xref: /openbmc/linux/net/ipv4/icmp.c (revision a09d2831)
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 <net/snmp.h>
78 #include <net/ip.h>
79 #include <net/route.h>
80 #include <net/protocol.h>
81 #include <net/icmp.h>
82 #include <net/tcp.h>
83 #include <net/udp.h>
84 #include <net/raw.h>
85 #include <linux/skbuff.h>
86 #include <net/sock.h>
87 #include <linux/errno.h>
88 #include <linux/timer.h>
89 #include <linux/init.h>
90 #include <asm/system.h>
91 #include <asm/uaccess.h>
92 #include <net/checksum.h>
93 #include <net/xfrm.h>
94 #include <net/inet_common.h>
95 
96 /*
97  *	Build xmit assembly blocks
98  */
99 
100 struct icmp_bxm {
101 	struct sk_buff *skb;
102 	int offset;
103 	int data_len;
104 
105 	struct {
106 		struct icmphdr icmph;
107 		__be32	       times[3];
108 	} data;
109 	int head_len;
110 	struct ip_options replyopts;
111 	unsigned char  optbuf[40];
112 };
113 
114 /* An array of errno for error messages from dest unreach. */
115 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
116 
117 struct icmp_err icmp_err_convert[] = {
118 	{
119 		.errno = ENETUNREACH,	/* ICMP_NET_UNREACH */
120 		.fatal = 0,
121 	},
122 	{
123 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNREACH */
124 		.fatal = 0,
125 	},
126 	{
127 		.errno = ENOPROTOOPT	/* ICMP_PROT_UNREACH */,
128 		.fatal = 1,
129 	},
130 	{
131 		.errno = ECONNREFUSED,	/* ICMP_PORT_UNREACH */
132 		.fatal = 1,
133 	},
134 	{
135 		.errno = EMSGSIZE,	/* ICMP_FRAG_NEEDED */
136 		.fatal = 0,
137 	},
138 	{
139 		.errno = EOPNOTSUPP,	/* ICMP_SR_FAILED */
140 		.fatal = 0,
141 	},
142 	{
143 		.errno = ENETUNREACH,	/* ICMP_NET_UNKNOWN */
144 		.fatal = 1,
145 	},
146 	{
147 		.errno = EHOSTDOWN,	/* ICMP_HOST_UNKNOWN */
148 		.fatal = 1,
149 	},
150 	{
151 		.errno = ENONET,	/* ICMP_HOST_ISOLATED */
152 		.fatal = 1,
153 	},
154 	{
155 		.errno = ENETUNREACH,	/* ICMP_NET_ANO	*/
156 		.fatal = 1,
157 	},
158 	{
159 		.errno = EHOSTUNREACH,	/* ICMP_HOST_ANO */
160 		.fatal = 1,
161 	},
162 	{
163 		.errno = ENETUNREACH,	/* ICMP_NET_UNR_TOS */
164 		.fatal = 0,
165 	},
166 	{
167 		.errno = EHOSTUNREACH,	/* ICMP_HOST_UNR_TOS */
168 		.fatal = 0,
169 	},
170 	{
171 		.errno = EHOSTUNREACH,	/* ICMP_PKT_FILTERED */
172 		.fatal = 1,
173 	},
174 	{
175 		.errno = EHOSTUNREACH,	/* ICMP_PREC_VIOLATION */
176 		.fatal = 1,
177 	},
178 	{
179 		.errno = EHOSTUNREACH,	/* ICMP_PREC_CUTOFF */
180 		.fatal = 1,
181 	},
182 };
183 
184 /*
185  *	ICMP control array. This specifies what to do with each ICMP.
186  */
187 
188 struct icmp_control {
189 	void (*handler)(struct sk_buff *skb);
190 	short   error;		/* This ICMP is classed as an error message */
191 };
192 
193 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
194 
195 /*
196  *	The ICMP socket(s). This is the most convenient way to flow control
197  *	our ICMP output as well as maintain a clean interface throughout
198  *	all layers. All Socketless IP sends will soon be gone.
199  *
200  *	On SMP we have one ICMP socket per-cpu.
201  */
202 static struct sock *icmp_sk(struct net *net)
203 {
204 	return net->ipv4.icmp_sk[smp_processor_id()];
205 }
206 
207 static inline struct sock *icmp_xmit_lock(struct net *net)
208 {
209 	struct sock *sk;
210 
211 	local_bh_disable();
212 
213 	sk = icmp_sk(net);
214 
215 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
216 		/* This can happen if the output path signals a
217 		 * dst_link_failure() for an outgoing ICMP packet.
218 		 */
219 		local_bh_enable();
220 		return NULL;
221 	}
222 	return sk;
223 }
224 
225 static inline void icmp_xmit_unlock(struct sock *sk)
226 {
227 	spin_unlock_bh(&sk->sk_lock.slock);
228 }
229 
230 /*
231  *	Send an ICMP frame.
232  */
233 
234 /*
235  *	Check transmit rate limitation for given message.
236  *	The rate information is held in the destination cache now.
237  *	This function is generic and could be used for other purposes
238  *	too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
239  *
240  *	Note that the same dst_entry fields are modified by functions in
241  *	route.c too, but these work for packet destinations while xrlim_allow
242  *	works for icmp destinations. This means the rate limiting information
243  *	for one "ip object" is shared - and these ICMPs are twice limited:
244  *	by source and by destination.
245  *
246  *	RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
247  *			  SHOULD allow setting of rate limits
248  *
249  * 	Shared between ICMPv4 and ICMPv6.
250  */
251 #define XRLIM_BURST_FACTOR 6
252 int xrlim_allow(struct dst_entry *dst, int timeout)
253 {
254 	unsigned long now, token = dst->rate_tokens;
255 	int rc = 0;
256 
257 	now = jiffies;
258 	token += now - dst->rate_last;
259 	dst->rate_last = now;
260 	if (token > XRLIM_BURST_FACTOR * timeout)
261 		token = XRLIM_BURST_FACTOR * timeout;
262 	if (token >= timeout) {
263 		token -= timeout;
264 		rc = 1;
265 	}
266 	dst->rate_tokens = token;
267 	return rc;
268 }
269 
270 static inline int icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
271 		int type, int code)
272 {
273 	struct dst_entry *dst = &rt->u.dst;
274 	int rc = 1;
275 
276 	if (type > NR_ICMP_TYPES)
277 		goto out;
278 
279 	/* Don't limit PMTU discovery. */
280 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
281 		goto out;
282 
283 	/* No rate limit on loopback */
284 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
285 		goto out;
286 
287 	/* Limit if icmp type is enabled in ratemask. */
288 	if ((1 << type) & net->ipv4.sysctl_icmp_ratemask)
289 		rc = xrlim_allow(dst, net->ipv4.sysctl_icmp_ratelimit);
290 out:
291 	return rc;
292 }
293 
294 /*
295  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
296  */
297 void icmp_out_count(struct net *net, unsigned char type)
298 {
299 	ICMPMSGOUT_INC_STATS(net, type);
300 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
301 }
302 
303 /*
304  *	Checksum each fragment, and on the first include the headers and final
305  *	checksum.
306  */
307 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
308 			  struct sk_buff *skb)
309 {
310 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
311 	__wsum csum;
312 
313 	csum = skb_copy_and_csum_bits(icmp_param->skb,
314 				      icmp_param->offset + offset,
315 				      to, len, 0);
316 
317 	skb->csum = csum_block_add(skb->csum, csum, odd);
318 	if (icmp_pointers[icmp_param->data.icmph.type].error)
319 		nf_ct_attach(skb, icmp_param->skb);
320 	return 0;
321 }
322 
323 static void icmp_push_reply(struct icmp_bxm *icmp_param,
324 			    struct ipcm_cookie *ipc, struct rtable **rt)
325 {
326 	struct sock *sk;
327 	struct sk_buff *skb;
328 
329 	sk = icmp_sk(dev_net((*rt)->u.dst.dev));
330 	if (ip_append_data(sk, icmp_glue_bits, icmp_param,
331 			   icmp_param->data_len+icmp_param->head_len,
332 			   icmp_param->head_len,
333 			   ipc, rt, MSG_DONTWAIT) < 0)
334 		ip_flush_pending_frames(sk);
335 	else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
336 		struct icmphdr *icmph = icmp_hdr(skb);
337 		__wsum csum = 0;
338 		struct sk_buff *skb1;
339 
340 		skb_queue_walk(&sk->sk_write_queue, skb1) {
341 			csum = csum_add(csum, skb1->csum);
342 		}
343 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
344 						 (char *)icmph,
345 						 icmp_param->head_len, csum);
346 		icmph->checksum = csum_fold(csum);
347 		skb->ip_summed = CHECKSUM_NONE;
348 		ip_push_pending_frames(sk);
349 	}
350 }
351 
352 /*
353  *	Driving logic for building and sending ICMP messages.
354  */
355 
356 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
357 {
358 	struct ipcm_cookie ipc;
359 	struct rtable *rt = skb_rtable(skb);
360 	struct net *net = dev_net(rt->u.dst.dev);
361 	struct sock *sk;
362 	struct inet_sock *inet;
363 	__be32 daddr;
364 
365 	if (ip_options_echo(&icmp_param->replyopts, skb))
366 		return;
367 
368 	sk = icmp_xmit_lock(net);
369 	if (sk == NULL)
370 		return;
371 	inet = inet_sk(sk);
372 
373 	icmp_param->data.icmph.checksum = 0;
374 
375 	inet->tos = ip_hdr(skb)->tos;
376 	daddr = ipc.addr = rt->rt_src;
377 	ipc.opt = NULL;
378 	ipc.shtx.flags = 0;
379 	if (icmp_param->replyopts.optlen) {
380 		ipc.opt = &icmp_param->replyopts;
381 		if (ipc.opt->srr)
382 			daddr = icmp_param->replyopts.faddr;
383 	}
384 	{
385 		struct flowi fl = { .nl_u = { .ip4_u =
386 					      { .daddr = daddr,
387 						.saddr = rt->rt_spec_dst,
388 						.tos = RT_TOS(ip_hdr(skb)->tos) } },
389 				    .proto = IPPROTO_ICMP };
390 		security_skb_classify_flow(skb, &fl);
391 		if (ip_route_output_key(net, &rt, &fl))
392 			goto out_unlock;
393 	}
394 	if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
395 			       icmp_param->data.icmph.code))
396 		icmp_push_reply(icmp_param, &ipc, &rt);
397 	ip_rt_put(rt);
398 out_unlock:
399 	icmp_xmit_unlock(sk);
400 }
401 
402 
403 /*
404  *	Send an ICMP message in response to a situation
405  *
406  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
407  *		  MAY send more (we do).
408  *			MUST NOT change this header information.
409  *			MUST NOT reply to a multicast/broadcast IP address.
410  *			MUST NOT reply to a multicast/broadcast MAC address.
411  *			MUST reply to only the first fragment.
412  */
413 
414 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
415 {
416 	struct iphdr *iph;
417 	int room;
418 	struct icmp_bxm icmp_param;
419 	struct rtable *rt = skb_rtable(skb_in);
420 	struct ipcm_cookie ipc;
421 	__be32 saddr;
422 	u8  tos;
423 	struct net *net;
424 	struct sock *sk;
425 
426 	if (!rt)
427 		goto out;
428 	net = dev_net(rt->u.dst.dev);
429 
430 	/*
431 	 *	Find the original header. It is expected to be valid, of course.
432 	 *	Check this, icmp_send is called from the most obscure devices
433 	 *	sometimes.
434 	 */
435 	iph = ip_hdr(skb_in);
436 
437 	if ((u8 *)iph < skb_in->head ||
438 	    (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
439 		goto out;
440 
441 	/*
442 	 *	No replies to physical multicast/broadcast
443 	 */
444 	if (skb_in->pkt_type != PACKET_HOST)
445 		goto out;
446 
447 	/*
448 	 *	Now check at the protocol level
449 	 */
450 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
451 		goto out;
452 
453 	/*
454 	 *	Only reply to fragment 0. We byte re-order the constant
455 	 *	mask for efficiency.
456 	 */
457 	if (iph->frag_off & htons(IP_OFFSET))
458 		goto out;
459 
460 	/*
461 	 *	If we send an ICMP error to an ICMP error a mess would result..
462 	 */
463 	if (icmp_pointers[type].error) {
464 		/*
465 		 *	We are an error, check if we are replying to an
466 		 *	ICMP error
467 		 */
468 		if (iph->protocol == IPPROTO_ICMP) {
469 			u8 _inner_type, *itp;
470 
471 			itp = skb_header_pointer(skb_in,
472 						 skb_network_header(skb_in) +
473 						 (iph->ihl << 2) +
474 						 offsetof(struct icmphdr,
475 							  type) -
476 						 skb_in->data,
477 						 sizeof(_inner_type),
478 						 &_inner_type);
479 			if (itp == NULL)
480 				goto out;
481 
482 			/*
483 			 *	Assume any unknown ICMP type is an error. This
484 			 *	isn't specified by the RFC, but think about it..
485 			 */
486 			if (*itp > NR_ICMP_TYPES ||
487 			    icmp_pointers[*itp].error)
488 				goto out;
489 		}
490 	}
491 
492 	sk = icmp_xmit_lock(net);
493 	if (sk == NULL)
494 		return;
495 
496 	/*
497 	 *	Construct source address and options.
498 	 */
499 
500 	saddr = iph->daddr;
501 	if (!(rt->rt_flags & RTCF_LOCAL)) {
502 		struct net_device *dev = NULL;
503 
504 		rcu_read_lock();
505 		if (rt->fl.iif &&
506 			net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
507 			dev = dev_get_by_index_rcu(net, rt->fl.iif);
508 
509 		if (dev)
510 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
511 		else
512 			saddr = 0;
513 		rcu_read_unlock();
514 	}
515 
516 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
517 					   IPTOS_PREC_INTERNETCONTROL) :
518 					  iph->tos;
519 
520 	if (ip_options_echo(&icmp_param.replyopts, skb_in))
521 		goto out_unlock;
522 
523 
524 	/*
525 	 *	Prepare data for ICMP header.
526 	 */
527 
528 	icmp_param.data.icmph.type	 = type;
529 	icmp_param.data.icmph.code	 = code;
530 	icmp_param.data.icmph.un.gateway = info;
531 	icmp_param.data.icmph.checksum	 = 0;
532 	icmp_param.skb	  = skb_in;
533 	icmp_param.offset = skb_network_offset(skb_in);
534 	inet_sk(sk)->tos = tos;
535 	ipc.addr = iph->saddr;
536 	ipc.opt = &icmp_param.replyopts;
537 	ipc.shtx.flags = 0;
538 
539 	{
540 		struct flowi fl = {
541 			.nl_u = {
542 				.ip4_u = {
543 					.daddr = icmp_param.replyopts.srr ?
544 						icmp_param.replyopts.faddr :
545 						iph->saddr,
546 					.saddr = saddr,
547 					.tos = RT_TOS(tos)
548 				}
549 			},
550 			.proto = IPPROTO_ICMP,
551 			.uli_u = {
552 				.icmpt = {
553 					.type = type,
554 					.code = code
555 				}
556 			}
557 		};
558 		int err;
559 		struct rtable *rt2;
560 
561 		security_skb_classify_flow(skb_in, &fl);
562 		if (__ip_route_output_key(net, &rt, &fl))
563 			goto out_unlock;
564 
565 		/* No need to clone since we're just using its address. */
566 		rt2 = rt;
567 
568 		err = xfrm_lookup(net, (struct dst_entry **)&rt, &fl, NULL, 0);
569 		switch (err) {
570 		case 0:
571 			if (rt != rt2)
572 				goto route_done;
573 			break;
574 		case -EPERM:
575 			rt = NULL;
576 			break;
577 		default:
578 			goto out_unlock;
579 		}
580 
581 		if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
582 			goto relookup_failed;
583 
584 		if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
585 			err = __ip_route_output_key(net, &rt2, &fl);
586 		else {
587 			struct flowi fl2 = {};
588 			struct dst_entry *odst;
589 
590 			fl2.fl4_dst = fl.fl4_src;
591 			if (ip_route_output_key(net, &rt2, &fl2))
592 				goto relookup_failed;
593 
594 			/* Ugh! */
595 			odst = skb_dst(skb_in);
596 			err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
597 					     RT_TOS(tos), rt2->u.dst.dev);
598 
599 			dst_release(&rt2->u.dst);
600 			rt2 = skb_rtable(skb_in);
601 			skb_dst_set(skb_in, odst);
602 		}
603 
604 		if (err)
605 			goto relookup_failed;
606 
607 		err = xfrm_lookup(net, (struct dst_entry **)&rt2, &fl, NULL,
608 				  XFRM_LOOKUP_ICMP);
609 		switch (err) {
610 		case 0:
611 			dst_release(&rt->u.dst);
612 			rt = rt2;
613 			break;
614 		case -EPERM:
615 			goto ende;
616 		default:
617 relookup_failed:
618 			if (!rt)
619 				goto out_unlock;
620 			break;
621 		}
622 	}
623 
624 route_done:
625 	if (!icmpv4_xrlim_allow(net, rt, type, code))
626 		goto ende;
627 
628 	/* RFC says return as much as we can without exceeding 576 bytes. */
629 
630 	room = dst_mtu(&rt->u.dst);
631 	if (room > 576)
632 		room = 576;
633 	room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
634 	room -= sizeof(struct icmphdr);
635 
636 	icmp_param.data_len = skb_in->len - icmp_param.offset;
637 	if (icmp_param.data_len > room)
638 		icmp_param.data_len = room;
639 	icmp_param.head_len = sizeof(struct icmphdr);
640 
641 	icmp_push_reply(&icmp_param, &ipc, &rt);
642 ende:
643 	ip_rt_put(rt);
644 out_unlock:
645 	icmp_xmit_unlock(sk);
646 out:;
647 }
648 
649 
650 /*
651  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
652  */
653 
654 static void icmp_unreach(struct sk_buff *skb)
655 {
656 	struct iphdr *iph;
657 	struct icmphdr *icmph;
658 	int hash, protocol;
659 	const struct net_protocol *ipprot;
660 	u32 info = 0;
661 	struct net *net;
662 
663 	net = dev_net(skb_dst(skb)->dev);
664 
665 	/*
666 	 *	Incomplete header ?
667 	 * 	Only checks for the IP header, there should be an
668 	 *	additional check for longer headers in upper levels.
669 	 */
670 
671 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
672 		goto out_err;
673 
674 	icmph = icmp_hdr(skb);
675 	iph   = (struct iphdr *)skb->data;
676 
677 	if (iph->ihl < 5) /* Mangled header, drop. */
678 		goto out_err;
679 
680 	if (icmph->type == ICMP_DEST_UNREACH) {
681 		switch (icmph->code & 15) {
682 		case ICMP_NET_UNREACH:
683 		case ICMP_HOST_UNREACH:
684 		case ICMP_PROT_UNREACH:
685 		case ICMP_PORT_UNREACH:
686 			break;
687 		case ICMP_FRAG_NEEDED:
688 			if (ipv4_config.no_pmtu_disc) {
689 				LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n",
690 					       &iph->daddr);
691 			} else {
692 				info = ip_rt_frag_needed(net, iph,
693 							 ntohs(icmph->un.frag.mtu),
694 							 skb->dev);
695 				if (!info)
696 					goto out;
697 			}
698 			break;
699 		case ICMP_SR_FAILED:
700 			LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n",
701 				       &iph->daddr);
702 			break;
703 		default:
704 			break;
705 		}
706 		if (icmph->code > NR_ICMP_UNREACH)
707 			goto out;
708 	} else if (icmph->type == ICMP_PARAMETERPROB)
709 		info = ntohl(icmph->un.gateway) >> 24;
710 
711 	/*
712 	 *	Throw it at our lower layers
713 	 *
714 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
715 	 *		  header.
716 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
717 	 *		  transport layer.
718 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
719 	 *		  transport layer.
720 	 */
721 
722 	/*
723 	 *	Check the other end isnt violating RFC 1122. Some routers send
724 	 *	bogus responses to broadcast frames. If you see this message
725 	 *	first check your netmask matches at both ends, if it does then
726 	 *	get the other vendor to fix their kit.
727 	 */
728 
729 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
730 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
731 		if (net_ratelimit())
732 			printk(KERN_WARNING "%pI4 sent an invalid ICMP "
733 					    "type %u, code %u "
734 					    "error to a broadcast: %pI4 on %s\n",
735 			       &ip_hdr(skb)->saddr,
736 			       icmph->type, icmph->code,
737 			       &iph->daddr,
738 			       skb->dev->name);
739 		goto out;
740 	}
741 
742 	/* Checkin full IP header plus 8 bytes of protocol to
743 	 * avoid additional coding at protocol handlers.
744 	 */
745 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
746 		goto out;
747 
748 	iph = (struct iphdr *)skb->data;
749 	protocol = iph->protocol;
750 
751 	/*
752 	 *	Deliver ICMP message to raw sockets. Pretty useless feature?
753 	 */
754 	raw_icmp_error(skb, protocol, info);
755 
756 	hash = protocol & (MAX_INET_PROTOS - 1);
757 	rcu_read_lock();
758 	ipprot = rcu_dereference(inet_protos[hash]);
759 	if (ipprot && ipprot->err_handler)
760 		ipprot->err_handler(skb, info);
761 	rcu_read_unlock();
762 
763 out:
764 	return;
765 out_err:
766 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
767 	goto out;
768 }
769 
770 
771 /*
772  *	Handle ICMP_REDIRECT.
773  */
774 
775 static void icmp_redirect(struct sk_buff *skb)
776 {
777 	struct iphdr *iph;
778 
779 	if (skb->len < sizeof(struct iphdr))
780 		goto out_err;
781 
782 	/*
783 	 *	Get the copied header of the packet that caused the redirect
784 	 */
785 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
786 		goto out;
787 
788 	iph = (struct iphdr *)skb->data;
789 
790 	switch (icmp_hdr(skb)->code & 7) {
791 	case ICMP_REDIR_NET:
792 	case ICMP_REDIR_NETTOS:
793 		/*
794 		 * As per RFC recommendations now handle it as a host redirect.
795 		 */
796 	case ICMP_REDIR_HOST:
797 	case ICMP_REDIR_HOSTTOS:
798 		ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
799 			       icmp_hdr(skb)->un.gateway,
800 			       iph->saddr, skb->dev);
801 		break;
802 	}
803 out:
804 	return;
805 out_err:
806 	ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
807 	goto out;
808 }
809 
810 /*
811  *	Handle ICMP_ECHO ("ping") requests.
812  *
813  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
814  *		  requests.
815  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
816  *		  included in the reply.
817  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
818  *		  echo requests, MUST have default=NOT.
819  *	See also WRT handling of options once they are done and working.
820  */
821 
822 static void icmp_echo(struct sk_buff *skb)
823 {
824 	struct net *net;
825 
826 	net = dev_net(skb_dst(skb)->dev);
827 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
828 		struct icmp_bxm icmp_param;
829 
830 		icmp_param.data.icmph	   = *icmp_hdr(skb);
831 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
832 		icmp_param.skb		   = skb;
833 		icmp_param.offset	   = 0;
834 		icmp_param.data_len	   = skb->len;
835 		icmp_param.head_len	   = sizeof(struct icmphdr);
836 		icmp_reply(&icmp_param, skb);
837 	}
838 }
839 
840 /*
841  *	Handle ICMP Timestamp requests.
842  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
843  *		  SHOULD be in the kernel for minimum random latency.
844  *		  MUST be accurate to a few minutes.
845  *		  MUST be updated at least at 15Hz.
846  */
847 static void icmp_timestamp(struct sk_buff *skb)
848 {
849 	struct timespec tv;
850 	struct icmp_bxm icmp_param;
851 	/*
852 	 *	Too short.
853 	 */
854 	if (skb->len < 4)
855 		goto out_err;
856 
857 	/*
858 	 *	Fill in the current time as ms since midnight UT:
859 	 */
860 	getnstimeofday(&tv);
861 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
862 					 tv.tv_nsec / NSEC_PER_MSEC);
863 	icmp_param.data.times[2] = icmp_param.data.times[1];
864 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
865 		BUG();
866 	icmp_param.data.icmph	   = *icmp_hdr(skb);
867 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
868 	icmp_param.data.icmph.code = 0;
869 	icmp_param.skb		   = skb;
870 	icmp_param.offset	   = 0;
871 	icmp_param.data_len	   = 0;
872 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
873 	icmp_reply(&icmp_param, skb);
874 out:
875 	return;
876 out_err:
877 	ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
878 	goto out;
879 }
880 
881 
882 /*
883  *	Handle ICMP_ADDRESS_MASK requests.  (RFC950)
884  *
885  * RFC1122 (3.2.2.9).  A host MUST only send replies to
886  * ADDRESS_MASK requests if it's been configured as an address mask
887  * agent.  Receiving a request doesn't constitute implicit permission to
888  * act as one. Of course, implementing this correctly requires (SHOULD)
889  * a way to turn the functionality on and off.  Another one for sysctl(),
890  * I guess. -- MS
891  *
892  * RFC1812 (4.3.3.9).	A router MUST implement it.
893  *			A router SHOULD have switch turning it on/off.
894  *		      	This switch MUST be ON by default.
895  *
896  * Gratuitous replies, zero-source replies are not implemented,
897  * that complies with RFC. DO NOT implement them!!! All the idea
898  * of broadcast addrmask replies as specified in RFC950 is broken.
899  * The problem is that it is not uncommon to have several prefixes
900  * on one physical interface. Moreover, addrmask agent can even be
901  * not aware of existing another prefixes.
902  * If source is zero, addrmask agent cannot choose correct prefix.
903  * Gratuitous mask announcements suffer from the same problem.
904  * RFC1812 explains it, but still allows to use ADDRMASK,
905  * that is pretty silly. --ANK
906  *
907  * All these rules are so bizarre, that I removed kernel addrmask
908  * support at all. It is wrong, it is obsolete, nobody uses it in
909  * any case. --ANK
910  *
911  * Furthermore you can do it with a usermode address agent program
912  * anyway...
913  */
914 
915 static void icmp_address(struct sk_buff *skb)
916 {
917 #if 0
918 	if (net_ratelimit())
919 		printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
920 #endif
921 }
922 
923 /*
924  * RFC1812 (4.3.3.9).	A router SHOULD listen all replies, and complain
925  *			loudly if an inconsistency is found.
926  */
927 
928 static void icmp_address_reply(struct sk_buff *skb)
929 {
930 	struct rtable *rt = skb_rtable(skb);
931 	struct net_device *dev = skb->dev;
932 	struct in_device *in_dev;
933 	struct in_ifaddr *ifa;
934 
935 	if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
936 		goto out;
937 
938 	in_dev = in_dev_get(dev);
939 	if (!in_dev)
940 		goto out;
941 	rcu_read_lock();
942 	if (in_dev->ifa_list &&
943 	    IN_DEV_LOG_MARTIANS(in_dev) &&
944 	    IN_DEV_FORWARD(in_dev)) {
945 		__be32 _mask, *mp;
946 
947 		mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
948 		BUG_ON(mp == NULL);
949 		for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
950 			if (*mp == ifa->ifa_mask &&
951 			    inet_ifa_match(rt->rt_src, ifa))
952 				break;
953 		}
954 		if (!ifa && net_ratelimit()) {
955 			printk(KERN_INFO "Wrong address mask %pI4 from %s/%pI4\n",
956 			       mp, dev->name, &rt->rt_src);
957 		}
958 	}
959 	rcu_read_unlock();
960 	in_dev_put(in_dev);
961 out:;
962 }
963 
964 static void icmp_discard(struct sk_buff *skb)
965 {
966 }
967 
968 /*
969  *	Deal with incoming ICMP packets.
970  */
971 int icmp_rcv(struct sk_buff *skb)
972 {
973 	struct icmphdr *icmph;
974 	struct rtable *rt = skb_rtable(skb);
975 	struct net *net = dev_net(rt->u.dst.dev);
976 
977 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
978 		struct sec_path *sp = skb_sec_path(skb);
979 		int nh;
980 
981 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
982 				 XFRM_STATE_ICMP))
983 			goto drop;
984 
985 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
986 			goto drop;
987 
988 		nh = skb_network_offset(skb);
989 		skb_set_network_header(skb, sizeof(*icmph));
990 
991 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
992 			goto drop;
993 
994 		skb_set_network_header(skb, nh);
995 	}
996 
997 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
998 
999 	switch (skb->ip_summed) {
1000 	case CHECKSUM_COMPLETE:
1001 		if (!csum_fold(skb->csum))
1002 			break;
1003 		/* fall through */
1004 	case CHECKSUM_NONE:
1005 		skb->csum = 0;
1006 		if (__skb_checksum_complete(skb))
1007 			goto error;
1008 	}
1009 
1010 	if (!pskb_pull(skb, sizeof(*icmph)))
1011 		goto error;
1012 
1013 	icmph = icmp_hdr(skb);
1014 
1015 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1016 	/*
1017 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1018 	 *
1019 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1020 	 *		  discarded.
1021 	 */
1022 	if (icmph->type > NR_ICMP_TYPES)
1023 		goto error;
1024 
1025 
1026 	/*
1027 	 *	Parse the ICMP message
1028 	 */
1029 
1030 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1031 		/*
1032 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1033 		 *	  silently ignored (we let user decide with a sysctl).
1034 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1035 		 *	  discarded if to broadcast/multicast.
1036 		 */
1037 		if ((icmph->type == ICMP_ECHO ||
1038 		     icmph->type == ICMP_TIMESTAMP) &&
1039 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1040 			goto error;
1041 		}
1042 		if (icmph->type != ICMP_ECHO &&
1043 		    icmph->type != ICMP_TIMESTAMP &&
1044 		    icmph->type != ICMP_ADDRESS &&
1045 		    icmph->type != ICMP_ADDRESSREPLY) {
1046 			goto error;
1047 		}
1048 	}
1049 
1050 	icmp_pointers[icmph->type].handler(skb);
1051 
1052 drop:
1053 	kfree_skb(skb);
1054 	return 0;
1055 error:
1056 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1057 	goto drop;
1058 }
1059 
1060 /*
1061  *	This table is the definition of how we handle ICMP.
1062  */
1063 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1064 	[ICMP_ECHOREPLY] = {
1065 		.handler = icmp_discard,
1066 	},
1067 	[1] = {
1068 		.handler = icmp_discard,
1069 		.error = 1,
1070 	},
1071 	[2] = {
1072 		.handler = icmp_discard,
1073 		.error = 1,
1074 	},
1075 	[ICMP_DEST_UNREACH] = {
1076 		.handler = icmp_unreach,
1077 		.error = 1,
1078 	},
1079 	[ICMP_SOURCE_QUENCH] = {
1080 		.handler = icmp_unreach,
1081 		.error = 1,
1082 	},
1083 	[ICMP_REDIRECT] = {
1084 		.handler = icmp_redirect,
1085 		.error = 1,
1086 	},
1087 	[6] = {
1088 		.handler = icmp_discard,
1089 		.error = 1,
1090 	},
1091 	[7] = {
1092 		.handler = icmp_discard,
1093 		.error = 1,
1094 	},
1095 	[ICMP_ECHO] = {
1096 		.handler = icmp_echo,
1097 	},
1098 	[9] = {
1099 		.handler = icmp_discard,
1100 		.error = 1,
1101 	},
1102 	[10] = {
1103 		.handler = icmp_discard,
1104 		.error = 1,
1105 	},
1106 	[ICMP_TIME_EXCEEDED] = {
1107 		.handler = icmp_unreach,
1108 		.error = 1,
1109 	},
1110 	[ICMP_PARAMETERPROB] = {
1111 		.handler = icmp_unreach,
1112 		.error = 1,
1113 	},
1114 	[ICMP_TIMESTAMP] = {
1115 		.handler = icmp_timestamp,
1116 	},
1117 	[ICMP_TIMESTAMPREPLY] = {
1118 		.handler = icmp_discard,
1119 	},
1120 	[ICMP_INFO_REQUEST] = {
1121 		.handler = icmp_discard,
1122 	},
1123 	[ICMP_INFO_REPLY] = {
1124 		.handler = icmp_discard,
1125 	},
1126 	[ICMP_ADDRESS] = {
1127 		.handler = icmp_address,
1128 	},
1129 	[ICMP_ADDRESSREPLY] = {
1130 		.handler = icmp_address_reply,
1131 	},
1132 };
1133 
1134 static void __net_exit icmp_sk_exit(struct net *net)
1135 {
1136 	int i;
1137 
1138 	for_each_possible_cpu(i)
1139 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1140 	kfree(net->ipv4.icmp_sk);
1141 	net->ipv4.icmp_sk = NULL;
1142 }
1143 
1144 static int __net_init icmp_sk_init(struct net *net)
1145 {
1146 	int i, err;
1147 
1148 	net->ipv4.icmp_sk =
1149 		kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1150 	if (net->ipv4.icmp_sk == NULL)
1151 		return -ENOMEM;
1152 
1153 	for_each_possible_cpu(i) {
1154 		struct sock *sk;
1155 
1156 		err = inet_ctl_sock_create(&sk, PF_INET,
1157 					   SOCK_RAW, IPPROTO_ICMP, net);
1158 		if (err < 0)
1159 			goto fail;
1160 
1161 		net->ipv4.icmp_sk[i] = sk;
1162 
1163 		/* Enough space for 2 64K ICMP packets, including
1164 		 * sk_buff struct overhead.
1165 		 */
1166 		sk->sk_sndbuf =
1167 			(2 * ((64 * 1024) + sizeof(struct sk_buff)));
1168 
1169 		/*
1170 		 * Speedup sock_wfree()
1171 		 */
1172 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1173 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1174 	}
1175 
1176 	/* Control parameters for ECHO replies. */
1177 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1178 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1179 
1180 	/* Control parameter - ignore bogus broadcast responses? */
1181 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1182 
1183 	/*
1184 	 * 	Configurable global rate limit.
1185 	 *
1186 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1187 	 *	bucket ratemask defines which icmp types are ratelimited by
1188 	 *	setting	it's bit position.
1189 	 *
1190 	 *	default:
1191 	 *	dest unreachable (3), source quench (4),
1192 	 *	time exceeded (11), parameter problem (12)
1193 	 */
1194 
1195 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1196 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1197 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1198 
1199 	return 0;
1200 
1201 fail:
1202 	for_each_possible_cpu(i)
1203 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1204 	kfree(net->ipv4.icmp_sk);
1205 	return err;
1206 }
1207 
1208 static struct pernet_operations __net_initdata icmp_sk_ops = {
1209        .init = icmp_sk_init,
1210        .exit = icmp_sk_exit,
1211 };
1212 
1213 int __init icmp_init(void)
1214 {
1215 	return register_pernet_subsys(&icmp_sk_ops);
1216 }
1217 
1218 EXPORT_SYMBOL(icmp_err_convert);
1219 EXPORT_SYMBOL(icmp_send);
1220 EXPORT_SYMBOL(xrlim_allow);
1221