xref: /linux/net/ipv4/icmp.c (revision 5e2b61f7)
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 flowi fl = { .fl4_dst= daddr,
357 				    .fl4_src = rt->rt_spec_dst,
358 				    .fl4_tos = RT_TOS(ip_hdr(skb)->tos),
359 				    .proto = IPPROTO_ICMP };
360 		security_skb_classify_flow(skb, &fl);
361 		rt = ip_route_output_key(net, &fl);
362 		if (IS_ERR(rt))
363 			goto out_unlock;
364 	}
365 	if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
366 			       icmp_param->data.icmph.code))
367 		icmp_push_reply(icmp_param, &ipc, &rt);
368 	ip_rt_put(rt);
369 out_unlock:
370 	icmp_xmit_unlock(sk);
371 }
372 
373 static struct rtable *icmp_route_lookup(struct net *net, struct sk_buff *skb_in,
374 					struct iphdr *iph,
375 					__be32 saddr, u8 tos,
376 					int type, int code,
377 					struct icmp_bxm *param)
378 {
379 	struct flowi fl = {
380 		.fl4_dst = (param->replyopts.srr ?
381 			    param->replyopts.faddr : iph->saddr),
382 		.fl4_src = saddr,
383 		.fl4_tos = RT_TOS(tos),
384 		.proto = IPPROTO_ICMP,
385 		.fl_icmp_type = type,
386 		.fl_icmp_code = code,
387 	};
388 	struct rtable *rt, *rt2;
389 	int err;
390 
391 	security_skb_classify_flow(skb_in, &fl);
392 	rt = __ip_route_output_key(net, &fl);
393 	if (IS_ERR(rt))
394 		return rt;
395 
396 	/* No need to clone since we're just using its address. */
397 	rt2 = rt;
398 
399 	if (!fl.fl4_src)
400 		fl.fl4_src = rt->rt_src;
401 
402 	rt = (struct rtable *) xfrm_lookup(net, &rt->dst, &fl, NULL, 0);
403 	if (!IS_ERR(rt)) {
404 		if (rt != rt2)
405 			return rt;
406 	} else if (PTR_ERR(rt) == -EPERM) {
407 		rt = NULL;
408 	} else
409 		return rt;
410 
411 	err = xfrm_decode_session_reverse(skb_in, &fl, AF_INET);
412 	if (err)
413 		goto relookup_failed;
414 
415 	if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL) {
416 		rt2 = __ip_route_output_key(net, &fl);
417 		if (IS_ERR(rt2))
418 			err = PTR_ERR(rt2);
419 	} else {
420 		struct flowi fl2 = {};
421 		unsigned long orefdst;
422 
423 		fl2.fl4_dst = fl.fl4_src;
424 		rt2 = ip_route_output_key(net, &fl2);
425 		if (IS_ERR(rt2)) {
426 			err = PTR_ERR(rt2);
427 			goto relookup_failed;
428 		}
429 		/* Ugh! */
430 		orefdst = skb_in->_skb_refdst; /* save old refdst */
431 		err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
432 				     RT_TOS(tos), rt2->dst.dev);
433 
434 		dst_release(&rt2->dst);
435 		rt2 = skb_rtable(skb_in);
436 		skb_in->_skb_refdst = orefdst; /* restore old refdst */
437 	}
438 
439 	if (err)
440 		goto relookup_failed;
441 
442 	rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst, &fl, NULL, XFRM_LOOKUP_ICMP);
443 	if (!IS_ERR(rt2)) {
444 		dst_release(&rt->dst);
445 		rt = rt2;
446 	} else if (PTR_ERR(rt2) == -EPERM) {
447 		if (rt)
448 			dst_release(&rt->dst);
449 		return rt2;
450 	} else {
451 		err = PTR_ERR(rt2);
452 		goto relookup_failed;
453 	}
454 	return rt;
455 
456 relookup_failed:
457 	if (rt)
458 		return rt;
459 	return ERR_PTR(err);
460 }
461 
462 /*
463  *	Send an ICMP message in response to a situation
464  *
465  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
466  *		  MAY send more (we do).
467  *			MUST NOT change this header information.
468  *			MUST NOT reply to a multicast/broadcast IP address.
469  *			MUST NOT reply to a multicast/broadcast MAC address.
470  *			MUST reply to only the first fragment.
471  */
472 
473 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
474 {
475 	struct iphdr *iph;
476 	int room;
477 	struct icmp_bxm icmp_param;
478 	struct rtable *rt = skb_rtable(skb_in);
479 	struct ipcm_cookie ipc;
480 	__be32 saddr;
481 	u8  tos;
482 	struct net *net;
483 	struct sock *sk;
484 
485 	if (!rt)
486 		goto out;
487 	net = dev_net(rt->dst.dev);
488 
489 	/*
490 	 *	Find the original header. It is expected to be valid, of course.
491 	 *	Check this, icmp_send is called from the most obscure devices
492 	 *	sometimes.
493 	 */
494 	iph = ip_hdr(skb_in);
495 
496 	if ((u8 *)iph < skb_in->head ||
497 	    (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
498 		goto out;
499 
500 	/*
501 	 *	No replies to physical multicast/broadcast
502 	 */
503 	if (skb_in->pkt_type != PACKET_HOST)
504 		goto out;
505 
506 	/*
507 	 *	Now check at the protocol level
508 	 */
509 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
510 		goto out;
511 
512 	/*
513 	 *	Only reply to fragment 0. We byte re-order the constant
514 	 *	mask for efficiency.
515 	 */
516 	if (iph->frag_off & htons(IP_OFFSET))
517 		goto out;
518 
519 	/*
520 	 *	If we send an ICMP error to an ICMP error a mess would result..
521 	 */
522 	if (icmp_pointers[type].error) {
523 		/*
524 		 *	We are an error, check if we are replying to an
525 		 *	ICMP error
526 		 */
527 		if (iph->protocol == IPPROTO_ICMP) {
528 			u8 _inner_type, *itp;
529 
530 			itp = skb_header_pointer(skb_in,
531 						 skb_network_header(skb_in) +
532 						 (iph->ihl << 2) +
533 						 offsetof(struct icmphdr,
534 							  type) -
535 						 skb_in->data,
536 						 sizeof(_inner_type),
537 						 &_inner_type);
538 			if (itp == NULL)
539 				goto out;
540 
541 			/*
542 			 *	Assume any unknown ICMP type is an error. This
543 			 *	isn't specified by the RFC, but think about it..
544 			 */
545 			if (*itp > NR_ICMP_TYPES ||
546 			    icmp_pointers[*itp].error)
547 				goto out;
548 		}
549 	}
550 
551 	sk = icmp_xmit_lock(net);
552 	if (sk == NULL)
553 		return;
554 
555 	/*
556 	 *	Construct source address and options.
557 	 */
558 
559 	saddr = iph->daddr;
560 	if (!(rt->rt_flags & RTCF_LOCAL)) {
561 		struct net_device *dev = NULL;
562 
563 		rcu_read_lock();
564 		if (rt_is_input_route(rt) &&
565 		    net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
566 			dev = dev_get_by_index_rcu(net, rt->rt_iif);
567 
568 		if (dev)
569 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
570 		else
571 			saddr = 0;
572 		rcu_read_unlock();
573 	}
574 
575 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
576 					   IPTOS_PREC_INTERNETCONTROL) :
577 					  iph->tos;
578 
579 	if (ip_options_echo(&icmp_param.replyopts, skb_in))
580 		goto out_unlock;
581 
582 
583 	/*
584 	 *	Prepare data for ICMP header.
585 	 */
586 
587 	icmp_param.data.icmph.type	 = type;
588 	icmp_param.data.icmph.code	 = code;
589 	icmp_param.data.icmph.un.gateway = info;
590 	icmp_param.data.icmph.checksum	 = 0;
591 	icmp_param.skb	  = skb_in;
592 	icmp_param.offset = skb_network_offset(skb_in);
593 	inet_sk(sk)->tos = tos;
594 	ipc.addr = iph->saddr;
595 	ipc.opt = &icmp_param.replyopts;
596 	ipc.tx_flags = 0;
597 
598 	rt = icmp_route_lookup(net, skb_in, iph, saddr, tos,
599 			       type, code, &icmp_param);
600 	if (IS_ERR(rt))
601 		goto out_unlock;
602 
603 	if (!icmpv4_xrlim_allow(net, rt, type, code))
604 		goto ende;
605 
606 	/* RFC says return as much as we can without exceeding 576 bytes. */
607 
608 	room = dst_mtu(&rt->dst);
609 	if (room > 576)
610 		room = 576;
611 	room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
612 	room -= sizeof(struct icmphdr);
613 
614 	icmp_param.data_len = skb_in->len - icmp_param.offset;
615 	if (icmp_param.data_len > room)
616 		icmp_param.data_len = room;
617 	icmp_param.head_len = sizeof(struct icmphdr);
618 
619 	icmp_push_reply(&icmp_param, &ipc, &rt);
620 ende:
621 	ip_rt_put(rt);
622 out_unlock:
623 	icmp_xmit_unlock(sk);
624 out:;
625 }
626 EXPORT_SYMBOL(icmp_send);
627 
628 
629 /*
630  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
631  */
632 
633 static void icmp_unreach(struct sk_buff *skb)
634 {
635 	struct iphdr *iph;
636 	struct icmphdr *icmph;
637 	int hash, protocol;
638 	const struct net_protocol *ipprot;
639 	u32 info = 0;
640 	struct net *net;
641 
642 	net = dev_net(skb_dst(skb)->dev);
643 
644 	/*
645 	 *	Incomplete header ?
646 	 * 	Only checks for the IP header, there should be an
647 	 *	additional check for longer headers in upper levels.
648 	 */
649 
650 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
651 		goto out_err;
652 
653 	icmph = icmp_hdr(skb);
654 	iph   = (struct iphdr *)skb->data;
655 
656 	if (iph->ihl < 5) /* Mangled header, drop. */
657 		goto out_err;
658 
659 	if (icmph->type == ICMP_DEST_UNREACH) {
660 		switch (icmph->code & 15) {
661 		case ICMP_NET_UNREACH:
662 		case ICMP_HOST_UNREACH:
663 		case ICMP_PROT_UNREACH:
664 		case ICMP_PORT_UNREACH:
665 			break;
666 		case ICMP_FRAG_NEEDED:
667 			if (ipv4_config.no_pmtu_disc) {
668 				LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n",
669 					       &iph->daddr);
670 			} else {
671 				info = ip_rt_frag_needed(net, iph,
672 							 ntohs(icmph->un.frag.mtu),
673 							 skb->dev);
674 				if (!info)
675 					goto out;
676 			}
677 			break;
678 		case ICMP_SR_FAILED:
679 			LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n",
680 				       &iph->daddr);
681 			break;
682 		default:
683 			break;
684 		}
685 		if (icmph->code > NR_ICMP_UNREACH)
686 			goto out;
687 	} else if (icmph->type == ICMP_PARAMETERPROB)
688 		info = ntohl(icmph->un.gateway) >> 24;
689 
690 	/*
691 	 *	Throw it at our lower layers
692 	 *
693 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
694 	 *		  header.
695 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
696 	 *		  transport layer.
697 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
698 	 *		  transport layer.
699 	 */
700 
701 	/*
702 	 *	Check the other end isnt violating RFC 1122. Some routers send
703 	 *	bogus responses to broadcast frames. If you see this message
704 	 *	first check your netmask matches at both ends, if it does then
705 	 *	get the other vendor to fix their kit.
706 	 */
707 
708 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
709 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
710 		if (net_ratelimit())
711 			printk(KERN_WARNING "%pI4 sent an invalid ICMP "
712 					    "type %u, code %u "
713 					    "error to a broadcast: %pI4 on %s\n",
714 			       &ip_hdr(skb)->saddr,
715 			       icmph->type, icmph->code,
716 			       &iph->daddr,
717 			       skb->dev->name);
718 		goto out;
719 	}
720 
721 	/* Checkin full IP header plus 8 bytes of protocol to
722 	 * avoid additional coding at protocol handlers.
723 	 */
724 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
725 		goto out;
726 
727 	iph = (struct iphdr *)skb->data;
728 	protocol = iph->protocol;
729 
730 	/*
731 	 *	Deliver ICMP message to raw sockets. Pretty useless feature?
732 	 */
733 	raw_icmp_error(skb, protocol, info);
734 
735 	hash = protocol & (MAX_INET_PROTOS - 1);
736 	rcu_read_lock();
737 	ipprot = rcu_dereference(inet_protos[hash]);
738 	if (ipprot && ipprot->err_handler)
739 		ipprot->err_handler(skb, info);
740 	rcu_read_unlock();
741 
742 out:
743 	return;
744 out_err:
745 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
746 	goto out;
747 }
748 
749 
750 /*
751  *	Handle ICMP_REDIRECT.
752  */
753 
754 static void icmp_redirect(struct sk_buff *skb)
755 {
756 	struct iphdr *iph;
757 
758 	if (skb->len < sizeof(struct iphdr))
759 		goto out_err;
760 
761 	/*
762 	 *	Get the copied header of the packet that caused the redirect
763 	 */
764 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
765 		goto out;
766 
767 	iph = (struct iphdr *)skb->data;
768 
769 	switch (icmp_hdr(skb)->code & 7) {
770 	case ICMP_REDIR_NET:
771 	case ICMP_REDIR_NETTOS:
772 		/*
773 		 * As per RFC recommendations now handle it as a host redirect.
774 		 */
775 	case ICMP_REDIR_HOST:
776 	case ICMP_REDIR_HOSTTOS:
777 		ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
778 			       icmp_hdr(skb)->un.gateway,
779 			       iph->saddr, skb->dev);
780 		break;
781 	}
782 out:
783 	return;
784 out_err:
785 	ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
786 	goto out;
787 }
788 
789 /*
790  *	Handle ICMP_ECHO ("ping") requests.
791  *
792  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
793  *		  requests.
794  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
795  *		  included in the reply.
796  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
797  *		  echo requests, MUST have default=NOT.
798  *	See also WRT handling of options once they are done and working.
799  */
800 
801 static void icmp_echo(struct sk_buff *skb)
802 {
803 	struct net *net;
804 
805 	net = dev_net(skb_dst(skb)->dev);
806 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
807 		struct icmp_bxm icmp_param;
808 
809 		icmp_param.data.icmph	   = *icmp_hdr(skb);
810 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
811 		icmp_param.skb		   = skb;
812 		icmp_param.offset	   = 0;
813 		icmp_param.data_len	   = skb->len;
814 		icmp_param.head_len	   = sizeof(struct icmphdr);
815 		icmp_reply(&icmp_param, skb);
816 	}
817 }
818 
819 /*
820  *	Handle ICMP Timestamp requests.
821  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
822  *		  SHOULD be in the kernel for minimum random latency.
823  *		  MUST be accurate to a few minutes.
824  *		  MUST be updated at least at 15Hz.
825  */
826 static void icmp_timestamp(struct sk_buff *skb)
827 {
828 	struct timespec tv;
829 	struct icmp_bxm icmp_param;
830 	/*
831 	 *	Too short.
832 	 */
833 	if (skb->len < 4)
834 		goto out_err;
835 
836 	/*
837 	 *	Fill in the current time as ms since midnight UT:
838 	 */
839 	getnstimeofday(&tv);
840 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
841 					 tv.tv_nsec / NSEC_PER_MSEC);
842 	icmp_param.data.times[2] = icmp_param.data.times[1];
843 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
844 		BUG();
845 	icmp_param.data.icmph	   = *icmp_hdr(skb);
846 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
847 	icmp_param.data.icmph.code = 0;
848 	icmp_param.skb		   = skb;
849 	icmp_param.offset	   = 0;
850 	icmp_param.data_len	   = 0;
851 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
852 	icmp_reply(&icmp_param, skb);
853 out:
854 	return;
855 out_err:
856 	ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
857 	goto out;
858 }
859 
860 
861 /*
862  *	Handle ICMP_ADDRESS_MASK requests.  (RFC950)
863  *
864  * RFC1122 (3.2.2.9).  A host MUST only send replies to
865  * ADDRESS_MASK requests if it's been configured as an address mask
866  * agent.  Receiving a request doesn't constitute implicit permission to
867  * act as one. Of course, implementing this correctly requires (SHOULD)
868  * a way to turn the functionality on and off.  Another one for sysctl(),
869  * I guess. -- MS
870  *
871  * RFC1812 (4.3.3.9).	A router MUST implement it.
872  *			A router SHOULD have switch turning it on/off.
873  *		      	This switch MUST be ON by default.
874  *
875  * Gratuitous replies, zero-source replies are not implemented,
876  * that complies with RFC. DO NOT implement them!!! All the idea
877  * of broadcast addrmask replies as specified in RFC950 is broken.
878  * The problem is that it is not uncommon to have several prefixes
879  * on one physical interface. Moreover, addrmask agent can even be
880  * not aware of existing another prefixes.
881  * If source is zero, addrmask agent cannot choose correct prefix.
882  * Gratuitous mask announcements suffer from the same problem.
883  * RFC1812 explains it, but still allows to use ADDRMASK,
884  * that is pretty silly. --ANK
885  *
886  * All these rules are so bizarre, that I removed kernel addrmask
887  * support at all. It is wrong, it is obsolete, nobody uses it in
888  * any case. --ANK
889  *
890  * Furthermore you can do it with a usermode address agent program
891  * anyway...
892  */
893 
894 static void icmp_address(struct sk_buff *skb)
895 {
896 #if 0
897 	if (net_ratelimit())
898 		printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
899 #endif
900 }
901 
902 /*
903  * RFC1812 (4.3.3.9).	A router SHOULD listen all replies, and complain
904  *			loudly if an inconsistency is found.
905  * called with rcu_read_lock()
906  */
907 
908 static void icmp_address_reply(struct sk_buff *skb)
909 {
910 	struct rtable *rt = skb_rtable(skb);
911 	struct net_device *dev = skb->dev;
912 	struct in_device *in_dev;
913 	struct in_ifaddr *ifa;
914 
915 	if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
916 		return;
917 
918 	in_dev = __in_dev_get_rcu(dev);
919 	if (!in_dev)
920 		return;
921 
922 	if (in_dev->ifa_list &&
923 	    IN_DEV_LOG_MARTIANS(in_dev) &&
924 	    IN_DEV_FORWARD(in_dev)) {
925 		__be32 _mask, *mp;
926 
927 		mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
928 		BUG_ON(mp == NULL);
929 		for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
930 			if (*mp == ifa->ifa_mask &&
931 			    inet_ifa_match(rt->rt_src, ifa))
932 				break;
933 		}
934 		if (!ifa && net_ratelimit()) {
935 			printk(KERN_INFO "Wrong address mask %pI4 from %s/%pI4\n",
936 			       mp, dev->name, &rt->rt_src);
937 		}
938 	}
939 }
940 
941 static void icmp_discard(struct sk_buff *skb)
942 {
943 }
944 
945 /*
946  *	Deal with incoming ICMP packets.
947  */
948 int icmp_rcv(struct sk_buff *skb)
949 {
950 	struct icmphdr *icmph;
951 	struct rtable *rt = skb_rtable(skb);
952 	struct net *net = dev_net(rt->dst.dev);
953 
954 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
955 		struct sec_path *sp = skb_sec_path(skb);
956 		int nh;
957 
958 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
959 				 XFRM_STATE_ICMP))
960 			goto drop;
961 
962 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
963 			goto drop;
964 
965 		nh = skb_network_offset(skb);
966 		skb_set_network_header(skb, sizeof(*icmph));
967 
968 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
969 			goto drop;
970 
971 		skb_set_network_header(skb, nh);
972 	}
973 
974 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
975 
976 	switch (skb->ip_summed) {
977 	case CHECKSUM_COMPLETE:
978 		if (!csum_fold(skb->csum))
979 			break;
980 		/* fall through */
981 	case CHECKSUM_NONE:
982 		skb->csum = 0;
983 		if (__skb_checksum_complete(skb))
984 			goto error;
985 	}
986 
987 	if (!pskb_pull(skb, sizeof(*icmph)))
988 		goto error;
989 
990 	icmph = icmp_hdr(skb);
991 
992 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
993 	/*
994 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
995 	 *
996 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
997 	 *		  discarded.
998 	 */
999 	if (icmph->type > NR_ICMP_TYPES)
1000 		goto error;
1001 
1002 
1003 	/*
1004 	 *	Parse the ICMP message
1005 	 */
1006 
1007 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1008 		/*
1009 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1010 		 *	  silently ignored (we let user decide with a sysctl).
1011 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1012 		 *	  discarded if to broadcast/multicast.
1013 		 */
1014 		if ((icmph->type == ICMP_ECHO ||
1015 		     icmph->type == ICMP_TIMESTAMP) &&
1016 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1017 			goto error;
1018 		}
1019 		if (icmph->type != ICMP_ECHO &&
1020 		    icmph->type != ICMP_TIMESTAMP &&
1021 		    icmph->type != ICMP_ADDRESS &&
1022 		    icmph->type != ICMP_ADDRESSREPLY) {
1023 			goto error;
1024 		}
1025 	}
1026 
1027 	icmp_pointers[icmph->type].handler(skb);
1028 
1029 drop:
1030 	kfree_skb(skb);
1031 	return 0;
1032 error:
1033 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1034 	goto drop;
1035 }
1036 
1037 /*
1038  *	This table is the definition of how we handle ICMP.
1039  */
1040 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1041 	[ICMP_ECHOREPLY] = {
1042 		.handler = icmp_discard,
1043 	},
1044 	[1] = {
1045 		.handler = icmp_discard,
1046 		.error = 1,
1047 	},
1048 	[2] = {
1049 		.handler = icmp_discard,
1050 		.error = 1,
1051 	},
1052 	[ICMP_DEST_UNREACH] = {
1053 		.handler = icmp_unreach,
1054 		.error = 1,
1055 	},
1056 	[ICMP_SOURCE_QUENCH] = {
1057 		.handler = icmp_unreach,
1058 		.error = 1,
1059 	},
1060 	[ICMP_REDIRECT] = {
1061 		.handler = icmp_redirect,
1062 		.error = 1,
1063 	},
1064 	[6] = {
1065 		.handler = icmp_discard,
1066 		.error = 1,
1067 	},
1068 	[7] = {
1069 		.handler = icmp_discard,
1070 		.error = 1,
1071 	},
1072 	[ICMP_ECHO] = {
1073 		.handler = icmp_echo,
1074 	},
1075 	[9] = {
1076 		.handler = icmp_discard,
1077 		.error = 1,
1078 	},
1079 	[10] = {
1080 		.handler = icmp_discard,
1081 		.error = 1,
1082 	},
1083 	[ICMP_TIME_EXCEEDED] = {
1084 		.handler = icmp_unreach,
1085 		.error = 1,
1086 	},
1087 	[ICMP_PARAMETERPROB] = {
1088 		.handler = icmp_unreach,
1089 		.error = 1,
1090 	},
1091 	[ICMP_TIMESTAMP] = {
1092 		.handler = icmp_timestamp,
1093 	},
1094 	[ICMP_TIMESTAMPREPLY] = {
1095 		.handler = icmp_discard,
1096 	},
1097 	[ICMP_INFO_REQUEST] = {
1098 		.handler = icmp_discard,
1099 	},
1100 	[ICMP_INFO_REPLY] = {
1101 		.handler = icmp_discard,
1102 	},
1103 	[ICMP_ADDRESS] = {
1104 		.handler = icmp_address,
1105 	},
1106 	[ICMP_ADDRESSREPLY] = {
1107 		.handler = icmp_address_reply,
1108 	},
1109 };
1110 
1111 static void __net_exit icmp_sk_exit(struct net *net)
1112 {
1113 	int i;
1114 
1115 	for_each_possible_cpu(i)
1116 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1117 	kfree(net->ipv4.icmp_sk);
1118 	net->ipv4.icmp_sk = NULL;
1119 }
1120 
1121 static int __net_init icmp_sk_init(struct net *net)
1122 {
1123 	int i, err;
1124 
1125 	net->ipv4.icmp_sk =
1126 		kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1127 	if (net->ipv4.icmp_sk == NULL)
1128 		return -ENOMEM;
1129 
1130 	for_each_possible_cpu(i) {
1131 		struct sock *sk;
1132 
1133 		err = inet_ctl_sock_create(&sk, PF_INET,
1134 					   SOCK_RAW, IPPROTO_ICMP, net);
1135 		if (err < 0)
1136 			goto fail;
1137 
1138 		net->ipv4.icmp_sk[i] = sk;
1139 
1140 		/* Enough space for 2 64K ICMP packets, including
1141 		 * sk_buff struct overhead.
1142 		 */
1143 		sk->sk_sndbuf =
1144 			(2 * ((64 * 1024) + sizeof(struct sk_buff)));
1145 
1146 		/*
1147 		 * Speedup sock_wfree()
1148 		 */
1149 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1150 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1151 	}
1152 
1153 	/* Control parameters for ECHO replies. */
1154 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1155 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1156 
1157 	/* Control parameter - ignore bogus broadcast responses? */
1158 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1159 
1160 	/*
1161 	 * 	Configurable global rate limit.
1162 	 *
1163 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1164 	 *	bucket ratemask defines which icmp types are ratelimited by
1165 	 *	setting	it's bit position.
1166 	 *
1167 	 *	default:
1168 	 *	dest unreachable (3), source quench (4),
1169 	 *	time exceeded (11), parameter problem (12)
1170 	 */
1171 
1172 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1173 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1174 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1175 
1176 	return 0;
1177 
1178 fail:
1179 	for_each_possible_cpu(i)
1180 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1181 	kfree(net->ipv4.icmp_sk);
1182 	return err;
1183 }
1184 
1185 static struct pernet_operations __net_initdata icmp_sk_ops = {
1186        .init = icmp_sk_init,
1187        .exit = icmp_sk_exit,
1188 };
1189 
1190 int __init icmp_init(void)
1191 {
1192 	return register_pernet_subsys(&icmp_sk_ops);
1193 }
1194