *	Other than that this module is a complete rewrite.
15  *
16  *	Fixes:
17  *	Clemens Fruhwirth	:	introduce global icmp rate limiting
18  *					with icmp type masking ability instead
19  *					of broken per type icmp timeouts.
20  *		Mike Shaver	:	RFC1122 checks.
21  *		Alan Cox	:	Multicast ping reply as self.
22  *		Alan Cox	:	Fix atomicity lockup in ip_build_xmit
23  *					call.
24  *		Alan Cox	:	Added 216,128 byte paths to the MTU
25  *					code.
26  *		Martin Mares	:	RFC1812 checks.
27  *		Martin Mares	:	Can be configured to follow redirects
28  *					if acting as a router _without_ a
29  *					routing protocol (RFC 1812).
30  *		Martin Mares	:	Echo requests may be configured to
31  *					be ignored (RFC 1812).
32  *		Martin Mares	:	Limitation of ICMP error message
33  *					transmit rate (RFC 1812).
34  *		Martin Mares	:	TOS and Precedence set correctly
35  *					(RFC 1812).
36  *		Martin Mares	:	Now copying as much data from the
37  *					original packet as we can without
38  *					exceeding 576 bytes (RFC 1812).
39  *	Willy Konynenberg	:	Transparent proxying support.
40  *		Keith Owens	:	RFC1191 correction for 4.2BSD based
41  *					path MTU bug.
42  *		Thomas Quinot	:	ICMP Dest Unreach codes up to 15 are
43  *					valid (RFC 1812).
44  *		Andi Kleen	:	Check all packet lengths properly
45  *					and moved all kfree_skb() up to
46  *					icmp_rcv.
47  *		Andi Kleen	:	Move the rate limit bookkeeping
48  *					into the dest entry and use a token
49  *					bucket filter (thanks to ANK). Make
50  *					the rates sysctl configurable.
51  *		Yu Tianli	:	Fixed two ugly bugs in icmp_send
52  *					- IP option length was accounted wrongly
53  *					- ICMP header length was not accounted
54  *					  at all.
55  *              Tristan Greaves :       Added sysctl option to ignore bogus
56  *              			broadcast responses from broken routers.
57  *
58  * To Fix:
59  *
60  *	- Should use skb_pull() instead of all the manual checking.
61  *	  This would also greatly simply some upper layer error handlers. --AK
62  *
63  */
64 
65 #include <linux/module.h>
66 #include <linux/types.h>
67 #include <linux/jiffies.h>
68 #include <linux/kernel.h>
69 #include <linux/fcntl.h>
70 #include <linux/socket.h>
71 #include <linux/in.h>
72 #include <linux/inet.h>
73 #include <linux/inetdevice.h>
74 #include <linux/netdevice.h>
75 #include <linux/string.h>
76 #include <linux/netfilter_ipv4.h>
77 #include <linux/slab.h>
78 #include <net/snmp.h>
79 #include <net/ip.h>
80 #include <net/route.h>
81 #include <net/protocol.h>
82 #include <net/icmp.h>
83 #include <net/tcp.h>
84 #include <net/udp.h>
85 #include <net/raw.h>
86 #include <linux/skbuff.h>
87 #include <net/sock.h>
88 #include <linux/errno.h>
89 #include <linux/timer.h>
90 #include <linux/init.h>
91 #include <asm/system.h>
92 #include <asm/uaccess.h>
93 #include <net/checksum.h>
94 #include <net/xfrm.h>
95 #include <net/inet_common.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 /* An array of errno for error messages from dest unreach. */
116 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
117 
118 const struct icmp_err icmp_err_convert[] = {
119 	{
120 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
121 		.fatal = 0,
122 	},
123 	{
124 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
125 		.fatal = 0,
126 	},
127 	{
128 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
129 		.fatal = 1,
130 	},
131 	{
132 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
133 		.fatal = 1,
134 	},
135 	{
136 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
137 		.fatal = 0,
138 	},
139 	{
140 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
141 		.fatal = 0,
142 	},
143 	{
144 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
145 		.fatal = 1,
146 	},
147 	{
148 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
149 		.fatal = 1,
150 	},
151 	{
152 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
153 		.fatal = 1,
154 	},
155 	{
156 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
157 		.fatal = 1,
158 	},
159 	{
160 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
161 		.fatal = 1,
162 	},
163 	{
164 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
165 		.fatal = 0,
166 	},
167 	{
168 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
169 		.fatal = 0,
170 	},
171 	{
172 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
173 		.fatal = 1,
174 	},
175 	{
176 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
177 		.fatal = 1,
178 	},
179 	{
180 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
181 		.fatal = 1,
182 	},
183 };
184 EXPORT_SYMBOL(icmp_err_convert);
185 
186 /*
187  *	ICMP control array. This specifies what to do with each ICMP.
188  */
189 
190 struct icmp_control {
191 	void (*handler)(struct sk_buff *skb);
192 	short   error;		/* This ICMP is classed as an error message */
193 };
194 
195 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
196 
197 /*
198  *	The ICMP socket(s). This is the most convenient way to flow control
199  *	our ICMP output as well as maintain a clean interface throughout
200  *	all layers. All Socketless IP sends will soon be gone.
201  *
202  *	On SMP we have one ICMP socket per-cpu.
203  */
204 static struct sock *icmp_sk(struct net *net)
205 {
206 	return net->ipv4.icmp_sk[smp_processor_id()];
207 }
208 
209 static inline struct sock *icmp_xmit_lock(struct net *net)
210 {
211 	struct sock *sk;
212 
213 	local_bh_disable();
214 
215 	sk = icmp_sk(net);
216 
217 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
218 		/* This can happen if the output path signals a
219 		 * dst_link_failure() for an outgoing ICMP packet.
220 		 */
221 		local_bh_enable();
222 		return NULL;
223 	}
224 	return sk;
225 }
226 
227 static inline void icmp_xmit_unlock(struct sock *sk)
228 {
229 	spin_unlock_bh(&sk->sk_lock.slock);
230 }
231 
232 /*
233  *	Send an ICMP frame.
234  */
235 
236 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
237 		int type, int code)
238 {
239 	struct dst_entry *dst = &rt->dst;
240 	bool rc = true;
241 
242 	if (type > NR_ICMP_TYPES)
243 		goto out;
244 
245 	/* Don't limit PMTU discovery. */
246 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
247 		goto out;
248 
249 	/* No rate limit on loopback */
250 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
251 		goto out;
252 
253 	/* Limit if icmp type is enabled in ratemask. */
254 	if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
255 		if (!rt->peer)
256 			rt_bind_peer(rt, 1);
257 		rc = inet_peer_xrlim_allow(rt->peer,
258 					   net->ipv4.sysctl_icmp_ratelimit);
259 	}
260 out:
261 	return rc;
262 }
263 
264 /*
265  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
266  */
267 void icmp_out_count(struct net *net, unsigned char type)
268 {
269 	ICMPMSGOUT_INC_STATS(net, type);
270 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
271 }
272 
273 /*
274  *	Checksum each fragment, and on the first include the headers and final
275  *	checksum.
276  */
277 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
278 			  struct sk_buff *skb)
279 {
280 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
281 	__wsum csum;
282 
283 	csum = skb_copy_and_csum_bits(icmp_param->skb,
284 				      icmp_param->offset + offset,
285 				      to, len, 0);
286 
287 	skb->csum = csum_block_add(skb->csum, csum, odd);
288 	if (icmp_pointers[icmp_param->data.icmph.type].error)
289 		nf_ct_attach(skb, icmp_param->skb);
290 	return 0;
291 }
292 
293 static void icmp_push_reply(struct icmp_bxm *icmp_param,
294 			    struct ipcm_cookie *ipc, struct rtable **rt)
295 {
296 	struct sock *sk;
297 	struct sk_buff *skb;
298 
299 	sk = icmp_sk(dev_net((*rt)->dst.dev));
300 	if (ip_append_data(sk, icmp_glue_bits, icmp_param,
301 			   icmp_param->data_len+icmp_param->head_len,
302 			   icmp_param->head_len,
303 			   ipc, rt, MSG_DONTWAIT) < 0) {
304 		ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
305 		ip_flush_pending_frames(sk);
306 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
307 		struct icmphdr *icmph = icmp_hdr(skb);
308 		__wsum csum = 0;
309 		struct sk_buff *skb1;
310 
311 		skb_queue_walk(&sk->sk_write_queue, skb1) {
312 			csum = csum_add(csum, skb1->csum);
313 		}
314 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
315 						 (char *)icmph,
316 						 icmp_param->head_len, csum);
317 		icmph->checksum = csum_fold(csum);
318 		skb->ip_summed = CHECKSUM_NONE;
319 		ip_push_pending_frames(sk);
320 	}
321 }
322 
323 /*
324  *	Driving logic for building and sending ICMP messages.
325  */
326 
327 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
328 {
329 	struct ipcm_cookie ipc;
330 	struct rtable *rt = skb_rtable(skb);
331 	struct net *net = dev_net(rt->dst.dev);
332 	struct sock *sk;
333 	struct inet_sock *inet;
334 	__be32 daddr;
335 
336 	if (ip_options_echo(&icmp_param->replyopts, skb))
337 		return;
338 
339 	sk = icmp_xmit_lock(net);
340 	if (sk == NULL)
341 		return;
342 	inet = inet_sk(sk);
343 
344 	icmp_param->data.icmph.checksum = 0;
345 
346 	inet->tos = ip_hdr(skb)->tos;
347 	daddr = ipc.addr = rt->rt_src;
348 	ipc.opt = NULL;
349 	ipc.tx_flags = 0;
350 	if (icmp_param->replyopts.optlen) {
351 		ipc.opt = &icmp_param->replyopts;
352 		if (ipc.opt->srr)
353 			daddr = icmp_param->replyopts.faddr;
354 	}
355 	{
356 		struct flowi4 fl4 = {
357 			.daddr = daddr,
358 			.saddr = rt->rt_spec_dst,
359 			.flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
360 			.flowi4_proto = IPPROTO_ICMP,
361 		};
362 		security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
363 		rt = ip_route_output_key(net, &fl4);
364 		if (IS_ERR(rt))
365 			goto out_unlock;
366 	}
367 	if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
368 			       icmp_param->data.icmph.code))
369 		icmp_push_reply(icmp_param, &ipc, &rt);
370 	ip_rt_put(rt);
371 out_unlock:
372 	icmp_xmit_unlock(sk);
373 }
374 
375 static struct rtable *icmp_route_lookup(struct net *net, struct sk_buff *skb_in,
376 					struct iphdr *iph,
377 					__be32 saddr, u8 tos,
378 					int type, int code,
379 					struct icmp_bxm *param)
380 {
381 	struct flowi4 fl4 = {
382 		.daddr = (param->replyopts.srr ?
383 			  param->replyopts.faddr : iph->saddr),
384 		.saddr = saddr,
385 		.flowi4_tos = RT_TOS(tos),
386 		.flowi4_proto = IPPROTO_ICMP,
387 		.fl4_icmp_type = type,
388 		.fl4_icmp_code = code,
389 	};
390 	struct rtable *rt, *rt2;
391 	int err;
392 
393 	security_skb_classify_flow(skb_in, flowi4_to_flowi(&fl4));
394 	rt = __ip_route_output_key(net, &fl4);
395 	if (IS_ERR(rt))
396 		return rt;
397 
398 	/* No need to clone since we're just using its address. */
399 	rt2 = rt;
400 
401 	if (!fl4.saddr)
402 		fl4.saddr = rt->rt_src;
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), AF_INET);
415 	if (err)
416 		goto relookup_failed;
417 
418 	if (inet_addr_type(net, fl4.saddr) == RTN_LOCAL) {
419 		rt2 = __ip_route_output_key(net, &fl4);
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.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.daddr, fl4.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), NULL,
447 					    XFRM_LOOKUP_ICMP);
448 	if (!IS_ERR(rt2)) {
449 		dst_release(&rt->dst);
450 		rt = rt2;
451 	} else if (PTR_ERR(rt2) == -EPERM) {
452 		if (rt)
453 			dst_release(&rt->dst);
454 		return rt2;
455 	} else {
456 		err = PTR_ERR(rt2);
457 		goto relookup_failed;
458 	}
459 	return rt;
460 
461 relookup_failed:
462 	if (rt)
463 		return rt;
464 	return ERR_PTR(err);
465 }
466 
467 /*
468  *	Send an ICMP message in response to a situation
469  *
470  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
471  *		  MAY send more (we do).
472  *			MUST NOT change this header information.
473  *			MUST NOT reply to a multicast/broadcast IP address.
474  *			MUST NOT reply to a multicast/broadcast MAC address.
475  *			MUST reply to only the first fragment.
476  */
477 
478 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
479 {
480 	struct iphdr *iph;
481 	int room;
482 	struct icmp_bxm icmp_param;
483 	struct rtable *rt = skb_rtable(skb_in);
484 	struct ipcm_cookie ipc;
485 	__be32 saddr;
486 	u8  tos;
487 	struct net *net;
488 	struct sock *sk;
489 
490 	if (!rt)
491 		goto out;
492 	net = dev_net(rt->dst.dev);
493 
494 	/*
495 	 *	Find the original header. It is expected to be valid, of course.
496 	 *	Check this, icmp_send is called from the most obscure devices
497 	 *	sometimes.
498 	 */
499 	iph = ip_hdr(skb_in);
500 
501 	if ((u8 *)iph < skb_in->head ||
502 	    (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
503 		goto out;
504 
505 	/*
506 	 *	No replies to physical multicast/broadcast
507 	 */
508 	if (skb_in->pkt_type != PACKET_HOST)
509 		goto out;
510 
511 	/*
512 	 *	Now check at the protocol level
513 	 */
514 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
515 		goto out;
516 
517 	/*
518 	 *	Only reply to fragment 0. We byte re-order the constant
519 	 *	mask for efficiency.
520 	 */
521 	if (iph->frag_off & htons(IP_OFFSET))
522 		goto out;
523 
524 	/*
525 	 *	If we send an ICMP error to an ICMP error a mess would result..
526 	 */
527 	if (icmp_pointers[type].error) {
528 		/*
529 		 *	We are an error, check if we are replying to an
530 		 *	ICMP error
531 		 */
532 		if (iph->protocol == IPPROTO_ICMP) {
533 			u8 _inner_type, *itp;
534 
535 			itp = skb_header_pointer(skb_in,
536 						 skb_network_header(skb_in) +
537 						 (iph->ihl << 2) +
538 						 offsetof(struct icmphdr,
539 							  type) -
540 						 skb_in->data,
541 						 sizeof(_inner_type),
542 						 &_inner_type);
543 			if (itp == NULL)
544 				goto out;
545 
546 			/*
547 			 *	Assume any unknown ICMP type is an error. This
548 			 *	isn't specified by the RFC, but think about it..
549 			 */
550 			if (*itp > NR_ICMP_TYPES ||
551 			    icmp_pointers[*itp].error)
552 				goto out;
553 		}
554 	}
555 
556 	sk = icmp_xmit_lock(net);
557 	if (sk == NULL)
558 		return;
559 
560 	/*
561 	 *	Construct source address and options.
562 	 */
563 
564 	saddr = iph->daddr;
565 	if (!(rt->rt_flags & RTCF_LOCAL)) {
566 		struct net_device *dev = NULL;
567 
568 		rcu_read_lock();
569 		if (rt_is_input_route(rt) &&
570 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
571 			dev = dev_get_by_index_rcu(net, rt->rt_iif);
572 
573 		if (dev)
574 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
575 		else
576 			saddr = 0;
577 		rcu_read_unlock();
578 	}
579 
580 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
581 					   IPTOS_PREC_INTERNETCONTROL) :
582 					  iph->tos;
583 
584 	if (ip_options_echo(&icmp_param.replyopts, skb_in))
585 		goto out_unlock;
586 
587 
588 	/*
589 	 *	Prepare data for ICMP header.
590 	 */
591 
592 	icmp_param.data.icmph.type	 = type;
593 	icmp_param.data.icmph.code	 = code;
594 	icmp_param.data.icmph.un.gateway = info;
595 	icmp_param.data.icmph.checksum	 = 0;
596 	icmp_param.skb	  = skb_in;
597 	icmp_param.offset = skb_network_offset(skb_in);
598 	inet_sk(sk)->tos = tos;
599 	ipc.addr = iph->saddr;
600 	ipc.opt = &icmp_param.replyopts;
601 	ipc.tx_flags = 0;
602 
603 	rt = icmp_route_lookup(net, skb_in, iph, saddr, tos,
604 			       type, code, &icmp_param);
605 	if (IS_ERR(rt))
606 		goto out_unlock;
607 
608 	if (!icmpv4_xrlim_allow(net, rt, type, code))
609 		goto ende;
610 
611 	/* RFC says return as much as we can without exceeding 576 bytes. */
612 
613 	room = dst_mtu(&rt->dst);
614 	if (room > 576)
615 		room = 576;
616 	room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
617 	room -= sizeof(struct icmphdr);
618 
619 	icmp_param.data_len = skb_in->len - icmp_param.offset;
620 	if (icmp_param.data_len > room)
621 		icmp_param.data_len = room;
622 	icmp_param.head_len = sizeof(struct icmphdr);
623 
624 	icmp_push_reply(&icmp_param, &ipc, &rt);
625 ende:
626 	ip_rt_put(rt);
627 out_unlock:
628 	icmp_xmit_unlock(sk);
629 out:;
630 }
631 EXPORT_SYMBOL(icmp_send);
632 
633 
634 /*
635  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
636  */
637 
638 static void icmp_unreach(struct sk_buff *skb)
639 {
640 	struct iphdr *iph;
641 	struct icmphdr *icmph;
642 	int hash, protocol;
643 	const struct net_protocol *ipprot;
644 	u32 info = 0;
645 	struct net *net;
646 
647 	net = dev_net(skb_dst(skb)->dev);
648 
649 	/*
650 	 *	Incomplete header ?
651 	 * 	Only checks for the IP header, there should be an
652 	 *	additional check for longer headers in upper levels.
653 	 */
654 
655 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
656 		goto out_err;
657 
658 	icmph = icmp_hdr(skb);
659 	iph   = (struct iphdr *)skb->data;
660 
661 	if (iph->ihl < 5) /* Mangled header, drop. */
662 		goto out_err;
663 
664 	if (icmph->type == ICMP_DEST_UNREACH) {
665 		switch (icmph->code & 15) {
666 		case ICMP_NET_UNREACH:
667 		case ICMP_HOST_UNREACH:
668 		case ICMP_PROT_UNREACH:
669 		case ICMP_PORT_UNREACH:
670 			break;
671 		case ICMP_FRAG_NEEDED:
672 			if (ipv4_config.no_pmtu_disc) {
673 				LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n",
674 					       &iph->daddr);
675 			} else {
676 				info = ip_rt_frag_needed(net, iph,
677 							 ntohs(icmph->un.frag.mtu),
678 							 skb->dev);
679 				if (!info)
680 					goto out;
681 			}
682 			break;
683 		case ICMP_SR_FAILED:
684 			LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n",
685 				       &iph->daddr);
686 			break;
687 		default:
688 			break;
689 		}
690 		if (icmph->code > NR_ICMP_UNREACH)
691 			goto out;
692 	} else if (icmph->type == ICMP_PARAMETERPROB)
693 		info = ntohl(icmph->un.gateway) >> 24;
694 
695 	/*
696 	 *	Throw it at our lower layers
697 	 *
698 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
699 	 *		  header.
700 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
701 	 *		  transport layer.
702 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
703 	 *		  transport layer.
704 	 */
705 
706 	/*
707 	 *	Check the other end isn't violating RFC 1122. Some routers send
708 	 *	bogus responses to broadcast frames. If you see this message
709 	 *	first check your netmask matches at both ends, if it does then
710 	 *	get the other vendor to fix their kit.
711 	 */
712 
713 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
714 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
715 		if (net_ratelimit())
716 			printk(KERN_WARNING "%pI4 sent an invalid ICMP "
717 					    "type %u, code %u "
718 					    "error to a broadcast: %pI4 on %s\n",
719 			       &ip_hdr(skb)->saddr,
720 			       icmph->type, icmph->code,
721 			       &iph->daddr,
722 			       skb->dev->name);
723 		goto out;
724 	}
725 
726 	/* Checkin full IP header plus 8 bytes of protocol to
727 	 * avoid additional coding at protocol handlers.
728 	 */
729 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
730 		goto out;
731 
732 	iph = (struct iphdr *)skb->data;
733 	protocol = iph->protocol;
734 
735 	/*
736 	 *	Deliver ICMP message to raw sockets. Pretty useless feature?
737 	 */
738 	raw_icmp_error(skb, protocol, info);
739 
740 	hash = protocol & (MAX_INET_PROTOS - 1);
741 	rcu_read_lock();
742 	ipprot = rcu_dereference(inet_protos[hash]);
743 	if (ipprot && ipprot->err_handler)
744 		ipprot->err_handler(skb, info);
745 	rcu_read_unlock();
746 
747 out:
748 	return;
749 out_err:
750 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
751 	goto out;
752 }
753 
754 
755 /*
756  *	Handle ICMP_REDIRECT.
757  */
758 
759 static void icmp_redirect(struct sk_buff *skb)
760 {
761 	struct iphdr *iph;
762 
763 	if (skb->len < sizeof(struct iphdr))
764 		goto out_err;
765 
766 	/*
767 	 *	Get the copied header of the packet that caused the redirect
768 	 */
769 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
770 		goto out;
771 
772 	iph = (struct iphdr *)skb->data;
773 
774 	switch (icmp_hdr(skb)->code & 7) {
775 	case ICMP_REDIR_NET:
776 	case ICMP_REDIR_NETTOS:
777 		/*
778 		 * As per RFC recommendations now handle it as a host redirect.
779 		 */
780 	case ICMP_REDIR_HOST:
781 	case ICMP_REDIR_HOSTTOS:
782 		ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
783 			       icmp_hdr(skb)->un.gateway,
784 			       iph->saddr, skb->dev);
785 		break;
786 	}
787 out:
788 	return;
789 out_err:
790 	ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
791 	goto out;
792 }
793 
794 /*
795  *	Handle ICMP_ECHO ("ping") requests.
796  *
797  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
798  *		  requests.
799  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
800  *		  included in the reply.
801  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
802  *		  echo requests, MUST have default=NOT.
803  *	See also WRT handling of options once they are done and working.
804  */
805 
806 static void icmp_echo(struct sk_buff *skb)
807 {
808 	struct net *net;
809 
810 	net = dev_net(skb_dst(skb)->dev);
811 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
812 		struct icmp_bxm icmp_param;
813 
814 		icmp_param.data.icmph	   = *icmp_hdr(skb);
815 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
816 		icmp_param.skb		   = skb;
817 		icmp_param.offset	   = 0;
818 		icmp_param.data_len	   = skb->len;
819 		icmp_param.head_len	   = sizeof(struct icmphdr);
820 		icmp_reply(&icmp_param, skb);
821 	}
822 }
823 
824 /*
825  *	Handle ICMP Timestamp requests.
826  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
827  *		  SHOULD be in the kernel for minimum random latency.
828  *		  MUST be accurate to a few minutes.
829  *		  MUST be updated at least at 15Hz.
830  */
831 static void icmp_timestamp(struct sk_buff *skb)
832 {
833 	struct timespec tv;
834 	struct icmp_bxm icmp_param;
835 	/*
836 	 *	Too short.
837 	 */
838 	if (skb->len < 4)
839 		goto out_err;
840 
841 	/*
842 	 *	Fill in the current time as ms since midnight UT:
843 	 */
844 	getnstimeofday(&tv);
845 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
846 					 tv.tv_nsec / NSEC_PER_MSEC);
847 	icmp_param.data.times[2] = icmp_param.data.times[1];
848 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
849 		BUG();
850 	icmp_param.data.icmph	   = *icmp_hdr(skb);
851 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
852 	icmp_param.data.icmph.code = 0;
853 	icmp_param.skb		   = skb;
854 	icmp_param.offset	   = 0;
855 	icmp_param.data_len	   = 0;
856 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
857 	icmp_reply(&icmp_param, skb);
858 out:
859 	return;
860 out_err:
861 	ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
862 	goto out;
863 }
864 
865 
866 /*
867  *	Handle ICMP_ADDRESS_MASK requests.  (RFC950)
868  *
869  * RFC1122 (3.2.2.9).  A host MUST only send replies to
870  * ADDRESS_MASK requests if it's been configured as an address mask
871  * agent.  Receiving a request doesn't constitute implicit permission to
872  * act as one. Of course, implementing this correctly requires (SHOULD)
873  * a way to turn the functionality on and off.  Another one for sysctl(),
874  * I guess. -- MS
875  *
876  * RFC1812 (4.3.3.9).	A router MUST implement it.
877  *			A router SHOULD have switch turning it on/off.
878  *		      	This switch MUST be ON by default.
879  *
880  * Gratuitous replies, zero-source replies are not implemented,
881  * that complies with RFC. DO NOT implement them!!! All the idea
882  * of broadcast addrmask replies as specified in RFC950 is broken.
883  * The problem is that it is not uncommon to have several prefixes
884  * on one physical interface. Moreover, addrmask agent can even be
885  * not aware of existing another prefixes.
886  * If source is zero, addrmask agent cannot choose correct prefix.
887  * Gratuitous mask announcements suffer from the same problem.
888  * RFC1812 explains it, but still allows to use ADDRMASK,
889  * that is pretty silly. --ANK
890  *
891  * All these rules are so bizarre, that I removed kernel addrmask
892  * support at all. It is wrong, it is obsolete, nobody uses it in
893  * any case. --ANK
894  *
895  * Furthermore you can do it with a usermode address agent program
896  * anyway...
897  */
898 
899 static void icmp_address(struct sk_buff *skb)
900 {
901 #if 0
902 	if (net_ratelimit())
903 		printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
904 #endif
905 }
906 
907 /*
908  * RFC1812 (4.3.3.9).	A router SHOULD listen all replies, and complain
909  *			loudly if an inconsistency is found.
910  * called with rcu_read_lock()
911  */
912 
913 static void icmp_address_reply(struct sk_buff *skb)
914 {
915 	struct rtable *rt = skb_rtable(skb);
916 	struct net_device *dev = skb->dev;
917 	struct in_device *in_dev;
918 	struct in_ifaddr *ifa;
919 
920 	if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
921 		return;
922 
923 	in_dev = __in_dev_get_rcu(dev);
924 	if (!in_dev)
925 		return;
926 
927 	if (in_dev->ifa_list &&
928 	    IN_DEV_LOG_MARTIANS(in_dev) &&
929 	    IN_DEV_FORWARD(in_dev)) {
930 		__be32 _mask, *mp;
931 
932 		mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
933 		BUG_ON(mp == NULL);
934 		for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
935 			if (*mp == ifa->ifa_mask &&
936 			    inet_ifa_match(rt->rt_src, ifa))
937 				break;
938 		}
939 		if (!ifa && net_ratelimit()) {
940 			printk(KERN_INFO "Wrong address mask %pI4 from %s/%pI4\n",
941 			       mp, dev->name, &rt->rt_src);
942 		}
943 	}
944 }
945 
946 static void icmp_discard(struct sk_buff *skb)
947 {
948 }
949 
950 /*
951  *	Deal with incoming ICMP packets.
952  */
953 int icmp_rcv(struct sk_buff *skb)
954 {
955 	struct icmphdr *icmph;
956 	struct rtable *rt = skb_rtable(skb);
957 	struct net *net = dev_net(rt->dst.dev);
958 
959 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
960 		struct sec_path *sp = skb_sec_path(skb);
961 		int nh;
962 
963 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
964 				 XFRM_STATE_ICMP))
965 			goto drop;
966 
967 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
968 			goto drop;
969 
970 		nh = skb_network_offset(skb);
971 		skb_set_network_header(skb, sizeof(*icmph));
972 
973 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
974 			goto drop;
975 
976 		skb_set_network_header(skb, nh);
977 	}
978 
979 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
980 
981 	switch (skb->ip_summed) {
982 	case CHECKSUM_COMPLETE:
983 		if (!csum_fold(skb->csum))
984 			break;
985 		/* fall through */
986 	case CHECKSUM_NONE:
987 		skb->csum = 0;
988 		if (__skb_checksum_complete(skb))
989 			goto error;
990 	}
991 
992 	if (!pskb_pull(skb, sizeof(*icmph)))
993 		goto error;
994 
995 	icmph = icmp_hdr(skb);
996 
997 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
998 	/*
999 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1000 	 *
1001 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1002 	 *		  discarded.
1003 	 */
1004 	if (icmph->type > NR_ICMP_TYPES)
1005 		goto error;
1006 
1007 
1008 	/*
1009 	 *	Parse the ICMP message
1010 	 */
1011 
1012 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1013 		/*
1014 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1015 		 *	  silently ignored (we let user decide with a sysctl).
1016 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1017 		 *	  discarded if to broadcast/multicast.
1018 		 */
1019 		if ((icmph->type == ICMP_ECHO ||
1020 		     icmph->type == ICMP_TIMESTAMP) &&
1021 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1022 			goto error;
1023 		}
1024 		if (icmph->type != ICMP_ECHO &&
1025 		    icmph->type != ICMP_TIMESTAMP &&
1026 		    icmph->type != ICMP_ADDRESS &&
1027 		    icmph->type != ICMP_ADDRESSREPLY) {
1028 			goto error;
1029 		}
1030 	}
1031 
1032 	icmp_pointers[icmph->type].handler(skb);
1033 
1034 drop:
1035 	kfree_skb(skb);
1036 	return 0;
1037 error:
1038 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1039 	goto drop;
1040 }
1041 
1042 /*
1043  *	This table is the definition of how we handle ICMP.
1044  */
1045 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1046 	[ICMP_ECHOREPLY] = {
1047 		.handler = icmp_discard,
1048 	},
1049 	[1] = {
1050 		.handler = icmp_discard,
1051 		.error = 1,
1052 	},
1053 	[2] = {
1054 		.handler = icmp_discard,
1055 		.error = 1,
1056 	},
1057 	[ICMP_DEST_UNREACH] = {
1058 		.handler = icmp_unreach,
1059 		.error = 1,
1060 	},
1061 	[ICMP_SOURCE_QUENCH] = {
1062 		.handler = icmp_unreach,
1063 		.error = 1,
1064 	},
1065 	[ICMP_REDIRECT] = {
1066 		.handler = icmp_redirect,
1067 		.error = 1,
1068 	},
1069 	[6] = {
1070 		.handler = icmp_discard,
1071 		.error = 1,
1072 	},
1073 	[7] = {
1074 		.handler = icmp_discard,
1075 		.error = 1,
1076 	},
1077 	[ICMP_ECHO] = {
1078 		.handler = icmp_echo,
1079 	},
1080 	[9] = {
1081 		.handler = icmp_discard,
1082 		.error = 1,
1083 	},
1084 	[10] = {
1085 		.handler = icmp_discard,
1086 		.error = 1,
1087 	},
1088 	[ICMP_TIME_EXCEEDED] = {
1089 		.handler = icmp_unreach,
1090 		.error = 1,
1091 	},
1092 	[ICMP_PARAMETERPROB] = {
1093 		.handler = icmp_unreach,
1094 		.error = 1,
1095 	},
1096 	[ICMP_TIMESTAMP] = {
1097 		.handler = icmp_timestamp,
1098 	},
1099 	[ICMP_TIMESTAMPREPLY] = {
1100 		.handler = icmp_discard,
1101 	},
1102 	[ICMP_INFO_REQUEST] = {
1103 		.handler = icmp_discard,
1104 	},
1105 	[ICMP_INFO_REPLY] = {
1106 		.handler = icmp_discard,
1107 	},
1108 	[ICMP_ADDRESS] = {
1109 		.handler = icmp_address,
1110 	},
1111 	[ICMP_ADDRESSREPLY] = {
1112 		.handler = icmp_address_reply,
1113 	},
1114 };
1115 
1116 static void __net_exit icmp_sk_exit(struct net *net)
1117 {
1118 	int i;
1119 
1120 	for_each_possible_cpu(i)
1121 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1122 	kfree(net->ipv4.icmp_sk);
1123 	net->ipv4.icmp_sk = NULL;
1124 }
1125 
1126 static int __net_init icmp_sk_init(struct net *net)
1127 {
1128 	int i, err;
1129 
1130 	net->ipv4.icmp_sk =
1131 		kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1132 	if (net->ipv4.icmp_sk == NULL)
1133 		return -ENOMEM;
1134 
1135 	for_each_possible_cpu(i) {
1136 		struct sock *sk;
1137 
1138 		err = inet_ctl_sock_create(&sk, PF_INET,
1139 					   SOCK_RAW, IPPROTO_ICMP, net);
1140 		if (err < 0)
1141 			goto fail;
1142 
1143 		net->ipv4.icmp_sk[i] = sk;
1144 
1145 		/* Enough space for 2 64K ICMP packets, including
1146 		 * sk_buff struct overhead.
1147 		 */
1148 		sk->sk_sndbuf =
1149 			(2 * ((64 * 1024) + sizeof(struct sk_buff)));
1150 
1151 		/*
1152 		 * Speedup sock_wfree()
1153 		 */
1154 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1155 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1156 	}
1157 
1158 	/* Control parameters for ECHO replies. */
1159 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1160 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1161 
1162 	/* Control parameter - ignore bogus broadcast responses? */
1163 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1164 
1165 	/*
1166 	 * 	Configurable global rate limit.
1167 	 *
1168 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1169 	 *	bucket ratemask defines which icmp types are ratelimited by
1170 	 *	setting	it's bit position.
1171 	 *
1172 	 *	default:
1173 	 *	dest unreachable (3), source quench (4),
1174 	 *	time exceeded (11), parameter problem (12)
1175 	 */
1176 
1177 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1178 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1179 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1180 
1181 	return 0;
1182 
1183 fail:
1184 	for_each_possible_cpu(i)
1185 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1186 	kfree(net->ipv4.icmp_sk);
1187 	return err;
1188 }
1189 
1190 static struct pernet_operations __net_initdata icmp_sk_ops = {
1191        .init = icmp_sk_init,
1192        .exit = icmp_sk_exit,
1193 };
1194 
1195 int __init icmp_init(void)
1196 {
1197 	return register_pernet_subsys(&icmp_sk_ops);
1198 }
1199