xref: /openbmc/linux/net/ipv4/icmp.c (revision d5532ee7)
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 /*
237  *	Check transmit rate limitation for given message.
238  *	The rate information is held in the destination cache now.
239  *	This function is generic and could be used for other purposes
240  *	too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
241  *
242  *	Note that the same dst_entry fields are modified by functions in
243  *	route.c too, but these work for packet destinations while xrlim_allow
244  *	works for icmp destinations. This means the rate limiting information
245  *	for one "ip object" is shared - and these ICMPs are twice limited:
246  *	by source and by destination.
247  *
248  *	RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
249  *			  SHOULD allow setting of rate limits
250  *
251  * 	Shared between ICMPv4 and ICMPv6.
252  */
253 #define XRLIM_BURST_FACTOR 6
254 int xrlim_allow(struct dst_entry *dst, int timeout)
255 {
256 	unsigned long now, token = dst->rate_tokens;
257 	int rc = 0;
258 
259 	now = jiffies;
260 	token += now - dst->rate_last;
261 	dst->rate_last = now;
262 	if (token > XRLIM_BURST_FACTOR * timeout)
263 		token = XRLIM_BURST_FACTOR * timeout;
264 	if (token >= timeout) {
265 		token -= timeout;
266 		rc = 1;
267 	}
268 	dst->rate_tokens = token;
269 	return rc;
270 }
271 EXPORT_SYMBOL(xrlim_allow);
272 
273 static inline int icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
274 		int type, int code)
275 {
276 	struct dst_entry *dst = &rt->dst;
277 	int rc = 1;
278 
279 	if (type > NR_ICMP_TYPES)
280 		goto out;
281 
282 	/* Don't limit PMTU discovery. */
283 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
284 		goto out;
285 
286 	/* No rate limit on loopback */
287 	if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
288 		goto out;
289 
290 	/* Limit if icmp type is enabled in ratemask. */
291 	if ((1 << type) & net->ipv4.sysctl_icmp_ratemask)
292 		rc = xrlim_allow(dst, net->ipv4.sysctl_icmp_ratelimit);
293 out:
294 	return rc;
295 }
296 
297 /*
298  *	Maintain the counters used in the SNMP statistics for outgoing ICMP
299  */
300 void icmp_out_count(struct net *net, unsigned char type)
301 {
302 	ICMPMSGOUT_INC_STATS(net, type);
303 	ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
304 }
305 
306 /*
307  *	Checksum each fragment, and on the first include the headers and final
308  *	checksum.
309  */
310 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
311 			  struct sk_buff *skb)
312 {
313 	struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
314 	__wsum csum;
315 
316 	csum = skb_copy_and_csum_bits(icmp_param->skb,
317 				      icmp_param->offset + offset,
318 				      to, len, 0);
319 
320 	skb->csum = csum_block_add(skb->csum, csum, odd);
321 	if (icmp_pointers[icmp_param->data.icmph.type].error)
322 		nf_ct_attach(skb, icmp_param->skb);
323 	return 0;
324 }
325 
326 static void icmp_push_reply(struct icmp_bxm *icmp_param,
327 			    struct ipcm_cookie *ipc, struct rtable **rt)
328 {
329 	struct sock *sk;
330 	struct sk_buff *skb;
331 
332 	sk = icmp_sk(dev_net((*rt)->dst.dev));
333 	if (ip_append_data(sk, icmp_glue_bits, icmp_param,
334 			   icmp_param->data_len+icmp_param->head_len,
335 			   icmp_param->head_len,
336 			   ipc, rt, MSG_DONTWAIT) < 0) {
337 		ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
338 		ip_flush_pending_frames(sk);
339 	} else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
340 		struct icmphdr *icmph = icmp_hdr(skb);
341 		__wsum csum = 0;
342 		struct sk_buff *skb1;
343 
344 		skb_queue_walk(&sk->sk_write_queue, skb1) {
345 			csum = csum_add(csum, skb1->csum);
346 		}
347 		csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
348 						 (char *)icmph,
349 						 icmp_param->head_len, csum);
350 		icmph->checksum = csum_fold(csum);
351 		skb->ip_summed = CHECKSUM_NONE;
352 		ip_push_pending_frames(sk);
353 	}
354 }
355 
356 /*
357  *	Driving logic for building and sending ICMP messages.
358  */
359 
360 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
361 {
362 	struct ipcm_cookie ipc;
363 	struct rtable *rt = skb_rtable(skb);
364 	struct net *net = dev_net(rt->dst.dev);
365 	struct sock *sk;
366 	struct inet_sock *inet;
367 	__be32 daddr;
368 
369 	if (ip_options_echo(&icmp_param->replyopts, skb))
370 		return;
371 
372 	sk = icmp_xmit_lock(net);
373 	if (sk == NULL)
374 		return;
375 	inet = inet_sk(sk);
376 
377 	icmp_param->data.icmph.checksum = 0;
378 
379 	inet->tos = ip_hdr(skb)->tos;
380 	daddr = ipc.addr = rt->rt_src;
381 	ipc.opt = NULL;
382 	ipc.shtx.flags = 0;
383 	if (icmp_param->replyopts.optlen) {
384 		ipc.opt = &icmp_param->replyopts;
385 		if (ipc.opt->srr)
386 			daddr = icmp_param->replyopts.faddr;
387 	}
388 	{
389 		struct flowi fl = { .nl_u = { .ip4_u =
390 					      { .daddr = daddr,
391 						.saddr = rt->rt_spec_dst,
392 						.tos = RT_TOS(ip_hdr(skb)->tos) } },
393 				    .proto = IPPROTO_ICMP };
394 		security_skb_classify_flow(skb, &fl);
395 		if (ip_route_output_key(net, &rt, &fl))
396 			goto out_unlock;
397 	}
398 	if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
399 			       icmp_param->data.icmph.code))
400 		icmp_push_reply(icmp_param, &ipc, &rt);
401 	ip_rt_put(rt);
402 out_unlock:
403 	icmp_xmit_unlock(sk);
404 }
405 
406 
407 /*
408  *	Send an ICMP message in response to a situation
409  *
410  *	RFC 1122: 3.2.2	MUST send at least the IP header and 8 bytes of header.
411  *		  MAY send more (we do).
412  *			MUST NOT change this header information.
413  *			MUST NOT reply to a multicast/broadcast IP address.
414  *			MUST NOT reply to a multicast/broadcast MAC address.
415  *			MUST reply to only the first fragment.
416  */
417 
418 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
419 {
420 	struct iphdr *iph;
421 	int room;
422 	struct icmp_bxm icmp_param;
423 	struct rtable *rt = skb_rtable(skb_in);
424 	struct ipcm_cookie ipc;
425 	__be32 saddr;
426 	u8  tos;
427 	struct net *net;
428 	struct sock *sk;
429 
430 	if (!rt)
431 		goto out;
432 	net = dev_net(rt->dst.dev);
433 
434 	/*
435 	 *	Find the original header. It is expected to be valid, of course.
436 	 *	Check this, icmp_send is called from the most obscure devices
437 	 *	sometimes.
438 	 */
439 	iph = ip_hdr(skb_in);
440 
441 	if ((u8 *)iph < skb_in->head ||
442 	    (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
443 		goto out;
444 
445 	/*
446 	 *	No replies to physical multicast/broadcast
447 	 */
448 	if (skb_in->pkt_type != PACKET_HOST)
449 		goto out;
450 
451 	/*
452 	 *	Now check at the protocol level
453 	 */
454 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
455 		goto out;
456 
457 	/*
458 	 *	Only reply to fragment 0. We byte re-order the constant
459 	 *	mask for efficiency.
460 	 */
461 	if (iph->frag_off & htons(IP_OFFSET))
462 		goto out;
463 
464 	/*
465 	 *	If we send an ICMP error to an ICMP error a mess would result..
466 	 */
467 	if (icmp_pointers[type].error) {
468 		/*
469 		 *	We are an error, check if we are replying to an
470 		 *	ICMP error
471 		 */
472 		if (iph->protocol == IPPROTO_ICMP) {
473 			u8 _inner_type, *itp;
474 
475 			itp = skb_header_pointer(skb_in,
476 						 skb_network_header(skb_in) +
477 						 (iph->ihl << 2) +
478 						 offsetof(struct icmphdr,
479 							  type) -
480 						 skb_in->data,
481 						 sizeof(_inner_type),
482 						 &_inner_type);
483 			if (itp == NULL)
484 				goto out;
485 
486 			/*
487 			 *	Assume any unknown ICMP type is an error. This
488 			 *	isn't specified by the RFC, but think about it..
489 			 */
490 			if (*itp > NR_ICMP_TYPES ||
491 			    icmp_pointers[*itp].error)
492 				goto out;
493 		}
494 	}
495 
496 	sk = icmp_xmit_lock(net);
497 	if (sk == NULL)
498 		return;
499 
500 	/*
501 	 *	Construct source address and options.
502 	 */
503 
504 	saddr = iph->daddr;
505 	if (!(rt->rt_flags & RTCF_LOCAL)) {
506 		struct net_device *dev = NULL;
507 
508 		rcu_read_lock();
509 		if (rt->fl.iif &&
510 			net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
511 			dev = dev_get_by_index_rcu(net, rt->fl.iif);
512 
513 		if (dev)
514 			saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
515 		else
516 			saddr = 0;
517 		rcu_read_unlock();
518 	}
519 
520 	tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
521 					   IPTOS_PREC_INTERNETCONTROL) :
522 					  iph->tos;
523 
524 	if (ip_options_echo(&icmp_param.replyopts, skb_in))
525 		goto out_unlock;
526 
527 
528 	/*
529 	 *	Prepare data for ICMP header.
530 	 */
531 
532 	icmp_param.data.icmph.type	 = type;
533 	icmp_param.data.icmph.code	 = code;
534 	icmp_param.data.icmph.un.gateway = info;
535 	icmp_param.data.icmph.checksum	 = 0;
536 	icmp_param.skb	  = skb_in;
537 	icmp_param.offset = skb_network_offset(skb_in);
538 	inet_sk(sk)->tos = tos;
539 	ipc.addr = iph->saddr;
540 	ipc.opt = &icmp_param.replyopts;
541 	ipc.shtx.flags = 0;
542 
543 	{
544 		struct flowi fl = {
545 			.nl_u = {
546 				.ip4_u = {
547 					.daddr = icmp_param.replyopts.srr ?
548 						icmp_param.replyopts.faddr :
549 						iph->saddr,
550 					.saddr = saddr,
551 					.tos = RT_TOS(tos)
552 				}
553 			},
554 			.proto = IPPROTO_ICMP,
555 			.uli_u = {
556 				.icmpt = {
557 					.type = type,
558 					.code = code
559 				}
560 			}
561 		};
562 		int err;
563 		struct rtable *rt2;
564 
565 		security_skb_classify_flow(skb_in, &fl);
566 		if (__ip_route_output_key(net, &rt, &fl))
567 			goto out_unlock;
568 
569 		/* No need to clone since we're just using its address. */
570 		rt2 = rt;
571 
572 		err = xfrm_lookup(net, (struct dst_entry **)&rt, &fl, NULL, 0);
573 		switch (err) {
574 		case 0:
575 			if (rt != rt2)
576 				goto route_done;
577 			break;
578 		case -EPERM:
579 			rt = NULL;
580 			break;
581 		default:
582 			goto out_unlock;
583 		}
584 
585 		if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
586 			goto relookup_failed;
587 
588 		if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
589 			err = __ip_route_output_key(net, &rt2, &fl);
590 		else {
591 			struct flowi fl2 = {};
592 			unsigned long orefdst;
593 
594 			fl2.fl4_dst = fl.fl4_src;
595 			if (ip_route_output_key(net, &rt2, &fl2))
596 				goto relookup_failed;
597 
598 			/* Ugh! */
599 			orefdst = skb_in->_skb_refdst; /* save old refdst */
600 			err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
601 					     RT_TOS(tos), rt2->dst.dev);
602 
603 			dst_release(&rt2->dst);
604 			rt2 = skb_rtable(skb_in);
605 			skb_in->_skb_refdst = orefdst; /* restore old refdst */
606 		}
607 
608 		if (err)
609 			goto relookup_failed;
610 
611 		err = xfrm_lookup(net, (struct dst_entry **)&rt2, &fl, NULL,
612 				  XFRM_LOOKUP_ICMP);
613 		switch (err) {
614 		case 0:
615 			dst_release(&rt->dst);
616 			rt = rt2;
617 			break;
618 		case -EPERM:
619 			goto ende;
620 		default:
621 relookup_failed:
622 			if (!rt)
623 				goto out_unlock;
624 			break;
625 		}
626 	}
627 
628 route_done:
629 	if (!icmpv4_xrlim_allow(net, rt, type, code))
630 		goto ende;
631 
632 	/* RFC says return as much as we can without exceeding 576 bytes. */
633 
634 	room = dst_mtu(&rt->dst);
635 	if (room > 576)
636 		room = 576;
637 	room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
638 	room -= sizeof(struct icmphdr);
639 
640 	icmp_param.data_len = skb_in->len - icmp_param.offset;
641 	if (icmp_param.data_len > room)
642 		icmp_param.data_len = room;
643 	icmp_param.head_len = sizeof(struct icmphdr);
644 
645 	icmp_push_reply(&icmp_param, &ipc, &rt);
646 ende:
647 	ip_rt_put(rt);
648 out_unlock:
649 	icmp_xmit_unlock(sk);
650 out:;
651 }
652 EXPORT_SYMBOL(icmp_send);
653 
654 
655 /*
656  *	Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
657  */
658 
659 static void icmp_unreach(struct sk_buff *skb)
660 {
661 	struct iphdr *iph;
662 	struct icmphdr *icmph;
663 	int hash, protocol;
664 	const struct net_protocol *ipprot;
665 	u32 info = 0;
666 	struct net *net;
667 
668 	net = dev_net(skb_dst(skb)->dev);
669 
670 	/*
671 	 *	Incomplete header ?
672 	 * 	Only checks for the IP header, there should be an
673 	 *	additional check for longer headers in upper levels.
674 	 */
675 
676 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
677 		goto out_err;
678 
679 	icmph = icmp_hdr(skb);
680 	iph   = (struct iphdr *)skb->data;
681 
682 	if (iph->ihl < 5) /* Mangled header, drop. */
683 		goto out_err;
684 
685 	if (icmph->type == ICMP_DEST_UNREACH) {
686 		switch (icmph->code & 15) {
687 		case ICMP_NET_UNREACH:
688 		case ICMP_HOST_UNREACH:
689 		case ICMP_PROT_UNREACH:
690 		case ICMP_PORT_UNREACH:
691 			break;
692 		case ICMP_FRAG_NEEDED:
693 			if (ipv4_config.no_pmtu_disc) {
694 				LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n",
695 					       &iph->daddr);
696 			} else {
697 				info = ip_rt_frag_needed(net, iph,
698 							 ntohs(icmph->un.frag.mtu),
699 							 skb->dev);
700 				if (!info)
701 					goto out;
702 			}
703 			break;
704 		case ICMP_SR_FAILED:
705 			LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n",
706 				       &iph->daddr);
707 			break;
708 		default:
709 			break;
710 		}
711 		if (icmph->code > NR_ICMP_UNREACH)
712 			goto out;
713 	} else if (icmph->type == ICMP_PARAMETERPROB)
714 		info = ntohl(icmph->un.gateway) >> 24;
715 
716 	/*
717 	 *	Throw it at our lower layers
718 	 *
719 	 *	RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
720 	 *		  header.
721 	 *	RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
722 	 *		  transport layer.
723 	 *	RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
724 	 *		  transport layer.
725 	 */
726 
727 	/*
728 	 *	Check the other end isnt violating RFC 1122. Some routers send
729 	 *	bogus responses to broadcast frames. If you see this message
730 	 *	first check your netmask matches at both ends, if it does then
731 	 *	get the other vendor to fix their kit.
732 	 */
733 
734 	if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
735 	    inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
736 		if (net_ratelimit())
737 			printk(KERN_WARNING "%pI4 sent an invalid ICMP "
738 					    "type %u, code %u "
739 					    "error to a broadcast: %pI4 on %s\n",
740 			       &ip_hdr(skb)->saddr,
741 			       icmph->type, icmph->code,
742 			       &iph->daddr,
743 			       skb->dev->name);
744 		goto out;
745 	}
746 
747 	/* Checkin full IP header plus 8 bytes of protocol to
748 	 * avoid additional coding at protocol handlers.
749 	 */
750 	if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
751 		goto out;
752 
753 	iph = (struct iphdr *)skb->data;
754 	protocol = iph->protocol;
755 
756 	/*
757 	 *	Deliver ICMP message to raw sockets. Pretty useless feature?
758 	 */
759 	raw_icmp_error(skb, protocol, info);
760 
761 	hash = protocol & (MAX_INET_PROTOS - 1);
762 	rcu_read_lock();
763 	ipprot = rcu_dereference(inet_protos[hash]);
764 	if (ipprot && ipprot->err_handler)
765 		ipprot->err_handler(skb, info);
766 	rcu_read_unlock();
767 
768 out:
769 	return;
770 out_err:
771 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
772 	goto out;
773 }
774 
775 
776 /*
777  *	Handle ICMP_REDIRECT.
778  */
779 
780 static void icmp_redirect(struct sk_buff *skb)
781 {
782 	struct iphdr *iph;
783 
784 	if (skb->len < sizeof(struct iphdr))
785 		goto out_err;
786 
787 	/*
788 	 *	Get the copied header of the packet that caused the redirect
789 	 */
790 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
791 		goto out;
792 
793 	iph = (struct iphdr *)skb->data;
794 
795 	switch (icmp_hdr(skb)->code & 7) {
796 	case ICMP_REDIR_NET:
797 	case ICMP_REDIR_NETTOS:
798 		/*
799 		 * As per RFC recommendations now handle it as a host redirect.
800 		 */
801 	case ICMP_REDIR_HOST:
802 	case ICMP_REDIR_HOSTTOS:
803 		ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
804 			       icmp_hdr(skb)->un.gateway,
805 			       iph->saddr, skb->dev);
806 		break;
807 	}
808 out:
809 	return;
810 out_err:
811 	ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
812 	goto out;
813 }
814 
815 /*
816  *	Handle ICMP_ECHO ("ping") requests.
817  *
818  *	RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
819  *		  requests.
820  *	RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
821  *		  included in the reply.
822  *	RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
823  *		  echo requests, MUST have default=NOT.
824  *	See also WRT handling of options once they are done and working.
825  */
826 
827 static void icmp_echo(struct sk_buff *skb)
828 {
829 	struct net *net;
830 
831 	net = dev_net(skb_dst(skb)->dev);
832 	if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
833 		struct icmp_bxm icmp_param;
834 
835 		icmp_param.data.icmph	   = *icmp_hdr(skb);
836 		icmp_param.data.icmph.type = ICMP_ECHOREPLY;
837 		icmp_param.skb		   = skb;
838 		icmp_param.offset	   = 0;
839 		icmp_param.data_len	   = skb->len;
840 		icmp_param.head_len	   = sizeof(struct icmphdr);
841 		icmp_reply(&icmp_param, skb);
842 	}
843 }
844 
845 /*
846  *	Handle ICMP Timestamp requests.
847  *	RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
848  *		  SHOULD be in the kernel for minimum random latency.
849  *		  MUST be accurate to a few minutes.
850  *		  MUST be updated at least at 15Hz.
851  */
852 static void icmp_timestamp(struct sk_buff *skb)
853 {
854 	struct timespec tv;
855 	struct icmp_bxm icmp_param;
856 	/*
857 	 *	Too short.
858 	 */
859 	if (skb->len < 4)
860 		goto out_err;
861 
862 	/*
863 	 *	Fill in the current time as ms since midnight UT:
864 	 */
865 	getnstimeofday(&tv);
866 	icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
867 					 tv.tv_nsec / NSEC_PER_MSEC);
868 	icmp_param.data.times[2] = icmp_param.data.times[1];
869 	if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
870 		BUG();
871 	icmp_param.data.icmph	   = *icmp_hdr(skb);
872 	icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
873 	icmp_param.data.icmph.code = 0;
874 	icmp_param.skb		   = skb;
875 	icmp_param.offset	   = 0;
876 	icmp_param.data_len	   = 0;
877 	icmp_param.head_len	   = sizeof(struct icmphdr) + 12;
878 	icmp_reply(&icmp_param, skb);
879 out:
880 	return;
881 out_err:
882 	ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
883 	goto out;
884 }
885 
886 
887 /*
888  *	Handle ICMP_ADDRESS_MASK requests.  (RFC950)
889  *
890  * RFC1122 (3.2.2.9).  A host MUST only send replies to
891  * ADDRESS_MASK requests if it's been configured as an address mask
892  * agent.  Receiving a request doesn't constitute implicit permission to
893  * act as one. Of course, implementing this correctly requires (SHOULD)
894  * a way to turn the functionality on and off.  Another one for sysctl(),
895  * I guess. -- MS
896  *
897  * RFC1812 (4.3.3.9).	A router MUST implement it.
898  *			A router SHOULD have switch turning it on/off.
899  *		      	This switch MUST be ON by default.
900  *
901  * Gratuitous replies, zero-source replies are not implemented,
902  * that complies with RFC. DO NOT implement them!!! All the idea
903  * of broadcast addrmask replies as specified in RFC950 is broken.
904  * The problem is that it is not uncommon to have several prefixes
905  * on one physical interface. Moreover, addrmask agent can even be
906  * not aware of existing another prefixes.
907  * If source is zero, addrmask agent cannot choose correct prefix.
908  * Gratuitous mask announcements suffer from the same problem.
909  * RFC1812 explains it, but still allows to use ADDRMASK,
910  * that is pretty silly. --ANK
911  *
912  * All these rules are so bizarre, that I removed kernel addrmask
913  * support at all. It is wrong, it is obsolete, nobody uses it in
914  * any case. --ANK
915  *
916  * Furthermore you can do it with a usermode address agent program
917  * anyway...
918  */
919 
920 static void icmp_address(struct sk_buff *skb)
921 {
922 #if 0
923 	if (net_ratelimit())
924 		printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
925 #endif
926 }
927 
928 /*
929  * RFC1812 (4.3.3.9).	A router SHOULD listen all replies, and complain
930  *			loudly if an inconsistency is found.
931  * called with rcu_read_lock()
932  */
933 
934 static void icmp_address_reply(struct sk_buff *skb)
935 {
936 	struct rtable *rt = skb_rtable(skb);
937 	struct net_device *dev = skb->dev;
938 	struct in_device *in_dev;
939 	struct in_ifaddr *ifa;
940 
941 	if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
942 		return;
943 
944 	in_dev = __in_dev_get_rcu(dev);
945 	if (!in_dev)
946 		return;
947 
948 	if (in_dev->ifa_list &&
949 	    IN_DEV_LOG_MARTIANS(in_dev) &&
950 	    IN_DEV_FORWARD(in_dev)) {
951 		__be32 _mask, *mp;
952 
953 		mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
954 		BUG_ON(mp == NULL);
955 		for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
956 			if (*mp == ifa->ifa_mask &&
957 			    inet_ifa_match(rt->rt_src, ifa))
958 				break;
959 		}
960 		if (!ifa && net_ratelimit()) {
961 			printk(KERN_INFO "Wrong address mask %pI4 from %s/%pI4\n",
962 			       mp, dev->name, &rt->rt_src);
963 		}
964 	}
965 }
966 
967 static void icmp_discard(struct sk_buff *skb)
968 {
969 }
970 
971 /*
972  *	Deal with incoming ICMP packets.
973  */
974 int icmp_rcv(struct sk_buff *skb)
975 {
976 	struct icmphdr *icmph;
977 	struct rtable *rt = skb_rtable(skb);
978 	struct net *net = dev_net(rt->dst.dev);
979 
980 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
981 		struct sec_path *sp = skb_sec_path(skb);
982 		int nh;
983 
984 		if (!(sp && sp->xvec[sp->len - 1]->props.flags &
985 				 XFRM_STATE_ICMP))
986 			goto drop;
987 
988 		if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
989 			goto drop;
990 
991 		nh = skb_network_offset(skb);
992 		skb_set_network_header(skb, sizeof(*icmph));
993 
994 		if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
995 			goto drop;
996 
997 		skb_set_network_header(skb, nh);
998 	}
999 
1000 	ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
1001 
1002 	switch (skb->ip_summed) {
1003 	case CHECKSUM_COMPLETE:
1004 		if (!csum_fold(skb->csum))
1005 			break;
1006 		/* fall through */
1007 	case CHECKSUM_NONE:
1008 		skb->csum = 0;
1009 		if (__skb_checksum_complete(skb))
1010 			goto error;
1011 	}
1012 
1013 	if (!pskb_pull(skb, sizeof(*icmph)))
1014 		goto error;
1015 
1016 	icmph = icmp_hdr(skb);
1017 
1018 	ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1019 	/*
1020 	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
1021 	 *
1022 	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
1023 	 *		  discarded.
1024 	 */
1025 	if (icmph->type > NR_ICMP_TYPES)
1026 		goto error;
1027 
1028 
1029 	/*
1030 	 *	Parse the ICMP message
1031 	 */
1032 
1033 	if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1034 		/*
1035 		 *	RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1036 		 *	  silently ignored (we let user decide with a sysctl).
1037 		 *	RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1038 		 *	  discarded if to broadcast/multicast.
1039 		 */
1040 		if ((icmph->type == ICMP_ECHO ||
1041 		     icmph->type == ICMP_TIMESTAMP) &&
1042 		    net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1043 			goto error;
1044 		}
1045 		if (icmph->type != ICMP_ECHO &&
1046 		    icmph->type != ICMP_TIMESTAMP &&
1047 		    icmph->type != ICMP_ADDRESS &&
1048 		    icmph->type != ICMP_ADDRESSREPLY) {
1049 			goto error;
1050 		}
1051 	}
1052 
1053 	icmp_pointers[icmph->type].handler(skb);
1054 
1055 drop:
1056 	kfree_skb(skb);
1057 	return 0;
1058 error:
1059 	ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1060 	goto drop;
1061 }
1062 
1063 /*
1064  *	This table is the definition of how we handle ICMP.
1065  */
1066 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1067 	[ICMP_ECHOREPLY] = {
1068 		.handler = icmp_discard,
1069 	},
1070 	[1] = {
1071 		.handler = icmp_discard,
1072 		.error = 1,
1073 	},
1074 	[2] = {
1075 		.handler = icmp_discard,
1076 		.error = 1,
1077 	},
1078 	[ICMP_DEST_UNREACH] = {
1079 		.handler = icmp_unreach,
1080 		.error = 1,
1081 	},
1082 	[ICMP_SOURCE_QUENCH] = {
1083 		.handler = icmp_unreach,
1084 		.error = 1,
1085 	},
1086 	[ICMP_REDIRECT] = {
1087 		.handler = icmp_redirect,
1088 		.error = 1,
1089 	},
1090 	[6] = {
1091 		.handler = icmp_discard,
1092 		.error = 1,
1093 	},
1094 	[7] = {
1095 		.handler = icmp_discard,
1096 		.error = 1,
1097 	},
1098 	[ICMP_ECHO] = {
1099 		.handler = icmp_echo,
1100 	},
1101 	[9] = {
1102 		.handler = icmp_discard,
1103 		.error = 1,
1104 	},
1105 	[10] = {
1106 		.handler = icmp_discard,
1107 		.error = 1,
1108 	},
1109 	[ICMP_TIME_EXCEEDED] = {
1110 		.handler = icmp_unreach,
1111 		.error = 1,
1112 	},
1113 	[ICMP_PARAMETERPROB] = {
1114 		.handler = icmp_unreach,
1115 		.error = 1,
1116 	},
1117 	[ICMP_TIMESTAMP] = {
1118 		.handler = icmp_timestamp,
1119 	},
1120 	[ICMP_TIMESTAMPREPLY] = {
1121 		.handler = icmp_discard,
1122 	},
1123 	[ICMP_INFO_REQUEST] = {
1124 		.handler = icmp_discard,
1125 	},
1126 	[ICMP_INFO_REPLY] = {
1127 		.handler = icmp_discard,
1128 	},
1129 	[ICMP_ADDRESS] = {
1130 		.handler = icmp_address,
1131 	},
1132 	[ICMP_ADDRESSREPLY] = {
1133 		.handler = icmp_address_reply,
1134 	},
1135 };
1136 
1137 static void __net_exit icmp_sk_exit(struct net *net)
1138 {
1139 	int i;
1140 
1141 	for_each_possible_cpu(i)
1142 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1143 	kfree(net->ipv4.icmp_sk);
1144 	net->ipv4.icmp_sk = NULL;
1145 }
1146 
1147 static int __net_init icmp_sk_init(struct net *net)
1148 {
1149 	int i, err;
1150 
1151 	net->ipv4.icmp_sk =
1152 		kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1153 	if (net->ipv4.icmp_sk == NULL)
1154 		return -ENOMEM;
1155 
1156 	for_each_possible_cpu(i) {
1157 		struct sock *sk;
1158 
1159 		err = inet_ctl_sock_create(&sk, PF_INET,
1160 					   SOCK_RAW, IPPROTO_ICMP, net);
1161 		if (err < 0)
1162 			goto fail;
1163 
1164 		net->ipv4.icmp_sk[i] = sk;
1165 
1166 		/* Enough space for 2 64K ICMP packets, including
1167 		 * sk_buff struct overhead.
1168 		 */
1169 		sk->sk_sndbuf =
1170 			(2 * ((64 * 1024) + sizeof(struct sk_buff)));
1171 
1172 		/*
1173 		 * Speedup sock_wfree()
1174 		 */
1175 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1176 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1177 	}
1178 
1179 	/* Control parameters for ECHO replies. */
1180 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1181 	net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1182 
1183 	/* Control parameter - ignore bogus broadcast responses? */
1184 	net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1185 
1186 	/*
1187 	 * 	Configurable global rate limit.
1188 	 *
1189 	 *	ratelimit defines tokens/packet consumed for dst->rate_token
1190 	 *	bucket ratemask defines which icmp types are ratelimited by
1191 	 *	setting	it's bit position.
1192 	 *
1193 	 *	default:
1194 	 *	dest unreachable (3), source quench (4),
1195 	 *	time exceeded (11), parameter problem (12)
1196 	 */
1197 
1198 	net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1199 	net->ipv4.sysctl_icmp_ratemask = 0x1818;
1200 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1201 
1202 	return 0;
1203 
1204 fail:
1205 	for_each_possible_cpu(i)
1206 		inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1207 	kfree(net->ipv4.icmp_sk);
1208 	return err;
1209 }
1210 
1211 static struct pernet_operations __net_initdata icmp_sk_ops = {
1212        .init = icmp_sk_init,
1213        .exit = icmp_sk_exit,
1214 };
1215 
1216 int __init icmp_init(void)
1217 {
1218 	return register_pernet_subsys(&icmp_sk_ops);
1219 }
1220